A special line painting control method and system based on intelligent algorithm

Abstract

The application relates to the technical field of intelligent control, and discloses a special line painting control method and system of an intelligent algorithm. The method comprises the following steps: collecting painting line sensor data to construct a process digital model; training a deep deterministic policy gradient network based on the model to generate a multi-parameter collaborative control instruction; executing the control instruction and generating a feedforward signal through multi-level comparison of quality data; fusing the feedforward signal and the control instruction, and forming a real-time control strategy through network correction. The application solves the problem of digital modeling based on multi-dimensional parameter mapping in special line painting control, and overcomes the problem of insufficient precision of multi-parameter collaborative control of the painting process. Through the intelligent algorithm of the deep deterministic policy gradient network and the feedforward control signal fusion mechanism, the problems of lagging detection and insufficient dynamic response of quality abnormalities in the painting process are effectively solved.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and in particular to a method and system for controlling the painting of a dedicated line based on intelligent algorithms. Background Technology

[0002] Magnetite wire, as a key material in electrical equipment, is widely used in transformers, motors, new energy vehicle electrical components, robots, home appliances, and military products. In the production of magnetite wire, the coating process is the core step that determines the product's insulation performance, heat resistance rating, and service life. Traditional coating control methods mainly rely on fixed process parameters and manual experience, typically using a PID (Proportional-Integral-Derivative) controller to maintain key parameters such as temperature and speed.

[0003] Existing painting control technologies have several limitations. Traditional PID control strategies struggle to adapt to the nonlinear characteristics and multivariate coupling relationships in the painting process, making precise parameter control impossible. More importantly, existing control systems generally lack a deep understanding of the synergistic mechanisms of multiple parameters during painting, making it difficult to perform synergistic optimization when raw material properties fluctuate or environmental conditions change. Furthermore, traditional control systems exhibit lag in response to abnormal conditions and lack feedforward control capabilities. They cannot adjust control strategies promptly in situations such as changes in paint properties or fluctuations in wire surface condition, easily leading to unstable product quality.

[0004] In the control of painting on dedicated production lines, the lack of accurate acquisition and modeling capabilities for multi-source parameters makes it difficult to form a complete digital process model. Furthermore, traditional control methods lack quality prediction and feedforward control capabilities, failing to achieve intelligent fusion and dynamic adjustment of control signals, resulting in a delayed response to process fluctuations. Summary of the Invention

[0005] This application provides a method and system for controlling the coating of dedicated production lines using intelligent algorithms. It addresses the problem of digital modeling based on multi-dimensional parameter mapping in the coating control of dedicated production lines, overcoming the insufficient accuracy of multi-parameter collaborative control in the coating process. By fusing the intelligent algorithm of a deep deterministic policy gradient network with a feedforward control signal fusion mechanism, it effectively addresses the problems of delayed detection and insufficient dynamic response of quality anomalies during the coating process.

[0006] Firstly, this application provides a dedicated line painting control method based on intelligent algorithms. The method includes: inputting sensor data from the painting production line into a digital twin system for multi-dimensional parameter mapping to obtain a digital model of the painting process containing temperature parameters, viscosity parameters, tension parameters, and linear velocity; constructing a deep deterministic strategy gradient network structure based on the digital model, inputting the painting process state vector into the deep deterministic strategy gradient network to calculate control actions, and obtaining multi-parameter coordinated control commands for temperature adjustment, speed adjustment, and paint supply; adjusting the parameters of the painting production line in real time according to the multi-parameter coordinated control commands, while simultaneously performing multi-level comparison analysis between the painting quality data and preset thresholds to obtain feedforward control signals; fusing the feedforward control signals with the multi-parameter coordinated control commands, and correcting the control parameters according to the deep deterministic strategy gradient network to obtain a real-time control strategy.

[0007] Optionally, the step of inputting sensor data from the coating production line into a digital twin system for multi-dimensional parameter mapping to obtain a digital model of the coating process including a set of temperature parameters, viscosity parameters, tension parameters, and linear velocity includes: collecting temperature data using temperature sensors installed at the paint tank, oven inlet, middle section, and outlet of the coating production line to obtain a set of temperature parameters; measuring the viscosity of the paint using a viscosity sensor installed at the paint tank and collecting tension data using a tension sensor installed at the guide wheel to obtain viscosity and tension parameters; measuring the wire diameter using a laser diameter measuring instrument positioned at the wire diameter control location, and simultaneously detecting the paint film performance at the end of the production line using an infrared spectrometer to obtain... Abnormal noise is removed from the temperature parameter set, viscosity parameter, tension parameter, and linear velocity based on the linear velocity and product quality data, and each parameter is standardized to obtain a standardized parameter set. Based on the standardized parameter set, temperature-viscosity relationship graphs, viscosity-coating thickness relationship graphs, and linear velocity-curing degree relationship graphs are established, and the weighting coefficients of each parameter are determined to obtain a parameter influence matrix. The parameter influence matrix is ​​combined with the product quality data to construct a mapping function between process parameters and product quality. The mapping accuracy is verified by comparing with historical production data. When the verification error is less than a preset threshold, the mapping function is determined as the digital model of the coating process.

[0008] Optionally, the step of constructing a deep deterministic strategy gradient network structure based on the digital model of the painting process, and inputting the state vector of the painting process into the deep deterministic strategy gradient network to calculate control actions, and obtaining multi-parameter coordinated control commands for temperature adjustment, speed adjustment, and paint supply, includes: combining the temperature parameter set, viscosity parameter, tension parameter, and linear velocity in the digital model of the painting process into a state space, constructing a multi-dimensional state vector as input data for the deep deterministic strategy gradient network; defining the temperature adjustment, speed adjustment, and paint supply as an action space, setting a safety limit range for each action, and constructing an action vector as the output target of the deep deterministic strategy gradient network; and weighting and combining the coating thickness, paint film performance, and energy consumption in the product quality data to construct a reward function as the deep deterministic strategy gradient network. The optimization objective of the policy gradient network is as follows: A policy network structure is constructed based on the state vector, the action vector, and the reward function. A multi-layer neural network maps states to actions, where the input layer corresponds to the state vector dimension, the hidden layer uses an activation function for feature extraction, and the output layer limits the action range. A value network structure is constructed based on the policy network structure. By merging the inputs of the state vector and the action vector, the output policy evaluation value is used as a measure of policy quality, and update parameters are set to achieve smooth network updates. Historical data from the digital model of the painting process is input into an experience replay buffer. The policy network and value network are trained through batch sampling. When the reward value stabilizes after multiple consecutive rounds and the loss of the value function falls below a threshold, a trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are obtained.

[0009] Optionally, the step of inputting historical data from the digital model of the painting process into an experience replay buffer, training the policy network and the value network through batch sampling, and obtaining the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions when the reward value stabilizes for multiple consecutive rounds and the value function loss drops below a threshold, includes: sorting and segmenting the historical data in the digital model of the painting process over time, filtering out effective data sequences containing multiple working conditions, and constructing a fixed-capacity experience replay buffer as a training sample library; randomly sampling batches of samples from the experience replay buffer, each sample containing a state vector, action vector, reward value, and next state vector to form a training batch dataset; inputting the state vector and next state vector from the training batch dataset into the target policy network and target value network respectively, and calculating the target... The Q-value is used as the learning target of the value network. Based on the mean squared error between the target Q-value and the Q-value predicted by the current value network, the loss function of the value network is calculated, and the value network parameters are updated using gradient descent to optimize value assessment accuracy. Based on the gradient information of the value network, the policy network is updated using policy gradients. The update direction and step size of the policy network parameters are calculated using a deterministic policy gradient algorithm to maximize the value assessment of the actions output by the policy network. The target policy network and the target value network are soft-updated using a fixed update rate, and the current network parameters are fused to ensure training stability. The degree of convergence is determined by monitoring the reward value and value function loss in consecutive rounds. When the reward value fluctuation range limit is met and the loss value is lower than the training termination threshold, the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are derived.

