Hot galvanizing coating intelligent control system

Abstract

The invention relates to the technical field of automatic control, particularly discloses an intelligent control system for hot galvanizing coating, and aims to solve the problems of large coating thickness fluctuation, high zinc consumption and poor quality stability. The system comprises a multi-modal sensing acquisition module, a dynamic coupling modeling module, a distributed parallel decision module and a multi-target collaborative execution module, by collecting process data in real time, establishing a non-linear state space model, generating a multi-target optimization decision in parallel and outputting a final control instruction based on dynamic weight fusion, accurate control of plating quality and improvement of production energy efficiency are realized.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, specifically relating to an intelligent control system for hot-dip galvanizing. Background Technology

[0002] In the field of metal surface treatment technology, hot-dip galvanizing is a highly efficient and economical corrosion-resistant process widely used in industries such as construction, automobiles, and home appliances. This process forms a zinc-iron alloy layer on the surface of a steel substrate, effectively isolating the substrate from corrosive media and significantly extending the material's service life. The hot-dip galvanizing process involves multiple steps, including pretreatment, galvanizing, and cooling. Precise control of its process parameters has a decisive impact on coating quality and production efficiency.

[0003] Intelligent control of the hot-dip galvanizing process is a key technological direction for improving coating uniformity, reducing zinc consumption, and achieving green production. This technology aims to achieve precise control over coating thickness and microstructure by real-time monitoring and dynamic adjustment of core process parameters such as galvanizing temperature, immersion time, and air knife pressure, thereby meeting the differentiated performance requirements of coatings in different application scenarios.

[0004] Existing technologies mostly employ single-loop control systems based on traditional PID control, which are difficult to cope with the complex dynamic characteristics of strong nonlinearity, large time delay, and multivariable coupling in the galvanizing process; the coordinated control effect between air knife pressure, strip speed, and zinc liquid temperature is poor, which can easily lead to large fluctuations in coating thickness, excessively thick edges, or zinc wave defects.

[0005] Meanwhile, the existing system lacks the ability to deeply explore the mapping relationship between process parameters and coating quality, and cannot perform adaptive optimization according to real-time operating conditions, resulting in high zinc consumption and difficulty in ensuring quality stability.

[0006] In addition, traditional control systems are slow to respond to changes in production rhythm, and are prone to producing defective products during transition periods when switching specifications or adjusting speeds, which seriously affects production efficiency and cost control.

[0007] Therefore, there is an urgent need to develop an intelligent control system for hot-dip galvanizing coating that can achieve intelligent collaborative optimization of multiple parameters and has self-learning and self-adaptive capabilities. Summary of the Invention

[0008] The technical problem this invention aims to solve is to overcome the shortcomings of existing hot-dip galvanizing coating control processes, such as large fluctuations in coating thickness, high zinc consumption, poor quality stability, and insufficient adaptability to production rhythm caused by strong nonlinearity, large time delay, and multivariate coupling. The purpose of this invention is to provide an intelligent control system for hot-dip galvanizing coating. By constructing a full-process intelligent control architecture that integrates multimodal perception, dynamic coupling modeling, distributed parallel decision-making, and multi-objective collaborative optimization, this system achieves precise and stable control of coating quality and a significant improvement in production efficiency.

[0009] The technical solution of this invention is an intelligent control system for hot-dip galvanizing, comprising: The multimodal sensing and acquisition module is used to collect data in real time on strip speed, zinc liquid temperature, air knife pressure, air knife distance, strip tension, and temperature distribution of the plated surface after plating during the production line operation. The dynamic coupling modeling module is coupled to the output of the multimodal sensing and acquisition module. It is used to establish a nonlinear state-space model describing the dynamic coupling relationship between air knife pressure, strip speed, zinc liquid temperature and coating thickness based on the real-time data provided by the multimodal sensing and acquisition module. The distributed parallel decision-making module is coupled to the output of the dynamic coupling modeling module. It is used to receive the state prediction information output by the dynamic coupling modeling module and perform parallel calculations based on multiple preset performance indicators to generate multiple sets of optimized decision schemes for the air knife pressure setpoint, strip speed setpoint, and zinc liquid temperature setpoint. The multi-objective collaborative execution module is coupled to the output of the distributed parallel decision-making module. It is used to comprehensively evaluate and integrate multiple sets of optimized decision schemes output by the distributed parallel decision-making module, generate the final execution instructions, and send them to the corresponding execution mechanism.

[0010] Preferably, the multimodal sensing acquisition module includes a velocity detection unit, a temperature detection unit, a pressure detection unit, a distance detection unit, a tension detection unit, and an area array infrared temperature measurement unit; The speed detection unit uses a laser velocimeter to measure the running speed of the strip steel in real time; The temperature detection unit uses an immersion thermocouple array to measure the temperature of the molten zinc in different areas inside the zinc pot; The pressure detection unit uses a high-precision pressure transmitter to monitor the pressure values ​​of the main air supply pipe and the branch pipes on both sides of the air knife. The distance detection unit uses a laser displacement sensor to detect the distance between the air knife lip and the strip surface in real time; The tension detection unit acquires real-time tension data of the strip steel through the tension sensor attached to the tension roller; An area array infrared temperature measurement unit is installed at the inlet of the post-plating cooling section to acquire two-dimensional temperature field distribution data of the strip surface.