[0010] Optionally, the step of adjusting the parameters of the painting production line in real time according to the multi-parameter collaborative control command, and simultaneously performing multi-level comparison and analysis of the painting quality data with preset thresholds to obtain a feedforward control signal, includes: parsing the multi-parameter collaborative control command into temperature adjustment amount, speed adjustment amount, and paint supply amount; transmitting the control values ​​to the actuator through an industrial bus system to adjust the parameters of the painting production line in real time to obtain the adjusted painting process state; collecting coating thickness, paint film uniformity, and insulation strength data in the adjusted painting process state in real time through an online detection system; comparing each quality indicator with its corresponding early warning threshold at multiple levels to obtain a quality status assessment result; predicting the direction of quality change based on the quality status assessment result and historical quality change trends; and generating a feedforward control signal containing intervention level and intervention parameters when the prediction shows that any quality indicator deviates from the normal range.

[0011] Optionally, the step of comparing each quality indicator with its corresponding warning threshold at multiple levels includes: calculating the statistical distribution characteristics of the real-time collected coating thickness data to obtain the mean and standard deviation of the thickness, and comparing them with preset first-level, second-level, and critical thresholds for thickness to generate a thickness anomaly level identifier; performing spectral analysis on the real-time collected paint film uniformity data to extract the surface fluctuation frequency and amplitude characteristics, and comparing them with preset first-level, second-level, and critical thresholds for uniformity to generate a uniformity anomaly level identifier; establishing a time series model for the real-time collected insulation strength data to extract the strength change trend, and comparing it with preset first-level, second-level, and critical thresholds for insulation strength. Critical thresholds are compared to generate insulation strength anomaly level identifiers. The thickness anomaly level identifier, the uniformity anomaly level identifier, and the insulation strength anomaly level identifier are weighted together to calculate a comprehensive anomaly index. A time-series cumulative analysis is performed on the comprehensive anomaly index. When the anomaly index continuously rises and exceeds the first warning line, a minor intervention is triggered; when it exceeds the second warning line, a moderate intervention is triggered; and when it exceeds the third warning line, a strong intervention is triggered, forming a tiered intervention decision. Based on the tiered intervention decision and the processing strategies in the historical anomaly handling experience database, an intervention measure mapping table is constructed, and corresponding intervention parameter adjustment amounts are assigned to each anomaly situation, completing the conversion from quality status assessment results to feedforward control signals.

[0012] Optionally, the step of fusing the feedforward control signal with the multi-parameter collaborative control command and correcting the control parameters according to the deep deterministic strategy gradient network to obtain a real-time control strategy includes: parsing the feedforward control signal, extracting the intervention level, intervention parameters, and intervention direction, and converting the intervention parameters into a compensation vector compatible with the multi-parameter collaborative control command format to obtain a control correction amount; weighting and fusing the multi-parameter collaborative control command with the control correction amount, setting the weight coefficient of the feedforward control signal according to the intervention level, and generating a fused control command to ensure that the feedforward control priority is increased in the case of quality anomalies; combining the fused control command with the current painting process state vector, inputting it into the deep deterministic strategy gradient network for online inference to obtain control parameters corrected by the network; performing a safety check on the control parameters corrected by the network to ensure that each control parameter is within the set safety boundary to prevent abnormal control from causing equipment damage or product quality collapse, and obtaining safe control parameters; dividing the safe control parameters into immediate adjustment parameters and gradual adjustment parameters according to the degree of process influence, executing the immediate adjustment parameters directly, and using a ramp function to smoothly transition the gradual adjustment parameters to obtain a real-time control strategy.

[0013] Secondly, this application provides a dedicated line coating control system based on an intelligent algorithm, the dedicated line coating control system based on the intelligent algorithm comprising:

[0014] The mapping module is used to input sensor data from the painting production line into the digital twin system for multi-dimensional parameter mapping, resulting in a digital model of the painting process that includes temperature parameters, viscosity parameters, tension parameters, and linear velocity.

[0015] The input module is used to construct a deep deterministic strategy gradient network structure based on the digital model of the painting process, input the state vector of the painting process into the deep deterministic strategy gradient network to calculate the control action, and obtain multi-parameter coordinated control commands for temperature adjustment, speed adjustment and paint supply.

[0016] The comparison module is used to adjust the parameters of the painting production line in real time according to the multi-parameter collaborative control instructions, and at the same time, compare and analyze the painting quality data with the preset threshold to obtain the feedforward control signal.

[0017] The fusion module is used to fuse the feedforward control signal with the multi-parameter collaborative control command, and to correct the control parameters according to the deep deterministic policy gradient network to obtain the real-time control strategy.

[0018] Thirdly, a dedicated line coating control device based on an intelligent algorithm is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the dedicated line coating control device based on the intelligent algorithm to execute the aforementioned dedicated line coating control method based on the intelligent algorithm.

[0019] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to execute the aforementioned intelligent algorithm for a dedicated line painting control method.

[0020] The technical solution provided in this application effectively solves the problem of insufficient understanding of the painting process by inputting sensor data from the painting production line into a digital twin system for multi-dimensional parameter mapping. It constructs a complete digital model of the painting process, including temperature parameters, viscosity parameters, tension parameters, and linear velocity, enabling accurate description and prediction of process parameters. Based on this digital model, a deep deterministic policy gradient network structure is built, leveraging the advantages of combining deep learning and reinforcement learning algorithms to overcome the limitations of traditional PID control in handling nonlinear, multivariable coupled systems. By inputting the painting process state vector into the deep deterministic policy gradient network to calculate control actions, it significantly improves control accuracy and response speed. This algorithm is particularly suitable for industrial processes with highly nonlinear and multi-parameter coupling characteristics, such as dedicated line painting. It can automatically generate multi-parameter coordinated control commands for temperature regulation, speed regulation, and paint supply based on a complex state space, solving the technical problem of difficult multi-parameter coordinated optimization. In actual control execution, this method adjusts the parameters of the painting production line in real time according to multi-parameter collaborative control instructions. At the same time, it innovatively performs multi-level comparison and analysis of painting quality data with preset thresholds to generate feedforward control signals, realizing early prediction and intervention of quality anomalies, and effectively solving the problem of lagging quality anomaly detection in traditional control.

[0021] The deep deterministic policy gradient algorithm in this application, through an Actor-Critic network structure and an experience replay mechanism, can learn the optimal control strategy from historical data and continuously optimize the decision-making process through value evaluation. Its contribution to the coating control scheme lies in its ability to handle continuous motion space, adapt to process parameter fluctuations, and autonomously learn and optimize control strategies. By fusing feedforward control signals with multi-parameter collaborative control commands and correcting control parameters according to the deep deterministic policy gradient network, a real-time control strategy is obtained. This dual control mechanism combines the advantages of data-driven deep learning control and rule-based feedforward control, enhancing the system's robustness and adaptability. By deeply integrating deep reinforcement learning algorithms with the characteristics of the coating process, it creatively solves the problems of precise parameter control, multi-parameter collaborative optimization, and dynamic adaptability in dedicated line coating control, significantly improving coating quality stability and production efficiency. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1This is a schematic diagram of an embodiment of the dedicated line painting control method of the intelligent algorithm in this application;

[0024] Figure 2 This is a schematic diagram of an embodiment of the intelligent algorithm-based dedicated line painting control system in this application.

[0025] Figure 3 This is a schematic block diagram of the structure of a dedicated line painting control device with intelligent algorithm in an embodiment of the present invention. Detailed Implementation

[0026] This application provides a method and system for controlling the coating of a dedicated line using an intelligent algorithm. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0027] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the dedicated line painting control method of the intelligent algorithm in this application includes:

[0028] Step S101: Input the sensor data of the painting production line into the digital twin system for multi-dimensional parameter mapping to obtain a digital model of the painting process that includes temperature parameter set, viscosity parameter, tension parameter and linear velocity.

[0029] Step S102: Construct a deep deterministic strategy gradient network structure based on the digital model of the painting process, input the state vector of the painting process into the deep deterministic strategy gradient network to calculate the control action, and obtain multi-parameter coordinated control commands for temperature adjustment, speed adjustment and paint supply.

[0030] Step S103: Adjust the parameters of the painting production line in real time according to the multi-parameter collaborative control command, and at the same time perform multi-level comparison and analysis between the painting quality data and the preset threshold to obtain the feedforward control signal.

[0031] Step S104: The feedforward control signal and the multi-parameter collaborative control command are fused and processed, and the control parameters are corrected according to the deep deterministic strategy gradient network to obtain the real-time control strategy.

[0032] It is understood that the executing entity of this application can be a dedicated line painting control system based on intelligent algorithms, or it can be a terminal or a server; no specific limitation is made here. This application's embodiments use a server as an example for illustration.

[0033] Specifically, a sensor network is deployed at key locations in the coating production line to collect critical process parameters, including temperature, viscosity, tension, and linear velocity. Temperature parameters are collected by temperature sensors located at the paint tank, oven inlet, middle section, and outlet. Viscosity parameters are measured by viscosity sensors in the paint tank, tension parameters are obtained by tension sensors at the guide wheel, and linear velocity is provided by a laser diameter gauge at the wire diameter control position and an infrared spectrometer at the end. The collected raw data undergoes noise removal and standardization to form a standardized parameter set. Based on these parameters, temperature-viscosity, viscosity-coating thickness, and linear velocity-curing degree graphs are established. Multiple regression analysis is used to determine the weight coefficients of each parameter, resulting in a parameter influence matrix. Finally, the parameter influence matrix is ​​combined with product quality data to construct a mapping function between process parameters and product quality. When the verification error meets the conditions, it is determined as the digital model of the coating process. Based on the obtained digital model of the coating process, a deep deterministic strategy gradient network structure is constructed, which is particularly suitable for control problems in continuous motion spaces. In this step, the temperature, viscosity, tension, and linear velocity parameters from the digital model of the painting process are combined into a state vector, which serves as the input data for the deep deterministic policy gradient network. Simultaneously, temperature adjustment, velocity adjustment, and paint supply are defined as the action space, with safety limits set. Coating thickness, paint film performance, and energy consumption from the product quality data are weighted and combined to construct a reward function as the optimization objective. Based on these elements, a policy network and a value network are constructed. Batch sampling training is performed using historical data stored in the experience replay buffer. When the reward value and loss function meet the convergence condition, the trained network and its generated multi-parameter collaborative control commands are obtained.

[0034] The painting production line is adjusted in real time based on multi-parameter collaborative control commands, while quality monitoring is performed simultaneously. Control commands are parsed into specific temperature, speed, and paint supply amounts, and transmitted to actuators via an industrial bus system for parameter adjustment. After adjustment, online detection systems collect data on coating thickness, paint film uniformity, and insulation strength, comparing each quality indicator with preset multi-level warning thresholds. Statistical distribution characteristics are calculated for coating thickness data, spectral analysis is performed on paint film uniformity data, and a time series model is established for insulation strength data to generate respective anomaly level identifiers. A comprehensive anomaly index is then calculated using weighted averages. Through time-series cumulative analysis, graded intervention decisions are made based on the severity of the anomaly index, and feedforward control signals are generated by combining historical processing experience.

[0035] A real-time control strategy is obtained by fusing feedforward control signals with multi-parameter collaborative control commands. Intervention level, parameters, and direction are extracted from the feedforward control signals and converted into compensation vectors. These vectors are then weighted and fused with the multi-parameter collaborative control commands, with weight coefficients adjusted according to the intervention level. The fused control commands, along with the current process state vector, are input into a deep deterministic strategy gradient network for online inference to obtain corrected control parameters. These parameters undergo safety checks and are categorized into immediate-adjustment parameters and gradual-adjustment parameters based on their impact on the process. Different execution strategies are employed to achieve precise control of the painting process. This approach solves the problems of insufficient real-time parameter control accuracy, difficulty in multi-parameter collaborative optimization, insufficient dynamic adaptability, and lag in quality anomaly detection in traditional painting control, thus achieving precise control of the painting process.

[0036] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0037] Temperature data is collected by temperature sensors installed at the paint tank, oven inlet, middle section and outlet of the painting production line to obtain a set of temperature parameters;

[0038] The viscosity of the paint is measured by a viscosity sensor installed in the paint tank, and the tension data is collected by a tension sensor installed on the guide wheel to obtain viscosity and tension parameters.

[0039] The wire diameter is measured by a laser diameter measuring instrument placed at the wire diameter control position, and the coating performance is detected by an infrared spectrometer at the end of the production line to obtain line speed and product quality data.

[0040] Abnormal noise is removed from the temperature parameter set, viscosity parameter, tension parameter and linear velocity, and each parameter is standardized to obtain a standardized parameter set;

[0041] Based on the standardized parameter set, establish the temperature-viscosity relationship spectrum, viscosity-coating thickness relationship spectrum, and linear velocity-degree of cure relationship spectrum, and determine the weight coefficient of each parameter to obtain the parameter influence matrix;

[0042] By combining the parameter influence matrix with product quality data, a mapping function between process parameters and product quality is constructed. The mapping accuracy is verified by comparing it with historical production data. When the verification error is less than a preset threshold, the mapping function is determined as the digital model of the painting process.

[0043] Specifically, key process parameters are collected through a rationally deployed sensor network. High-precision temperature sensors are installed at the paint tank, oven inlet, middle section, and outlet of the coating production line, with a sampling frequency set to 10 times per second to ensure continuous recording of temperature changes, resulting in a set of temperature parameters covering multiple points. These parameter sets reflect the spatial distribution and temporal changes of temperature during the coating process, which is crucial for paint film curing. Simultaneously, viscosity sensors are installed in the paint tank to measure the paint viscosity value once per minute. Viscosity is a key factor affecting coating uniformity, and the measured value directly reflects the paint flow characteristics. Tension sensors are installed at the guide wheel, collecting tension data 5 times per second to obtain the tension changes of the wire during the coating process, which is of great significance for controlling the uniformity of paint film thickness.

[0044] Laser diameter gauges are deployed at the line diameter control positions to measure the line diameter 20 times per second with an accuracy of 10 micrometers, and the linear velocity is calculated from the changes in line diameter. Simultaneously, an infrared spectrometer is installed at the end of the production line to analyze the reflectance spectral characteristics and detect coating film performance, including quality parameters such as thickness uniformity, surface smoothness, and insulation strength, forming a product quality dataset. The collected raw data contains noise and outliers, requiring data preprocessing. A 5-point sliding window midpoint filtering algorithm is applied to the temperature parameter set, viscosity parameter, tension parameter, and linear velocity to remove short-term noise interference, resulting in smoother data. Then, standardization is performed to convert parameters of different dimensions into a standard distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensional differences on subsequent analysis and forming a standardized parameter set.

[0045] Based on a standardized parameter set, a multivariate regression analysis was used to establish relationship graphs between the parameters. The temperature-viscosity graph describes the changes in the rheological properties of the paint at different temperatures; the viscosity-coating thickness graph shows the influence of paint viscosity on the final coating thickness; and the linear velocity-degree of cure graph reflects the changes in the degree of paint film curing at different linear velocities. By analyzing these graphs, the weighting coefficients of each parameter's influence on the final quality were determined, and a parameter influence matrix was constructed. The parameter influence matrix was combined with product quality data to construct a mapping function between process parameters and product quality. This mapping function can predict product quality indicators under given process parameters. The mapping accuracy was verified by comparing with historical production data. When the root mean square error between the predicted and actual measured values ​​was less than a preset threshold of 3%, the mapping function was confirmed as effective and determined as the digital model of the painting process for subsequent training and validation of deep reinforcement learning algorithms.

[0046] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0047] The temperature parameter set, viscosity parameter, tension parameter and linear velocity in the digital model of the coating process are combined into a state space, and a multi-dimensional state vector is constructed as the input data of the deep deterministic policy gradient network.

[0048] The temperature regulation amount, speed regulation amount, and paint supply amount are defined as the action space, and a safety limit range is set for each action. The action vector is constructed as the output target of the deep deterministic policy gradient network.

[0049] We weighted and combined the coating thickness, paint film performance and energy consumption in the product quality data to construct a reward function as the optimization objective of the deep deterministic policy gradient network.

[0050] A policy network structure is constructed based on state vectors, action vectors, and reward functions. The state is mapped to action through a multi-layer neural network. The input layer corresponds to the dimension of the state vector, the hidden layer uses an activation function to process feature extraction, and the output layer limits the range of actions.

[0051] A value network structure is constructed based on the policy network structure. By merging the inputs of state vectors and action vectors, the output policy evaluation value is used as a measure of policy quality, and update parameters are set to achieve smooth network updates.

[0052] Historical data from the digital model of the painting process is input into the experience replay buffer. The policy network and value network are trained through batch sampling. When the reward value is stable for several consecutive rounds and the loss of the value function drops below the threshold, the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are obtained.

[0053] Specifically, the temperature parameter set, viscosity parameter, tension parameter, and linear velocity in the digital model of the painting process are combined into a state space. Each parameter includes not only its current value but also its rate of change, forming a multi-dimensional state vector. For example, the temperature parameter set consists of the paint tank temperature, the temperature of each zone in the oven, and their rates of change; the viscosity parameter includes the viscosity value and its trend; the tension parameter includes the tension magnitude and fluctuation amplitude; and the linear velocity includes the current velocity value and acceleration. These parameters together constitute a 16-dimensional state vector, serving as the input data for the deep deterministic policy gradient network. The action space consists of temperature adjustment, speed adjustment, and paint supply, which control key process parameters in the painting process. To ensure production safety, strict safety limits are set for each action: temperature adjustment is limited to ±5℃, speed adjustment does not exceed ±10% of the current speed, and paint supply adjustment is within ±15%. These limits ensure that the control actions do not cause process abrupt changes or equipment damage, and are transformed into an 8-dimensional action vector as the network's output target.

[0054] The reward function is the optimization objective of the deep deterministic policy gradient network, constructed by weighting and combining product quality data. Specifically, it involves four aspects: coating thickness uniformity, paint film surface smoothness, insulation strength, and energy consumption. Coating thickness uniformity is measured by standard deviation; the smaller the standard deviation, the higher the reward. Paint film smoothness is measured by surface roughness; the lower the roughness, the higher the reward. Insulation strength directly reflects product quality; the higher the strength, the higher the reward. Energy consumption monitors the energy consumption per unit of product; the lower the energy consumption, the higher the reward. These four indicators are assigned different weights according to product type and importance, and are combined to form the reward value. The policy network structure is constructed based on the state vector, action vector, and reward function. The policy network is a four-layer fully connected neural network. The number of nodes in the input layer is the same as the dimension of the state vector, receiving 16-dimensional state data. The hidden layer consists of two layers: the first layer uses 128 neurons, and the second layer uses 64 neurons, both using the ReLU activation function for nonlinear feature extraction. The number of nodes in the output layer is the same as the dimension of the action vector, and the tanh activation function is used to limit the output to the range [-1, 1], which is then mapped to the actual action range through a linear transformation.

[0055] The value network structure is similar to the policy network, but the input layer combines state and action vectors, resulting in an input dimension of 24 (16-dimensional state + 8-dimensional action). The value network evaluates the value (Q-value) of a given combination of state and action, guiding the policy network update. A soft update mechanism is used, with an update rate of 0.001 to ensure smooth changes in the target network parameters and improve training stability. During training, historical data from the digital model of the painting process is input into an experience replay buffer with a capacity of 100,000 records. Each time, 64 records are randomly selected from the buffer to form a training batch. Each record contains the current state, the action performed, the reward obtained, and the next state. The TD error is calculated using these data to update the value network parameters, and the policy network update is guided based on the gradient of the value network. When the average reward value fluctuates by no more than 5% over 50 consecutive training rounds, and the value function loss drops below a preset threshold, the network training is considered converged, resulting in a trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions.

[0056] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] The historical data in the digital model of the painting process are sorted and segmented by time, and effective data sequences containing multiple working conditions are selected to build a fixed-capacity experience playback buffer as a training sample library.

[0058] Randomly sample batches of samples from the experience replay buffer. Each sample contains a state vector, an action vector, a reward value, and a next state vector to form a training batch dataset.

[0059] The state vector and the next state vector in the training batch dataset are input into the target policy network and the target value network, respectively, and the target Q value is calculated as the learning target of the value network.

[0060] Based on the mean squared error between the target Q value and the Q value predicted by the current value network, the loss function of the value network is calculated, and the parameters of the value network are updated by gradient descent to optimize the accuracy of value assessment.

[0061] Based on the gradient information of the value network, the policy network is updated with policy gradient. The update direction and step size of the policy network parameters are calculated by the deterministic policy gradient algorithm so that the action output by the policy network maximizes the value assessment.

[0062] The target policy network and target value network are soft-updated using a fixed update rate. The current network parameters are fused to ensure training stability. The degree of convergence is judged by monitoring the reward value and value function loss in consecutive rounds. When the reward value fluctuation range limit is met and the loss value is lower than the training termination threshold, the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are derived.

[0063] Specifically, the historical data in the digital model of the painting process is sorted and segmented by time. The historical data is sorted according to the collection timestamp to ensure its temporal sequence, and then divided into complete process cycles including the initial state, executed actions, reward acquisition, and ending state. The screening process emphasizes coverage of various operating conditions, including normal production conditions, process parameter fluctuation conditions, raw material switching conditions, and conditions with different product specifications, ensuring the comprehensiveness of the training data. These filtered valid data sequences are stored in an experience replay buffer with a fixed capacity of 100,000 records. Old data is overwritten by new data to maintain the timeliness of the sample database. When randomly sampling batches from the experience replay buffer, 64 records are drawn each time to form the training batch dataset. Each record contains four key elements: the current state vector (including temperature parameters, viscosity parameters, tension parameters, and linear velocity), the executed action vector (including temperature adjustment, speed adjustment, and paint supply), the obtained reward value (based on product quality score), and the next state vector. Random sampling rather than sequential use of data reduces the correlation between samples and improves training stability.

[0064] The state vectors and next-state vectors from the training batch dataset are input into the target policy network and the target value network, respectively, to calculate the target Q-value. The target policy network first calculates the optimal action for the next time step based on the next state. Then, the target value network evaluates the value of this state-action pair, combining the current reward with a discount factor (set to 0.99) to calculate the target Q-value. The discount factor balances the importance of immediate rewards versus future rewards; a value closer to 1 indicates a greater focus on long-term gains. This process follows the Bellman optimality equation principle, providing a learning objective for the value network.

[0065] The mean squared error loss function is calculated based on the difference between the target Q-value and the Q-value predicted by the current value network. The loss function expresses the degree of deviation between the predicted and target values ​​and is the optimization objective during network training. The gradient of the loss function with respect to the network parameters is calculated using gradient descent, and the value network parameters are updated using the Adam optimizer with a learning rate set to 0.001. The Adam optimizer combines momentum and adaptive learning rate methods, effectively handling gradient sparsity and noise issues, and accelerating the convergence process.

[0066] The policy network is updated based on the gradient information of the value network. Using a deterministic policy gradient algorithm, the current state is fixed, the gradient of value with respect to action is calculated, and then the gradient of action with respect to policy network parameters is calculated. The product of these two gradients yields the final policy gradient. The policy network parameters are updated in this gradient direction to maximize the value assessment of the actions output by the policy network. The core idea of ​​this process is: how to adjust the policy network parameters so that, given a state, it produces actions that yield the highest value assessment. The target network is updated using a soft update mechanism with a fixed update rate of 0.001. The new values ​​of the target network parameters are a weighted average of the current network parameters and the original target network parameters, with weights of 0.001 and 0.999, respectively. This soft update method smooths the changes in the objective function and avoids oscillations during training. During training, the average reward value and the value function loss value are continuously monitored in each training round. When the reward value fluctuation range does not exceed 5% in 50 consecutive training rounds, and the loss value is below the training termination threshold of 0.01, the network training is considered converged, and the final deep deterministic policy gradient network and its control instructions are derived.

[0067] In the application of coating control for enameled wire used in new energy vehicle motors, training samples containing 25 different operating conditions were extracted by screening three months of production data. During the training process, the network gradually learned the complex relationships between temperature and viscosity, and between linear velocity and coating thickness. After approximately 300 training rounds, the strategy network was able to automatically generate optimal combinations of control parameters for different states. When the paint viscosity fluctuates and the wire tension deviates from the standard value, the network coordinates and adjusts the temperature and linear velocity to maintain coating uniformity, demonstrating a multi-parameter collaborative control capability that traditional PID control cannot achieve. This effectively solves the problem of insufficient accuracy of traditional control methods under parameter coupling conditions.

[0068] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0069] The multi-parameter collaborative control command is parsed into temperature regulation, speed regulation and paint supply, and the control value is transmitted to the actuator through the industrial bus system to adjust the parameters of the painting production line in real time and obtain the adjusted painting process status.

[0070] The coating thickness, paint film uniformity and insulation strength data under the adjusted painting process state are collected in real time by the online detection system. Each quality indicator is compared with its corresponding early warning threshold at multiple levels to obtain the quality status assessment result.

[0071] Based on the quality status assessment results and historical quality change trends, the direction of quality change is predicted. When the prediction shows that any quality indicator deviates from the normal range, a feedforward control signal containing the intervention level and intervention parameters is generated.

[0072] Specifically, the control commands are parsed to extract three core parameters: temperature regulation, speed regulation, and paint supply. The parsing process employs inverse normalization, converting the normalized values ​​from the network output (range -1 to 1) into actual control ranges. For example, temperature regulation is converted to an actual change from -5℃ to +5℃, speed regulation to a change from -10% to +10% of the current speed, and paint supply to a supply adjustment range from -15% to +15%. The converted control values ​​are transmitted to the corresponding actuators, including temperature controllers, frequency converters, and precision metering pumps, via an industrial bus system (such as PROFINET or EtherCAT). Upon receiving the commands, these actuators adjust the corresponding process parameters in real time, thereby precisely controlling the painting production line and ultimately establishing the adjusted painting process state. For the adjusted painting process state, key quality indicators are collected in real time by an online monitoring system deployed at the end of the production line. Coating thickness was measured using a laser thickness gauge, recording the wire diameter variation. Coating uniformity was assessed using a high-speed camera and image processing algorithms to evaluate surface condition and calculate surface waviness. Insulation strength was measured using an online insulation tester to determine the dielectric breakdown voltage. Multi-level comparisons were performed on each collected quality indicator, comparing the measured values ​​with preset multi-level warning thresholds. These included green zone thresholds (normal working range), yellow zone thresholds (minor anomalies requiring attention), orange zone thresholds (moderate anomalies requiring intervention), and red zone thresholds (severe anomalies requiring immediate action). The comparison results formed a quality status assessment matrix, including the anomaly level and deviation trend of each indicator. Each quality indicator in the matrix was assigned a corresponding anomaly level label (0 indicates normal, 1-3 indicate different degrees of severity of anomaly).

[0073] Based on the quality status assessment matrix and historical quality change data, time series analysis is used to predict the direction of quality changes. The time series analysis employs an exponentially weighted moving average method, assigning different weights to quality data within a recent period, with more recent data receiving greater weight, to calculate the quality change trend. When the prediction indicates that any quality indicator is about to deviate from the normal range, a feedforward control mechanism is activated. The generation of the feedforward control signal first determines the intervention level based on the severity of the anomaly, categorized as minor intervention (fine-tuning only a single parameter), moderate intervention (coordinated adjustment of multiple parameters), and strong intervention (significantly adjusting parameters and reducing production speed). Then, based on similar cases in the historical anomaly handling experience database, the most effective intervention parameter and its adjustment direction and magnitude are determined. The generated feedforward control signal includes an intervention level code, target parameter identifier, and specific adjustment amount.

[0074] In the production of power transformer windings, when the varnish uniformity index is detected to gradually decrease from a stable value of 0.92 to 0.87 (the normal lower limit is 0.85), time series analysis predicts that this index will exceed the yellow zone threshold within the next 30 minutes. Simultaneously, historical data analysis shows that such a downward trend is usually related to fluctuations in varnish viscosity. The feedforward control mechanism generates a control signal that includes a moderate intervention level and a +2℃ adjustment suggestion for the varnish bath temperature, while also suggesting a 5% reduction in linear velocity to increase varnish film formation time. This precise and targeted multi-parameter coordinated adjustment scheme overcomes the limitation of traditional PID control, which can only passively respond to anomalies.

[0075] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0076] The statistical distribution characteristics of the coating thickness data collected in real time are calculated to obtain the mean and standard deviation of the thickness. These values ​​are then compared with the preset first-level warning threshold, second-level warning threshold and critical threshold to generate a thickness anomaly level identifier.

[0077] Spectral analysis is performed on the real-time collected paint film uniformity data to extract the surface fluctuation frequency and amplitude characteristics, and these are compared with preset uniformity level 1 warning threshold, level 2 warning threshold and critical threshold to generate uniformity anomaly level identifiers.

[0078] A time series model is established for the real-time collected insulation strength data, the trend of strength change is extracted, and it is compared with the preset first-level warning threshold, second-level warning threshold and critical threshold of insulation strength to generate an insulation strength anomaly level identifier;

[0079] The thickness anomaly level indicator, uniformity anomaly level indicator, and insulation strength anomaly level indicator are weighted together to calculate the comprehensive anomaly index;

[0080] A time-series cumulative analysis of the comprehensive anomaly index is conducted. When the anomaly index continues to rise and exceeds the first warning line, a mild intervention is triggered; when it exceeds the second warning line, a moderate intervention is triggered; and when it exceeds the third warning line, a strong intervention is triggered, thus forming a graded intervention decision.

[0081] Based on the tiered intervention decision-making and the processing strategies in the historical anomaly handling experience base, an intervention measure mapping table is constructed to assign corresponding intervention parameter adjustment amounts to each anomaly, thus completing the conversion of quality status assessment results into feedforward control signals.

[0082] Specifically, a laser thickness gauge collects thickness data from multiple points per second, calculating the mean and standard deviation of the thickness within a fixed time window (usually 5 minutes). The mean reflects the overall thickness level, while the standard deviation reflects the dispersion of the thickness distribution and is an important indicator of uniformity. The calculated mean is compared with the standard specification value, and the standard deviation is compared with the preset consistency requirements, and then compared with three levels of thresholds: Level 1 warning threshold (slight deviation, such as a standard deviation exceeding 1 micrometer), Level 2 warning threshold (moderate deviation, such as a standard deviation exceeding 3 micrometers), and critical threshold (severe deviation, such as a standard deviation exceeding 5 micrometers). Based on the comparison results, a thickness anomaly level label is generated, using a numerical code of 0-3 (0 for normal, 3 for the most severe). Spectral analysis of the paint film uniformity data is an effective method for identifying periodic defects. Surface image data acquired by a high-speed camera is converted to the frequency domain using a fast Fourier transform, extracting the main frequency components and corresponding amplitude characteristics of surface fluctuations. The main frequency reflects the spatial distribution pattern of defects, such as low frequencies indicating large-scale fluctuations and high frequencies indicating small defects; the amplitude reflects the severity of the defects. The extracted spectral features are compared with preset thresholds and are also divided into three levels: Level 1 warning (e.g., low-frequency fluctuation amplitude exceeds 0.5), Level 2 warning (e.g., low-frequency fluctuation amplitude exceeds 1.0), and critical value (e.g., low-frequency fluctuation amplitude exceeds 1.5), generating a uniformity anomaly level label.

[0083] The insulation strength data processing employs time series modeling, continuously collecting data through an online insulation tester and applying an autoregressive integral moving average model to analyze its changing trends. This model, fitted to historical data, can predict the direction and rate of change in insulation strength over a future period. The predicted results are compared with preset thresholds: Level 1 warning (e.g., predicting insulation strength below the standard value by 5% after 30 minutes), Level 2 warning (e.g., predicting insulation strength below the standard value by 10%), and critical value (e.g., predicting insulation strength below the standard value by 15%), generating an insulation strength anomaly level identifier. The anomaly level identifiers of the three quality indicators are comprehensively weighted, assigning different weights based on product type and application field. For example, transformer wire may prioritize insulation strength (weight 0.5), while motor wire may prioritize uniformity (weight 0.4) and thickness (weight 0.3). By multiplying the weights and summing them, a comprehensive anomaly index is calculated, which comprehensively reflects the overall product quality status.

[0084] Time-series cumulative analysis of the comprehensive anomaly index focuses not only on the current value but also on its changing trend. A mild intervention is triggered when the anomaly index rises for three consecutive cycles and exceeds the first warning line (0.5); a moderate intervention is triggered when it exceeds the second warning line (1.0); and a strong intervention is triggered when it exceeds the third warning line (1.5). Time-series cumulative analysis can filter out random fluctuations, identify continuously deteriorating trends, and form more stable and reliable tiered intervention decisions.

[0085] Finally, based on tiered intervention decision-making and historical anomaly handling experience, an intervention measure mapping table is constructed. The experience base stores historical anomalies and their successful handling solutions. Through similarity matching, the historical cases most similar to the current situation are found. Based on the differences in handling parameters in the cases and the current process state, the appropriate parameter adjustment amounts for the current situation are calculated, generating feedforward control signals. For example, when the insulation strength index shows a downward trend, the mapping table might suggest increasing the oven temperature by 2°C and decreasing the linear speed by 5% to increase curing time and degree.

[0086] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0087] The feedforward control signal is analyzed to extract the intervention level, intervention parameters and intervention direction, and the intervention parameters are converted into compensation vectors compatible with the multi-parameter collaborative control command format to obtain the control correction amount;

[0088] The multi-parameter collaborative control command and control correction are weighted and fused together. The weight coefficient of the feedforward control signal is set according to the intervention level to generate the fused control command, ensuring that the feedforward control priority is increased in the case of quality abnormality.

[0089] The fused control commands are combined with the current painting process state vector and input into a deep deterministic policy gradient network for online inference to obtain control parameters corrected by the network.

[0090] The control parameters after network calibration are subjected to safety verification to ensure that all control parameters are within the set safety boundaries, preventing abnormal control from causing equipment damage or product quality failure, and thus obtaining safe control parameters.

[0091] The safety control parameters are divided into immediate adjustment parameters and gradual adjustment parameters according to their degree of influence on the process. Immediate adjustment parameters are executed directly, while gradual adjustment parameters are smoothed using a ramp function to obtain a real-time control strategy.

[0092] Specifically, the feedforward control signal is analyzed to extract its three key components: intervention level (slight, moderate, or strong), intervention parameters (temperature, speed, or paint supply, etc.), and intervention direction (increase or decrease). The analysis process uses a bitmasking method to decompose the encoded feedforward control signal into different fields, each corresponding to different information. After analysis, the intervention parameters are mapped to compensation vectors consistent with the structure of the multi-parameter collaborative control command, maintaining dimensionality consistency and enabling direct vector operations between the original control command and the feedforward correction. The converted compensation vector is then weighted and fused with the multi-parameter collaborative control command using an adaptive weight allocation mechanism. The weight coefficients of the feedforward control signal are automatically set according to the intervention level: smaller weights for slight intervention (e.g., 0.3), moderate weights for moderate intervention (e.g., 0.5), and larger weights for strong intervention (e.g., 0.8). The weight value directly determines the degree of influence of the feedforward control on the final command; the larger the weight, the higher the priority of the feedforward control. The weighted fusion adopts a linear combination method, calculating the fusion value for each control parameter separately: fusion value = original command value × (1 - weight) + feedforward correction amount × weight, ensuring that feedforward control can intervene in a timely manner and play a leading role in the event of quality abnormalities.

[0093] The fused control commands are combined with the current painting process state vector to form complete input data, which is then fed into a deep deterministic policy gradient network for online inference. Unlike the training process, online inference only performs forward computation, without requiring backpropagation to update network parameters. By mapping the current state and the fused commands through the policy network, network-corrected control parameters are obtained, enabling further optimization of the fused commands. This step utilizes the process characteristics learned by the deep network to refine the feedforward control signal, making it more consistent with the process features.

[0094] After network calibration, the control parameters need to undergo safety verification to avoid extreme values ​​exceeding the equipment's tolerance. Safety verification includes three levels: parameter range checking (ensuring each parameter is within allowable limits), parameter change rate checking (limiting the magnitude of single adjustments), and parameter combination checking (verifying whether multiple parameter combinations conform to process specifications). When a parameter is detected to be out of bounds, truncation is used to confine it within safe boundaries while retaining directional information to ensure the control intent is not completely lost. Parameters that pass the safety verification are considered safe control parameters and can be applied to the production line.

[0095] Safety control parameters are categorized according to their impact on the process. Parameters with rapid equipment response and minimal impact (such as paint supply) are classified as immediate adjustment parameters, and execution commands are sent directly. Parameters with high equipment inertia and far-reaching impact (such as temperature) are classified as gradual adjustment parameters, and ramp functions are used to achieve smooth transitions. The ramp function is set with different rates of change based on the parameter properties, such as a maximum temperature change of 0.5℃ per minute and a maximum linear velocity change of 2% per minute, ensuring stable changes in process parameters and avoiding quality fluctuations caused by abrupt changes. This hierarchical execution method ultimately forms a real-time control strategy that is both responsive and reliable.

[0096] The above describes the dedicated line coating control method using intelligent algorithms in the embodiments of this application. The following describes the dedicated line coating control system using intelligent algorithms in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the intelligent algorithm-based dedicated line painting control system in this application includes:

[0097] The mapping module 201 is used to input sensor data from the painting production line into the digital twin system for multi-dimensional parameter mapping, thereby obtaining a digital model of the painting process that includes a set of temperature parameters, viscosity parameters, tension parameters, and linear velocity.

[0098] Input module 202 is used to construct a deep deterministic strategy gradient network structure based on the digital model of the painting process, input the state vector of the painting process into the deep deterministic strategy gradient network to calculate the control action, and obtain multi-parameter coordinated control commands for temperature adjustment, speed adjustment and paint supply.

[0099] The comparison module 203 is used to adjust the parameters of the painting production line in real time according to the multi-parameter collaborative control command, and at the same time perform multi-level comparison analysis between the painting quality data and the preset threshold to obtain the feedforward control signal.

[0100] The fusion module 204 is used to fuse the feedforward control signal with the multi-parameter collaborative control command, and to correct the control parameters according to the deep deterministic strategy gradient network to obtain a real-time control strategy.

[0101] above Figure 2 The dedicated line painting control system of the intelligent algorithm in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The dedicated line painting control device of the intelligent algorithm in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0102] Figure 3This is a schematic diagram of a dedicated line coating control device for intelligent algorithms provided in an embodiment of the present invention. The dedicated line coating control device 300 for intelligent algorithms can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the dedicated line coating control device 300 for intelligent algorithms. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the dedicated line coating control device 300 for intelligent algorithms to implement the steps of the dedicated line coating control method for intelligent algorithms described above.

[0103] The dedicated line painting control device 300 for intelligent algorithms may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the dedicated line coating control device for the intelligent algorithm shown does not constitute a limitation on the dedicated line coating control device for the intelligent algorithm provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0104] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the dedicated line painting control method of the intelligent algorithm.

[0105] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0106] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a dedicated line painting control device (which may be a personal computer, server, or network device, etc.) with an intelligent algorithm to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dedicated line painting control method based on an intelligent algorithm, characterized in that, The method includes: Sensor data from the coating production line is input into a digital twin system for multi-dimensional parameter mapping to obtain a digital model of the coating process that includes a set of temperature parameters, viscosity parameters, tension parameters, and linear velocity. Based on the digital model of the painting process, a deep deterministic strategy gradient network structure is constructed. The state vector of the painting process is input into the deep deterministic strategy gradient network to calculate the control action, and a multi-parameter coordinated control command for temperature adjustment, speed adjustment and paint supply is obtained. The parameters of the painting production line are adjusted in real time according to the multi-parameter collaborative control command. At the same time, the painting quality data is compared and analyzed with the preset threshold at multiple levels to obtain the feedforward control signal. The feedforward control signal and the multi-parameter collaborative control command are fused together, and the control parameters are corrected according to the deep deterministic strategy gradient network to obtain a real-time control strategy. This includes: parsing the feedforward control signal to extract the intervention level, intervention parameters, and intervention direction; converting the intervention parameters into a compensation vector compatible with the multi-parameter collaborative control command format to obtain a control correction amount; weighting and fusing the multi-parameter collaborative control command and the control correction amount; setting the weight coefficients of the feedforward control signal according to the intervention level to generate a fused control command, ensuring that the feedforward control priority is increased in the event of quality anomalies; combining the fused control command with the current painting process state vector and inputting it into the deep deterministic strategy gradient network for online inference to obtain network-corrected control parameters; performing a safety check on the network-corrected control parameters to ensure that all control parameters are within the set safety boundaries to prevent abnormal control from causing equipment damage or product quality collapse, thus obtaining safe control parameters; and dividing the safe control parameters into immediate adjustment parameters and gradual adjustment parameters according to the degree of process impact. Immediate adjustment parameters are executed directly, while gradual adjustment parameters are smoothly transitioned using a ramp function to obtain the real-time control strategy.

2. The dedicated line painting control method of the intelligent algorithm according to claim 1, characterized in that, The process involves inputting sensor data from the coating production line into a digital twin system for multi-dimensional parameter mapping to obtain a digital model of the coating process that includes a set of temperature parameters, viscosity parameters, tension parameters, and linear velocity. This includes: Temperature data is collected by temperature sensors installed at the paint tank, oven inlet, middle section and outlet of the painting production line to obtain a set of temperature parameters; The viscosity of the paint is measured by a viscosity sensor installed in the paint tank, and the tension data is collected by a tension sensor installed on the guide wheel to obtain viscosity and tension parameters. The wire diameter is measured by a laser diameter measuring instrument placed at the wire diameter control position, and the coating performance is detected by an infrared spectrometer at the end of the production line to obtain line speed and product quality data. Abnormal noise is removed from the temperature parameter set, the viscosity parameter, the tension parameter, and the linear velocity, and each parameter is standardized to obtain a standardized parameter set; Based on the standardized parameter set, establish the temperature-viscosity relationship spectrum, the viscosity-coating thickness relationship spectrum, and the linear velocity-curing degree relationship spectrum, and determine the weight coefficient of each parameter to obtain the parameter influence matrix; The parameter influence matrix is ​​combined with the product quality data to construct a mapping function between process parameters and product quality. The mapping accuracy is verified by comparing it with historical production data. When the verification error is less than a preset threshold, the mapping function is determined as the digital model of the painting process.

3. The dedicated line painting control method of the intelligent algorithm according to claim 2, characterized in that, The deep deterministic policy gradient network structure is constructed based on the digital model of the painting process. The painting process state vector is input into the deep deterministic policy gradient network to calculate the control actions, resulting in multi-parameter coordinated control commands for temperature adjustment, speed adjustment, and paint supply. The temperature parameter set, viscosity parameter, tension parameter and linear velocity in the digital model of the coating process are combined into a state space, and a multi-dimensional state vector is constructed as the input data of the deep deterministic strategy gradient network. The temperature regulation amount, speed regulation amount, and paint supply amount are defined as the action space, and a safety limit range is set for each action. The action vector is constructed as the output target of the deep deterministic policy gradient network. The coating thickness, paint film performance, and energy consumption in the product quality data are weighted and combined to construct a reward function as the optimization objective of the deep deterministic policy gradient network; A policy network structure is constructed based on the state vector, the action vector, and the reward function. The state is mapped to the action through a multi-layer neural network, where the input layer corresponds to the dimension of the state vector, the hidden layer uses an activation function to process feature extraction, and the output layer limits the range of actions. Based on the policy network structure, a value network structure is constructed. By merging the inputs of state vectors and action vectors, the policy evaluation value is output as a measure of policy quality, and update parameters are set to achieve smooth network updates. Historical data from the digital model of the painting process is input into the experience replay buffer. The policy network and value network are trained by batch sampling. When the reward value is stable for several consecutive rounds and the loss of the value function drops below the threshold, the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are obtained.

4. The dedicated line painting control method of the intelligent algorithm according to claim 3, characterized in that, The process involves inputting historical data from the digital model of the painting process into an experience replay buffer, training the policy network and the value network through batch sampling, and obtaining a trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions when the reward value stabilizes for multiple consecutive rounds and the value function loss falls below a threshold. This includes: The historical data in the digital model of the painting process is sorted by time and segmented to filter out effective data sequences containing multiple working conditions, and a fixed-capacity experience playback buffer is constructed as a training sample library. Randomly select batch samples from the experience replay buffer. Each sample contains a state vector, an action vector, a reward value, and a next state vector to form a training batch dataset. The state vector and the next state vector in the training batch dataset are input into the target policy network and the target value network, respectively, and the target Q value is calculated as the learning target of the value network. Based on the mean squared error between the target Q value and the Q value predicted by the current value network, the loss function of the value network is calculated, and the parameters of the value network are updated by gradient descent to optimize the accuracy of value assessment. Based on the gradient information of the value network, the policy network is updated with policy gradient. The update direction and step size of the policy network parameters are calculated by a deterministic policy gradient algorithm so that the action output by the policy network maximizes the value assessment. The target policy network and the target value network are soft-updated using a fixed update rate. The current network parameters are fused to ensure training stability. The degree of convergence is judged by monitoring the reward value and value function loss in consecutive rounds. When the reward value fluctuation range limit is met and the loss value is lower than the training termination threshold, the trained deep deterministic policy gradient network and its generated multi-parameter collaborative control instructions are derived.

5. The dedicated line painting control method of the intelligent algorithm according to claim 1, characterized in that, The process involves real-time parameter adjustment of the coating production line based on multi-parameter collaborative control commands, and simultaneous multi-level comparison and analysis of coating quality data with preset thresholds to obtain feedforward control signals, including: The multi-parameter collaborative control command is parsed into temperature adjustment amount, speed adjustment amount and paint supply amount. The control value is transmitted to the actuator through the industrial bus system to adjust the parameters of the painting production line in real time and obtain the adjusted painting process status. The coating thickness, paint film uniformity and insulation strength data under the adjusted painting process state are collected in real time by the online detection system. Each quality indicator is compared with its corresponding early warning threshold in multiple levels to obtain the quality status assessment result. Based on the quality status assessment results and historical quality change trends, the direction of quality change is predicted. When the prediction shows that any quality indicator deviates from the normal range, a feedforward control signal containing the intervention level and intervention parameters is generated.

6. The dedicated line painting control method of the intelligent algorithm according to claim 5, characterized in that, The process of comparing each quality indicator with its corresponding early warning threshold at multiple levels includes: The statistical distribution characteristics of the coating thickness data collected in real time are calculated to obtain the mean and standard deviation of the thickness. These values ​​are then compared with the preset first-level warning threshold, second-level warning threshold and critical threshold to generate a thickness anomaly level identifier. Spectral analysis is performed on the real-time collected paint film uniformity data to extract the surface fluctuation frequency and amplitude characteristics, and the results are compared with preset uniformity level 1 warning threshold, level 2 warning threshold and critical threshold to generate uniformity anomaly level identifiers. A time series model is established for the real-time collected insulation strength data, the trend of strength change is extracted, and it is compared with the preset first-level warning threshold, second-level warning threshold and critical threshold of insulation strength to generate an insulation strength anomaly level identifier; The thickness anomaly level indicator, the uniformity anomaly level indicator, and the insulation strength anomaly level indicator are weighted together to calculate a comprehensive anomaly index; A time-series cumulative analysis is performed on the comprehensive anomaly index. When the anomaly index continues to rise and exceeds the first warning line, a slight intervention is triggered; when it exceeds the second warning line, a moderate intervention is triggered; and when it exceeds the third warning line, a strong intervention is triggered, thus forming a graded intervention decision. Based on the processing strategies in the hierarchical intervention decision-making and historical anomaly handling experience base, an intervention measure mapping table is constructed, and corresponding intervention parameter adjustment amounts are assigned to each anomaly situation to complete the conversion of quality status assessment results into feedforward control signals.

7. A dedicated line painting control system based on an intelligent algorithm, characterized in that, A dedicated line coating control method for implementing the intelligent algorithm as described in any one of claims 1-6, wherein the dedicated line coating control system for the intelligent algorithm comprises: The mapping module is used to input sensor data from the painting production line into the digital twin system for multi-dimensional parameter mapping, resulting in a digital model of the painting process that includes temperature parameters, viscosity parameters, tension parameters, and linear velocity. The input module is used to construct a deep deterministic strategy gradient network structure based on the digital model of the painting process, input the state vector of the painting process into the deep deterministic strategy gradient network to calculate the control action, and obtain multi-parameter coordinated control commands for temperature adjustment, speed adjustment and paint supply. The comparison module is used to adjust the parameters of the painting production line in real time according to the multi-parameter collaborative control instructions, and at the same time, compare and analyze the painting quality data with the preset threshold to obtain the feedforward control signal. The fusion module is used to fuse the feedforward control signal and the multi-parameter collaborative control command, and to correct the control parameters according to the deep deterministic strategy gradient network to obtain a real-time control strategy. This includes: parsing the feedforward control signal, extracting the intervention level, intervention parameters, and intervention direction, and converting the intervention parameters into a compensation vector compatible with the multi-parameter collaborative control command format to obtain a control correction amount; weighting and fusing the multi-parameter collaborative control command and the control correction amount, setting the weight coefficient of the feedforward control signal according to the intervention level, and generating a fused control command to ensure that the feedforward control priority is increased in the event of quality anomalies; combining the fused control command with the current painting process state vector, inputting it into the deep deterministic strategy gradient network for online inference to obtain network-corrected control parameters; performing a safety check on the network-corrected control parameters to ensure that all control parameters are within the set safety boundaries to prevent abnormal control from causing equipment damage or product quality collapse, thus obtaining safe control parameters; and dividing the safe control parameters into immediate adjustment parameters and gradual adjustment parameters according to the degree of process impact, executing the immediate adjustment parameters directly, and using a ramp function for smooth transition of the gradual adjustment parameters to obtain the real-time control strategy.

8. A dedicated line painting control device based on an intelligent algorithm, characterized in that, The method includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the intelligent algorithm of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the dedicated line painting control method of the intelligent algorithm as described in any one of claims 1 to 6.

Images