[0011] Preferably, the nonlinear state-space model constructed by the dynamic coupling modeling module includes four state variables, three control inputs, and two main outputs; The four state variables are coating thickness deviation, strip speed change rate, zinc liquid temperature gradient, and air knife pressure fluctuation. The three control inputs are the air knife pressure setting value, the strip speed setting value, and the zinc liquid temperature setting value; The two main outputs are the predicted coating thickness and the predicted coating uniformity index.

[0012] Preferably, the model building process of the dynamic coupling modeling module includes: Based on historical production data and real-time sensor data, the recursive least squares method is used to identify model parameters online. By introducing an adaptive mechanism with a forgetting factor, the model is able to track the slow time-varying characteristics of process conditions. In addition, Kalman filtering is used to make optimal estimates of the model state in order to suppress the interference of measurement noise.

[0013] Preferably, the distributed parallel decision-making module consists of three parallel decision sub-modules: the energy efficiency optimal decision sub-module, the quality optimal decision sub-module, and the stability optimal decision sub-module. The energy efficiency optimal decision-making submodule performs optimization calculations with the objective function of minimizing zinc consumption per unit area. The optimal quality decision submodule performs optimization calculations with the objective function of minimizing the weighted sum of coating thickness deviation and uniformity index; The stability-optimal decision-making submodule performs optimization calculations with the objective function of minimizing the sum of squared rates of change of key process parameters.

[0014] Preferably, each decision submodule adopts a model predictive control framework, which generates an optimized sequence of air knife pressure setpoints, strip speed setpoints, and zinc liquid temperature setpoints by solving a rolling optimization problem in a finite time domain.

[0015] Preferably, the decision fusion mechanism of the multi-objective collaborative execution module includes: A dynamic weighting coefficient is assigned to the decision schemes output by the energy efficiency optimal decision submodule, the quality optimal decision submodule, and the stability optimal decision submodule; the dynamic weighting coefficient is adaptively adjusted according to the current production stage and target priority. During the normal steady-state production phase, the weight coefficient of the quality-optimal decision-making submodule is set to 0.5, and the weight coefficients of the energy-optimal decision-making submodule and the stability-optimal decision-making submodule are each set to 0.25. During specification switching or acceleration / deceleration transition phases, the weight coefficient of the stability-optimal decision submodule is increased to 0.5, the weight coefficient of the quality-optimal decision submodule is adjusted to 0.3, and the weight coefficient of the energy-optimal decision submodule is adjusted to 0.2. Finally, a linear weighted average was performed on the weighted decision schemes to calculate the final air knife pressure setting, strip speed setting, and zinc liquid temperature setting.

[0016] Preferably, it also includes a process knowledge base and an adaptive learning engine; The process knowledge base is used to store historical optimal combinations of process parameters, process window limitations for different steel grades and specifications, and characteristic data and coping strategies for various defects; The adaptive learning engine is used to iteratively optimize and update the parameters of the dynamically coupled modeling module, the objective function weights of the distributed parallel decision-making module, and the dynamic weight coefficients of the multi-objective collaborative execution module based on the actual results of each production task.

[0017] Preferably, the update strategy of the adaptive learning engine adopts a gradient descent-based reinforcement learning algorithm, with the goal of maximizing cumulative production benefits in the long term.

[0018] Preferably, a smooth instruction transition unit is provided between the multi-target collaborative execution module and the underlying execution mechanism; The instruction smooth transition unit is used to ramp up the step changes of the set value. Its ramp rate is adaptively calculated according to the strip speed and specification changes to ensure smooth operation of the actuator and avoid impact on the strip running and coating formation process.

[0019] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention fundamentally solves the control problems caused by strong nonlinearity, large time delay and multivariable coupling in the hot-dip galvanizing process by constructing a full-process intelligent control architecture that integrates multimodal perception, dynamic coupling modeling, distributed parallel decision-making and multi-objective collaborative optimization. The dynamic coupling modeling module can accurately describe the complex dynamic relationship between key process parameters and coating quality indicators, providing a reliable model basis for advanced control strategies.

[0020] 2. The distributed parallel decision-making mechanism adopted in this invention achieves specialized optimization and collaborative consideration of different performance objectives through the parallel operation of three decision sub-modules: optimal energy efficiency, optimal quality, and optimal stability. Combined with the dynamic weight allocation strategy based on production stage adaptation in the multi-objective collaborative execution module, the system can flexibly balance zinc consumption and production stability while ensuring the core requirements of coating quality, significantly improving the overall operating efficiency of the system.

[0021] 3. The process knowledge base and adaptive learning engine introduced in this invention enable the system to continuously evolve, learn from historical data and real-time operating results and optimize its own model parameters and control strategies, thereby gradually adapting to changes in production line equipment characteristics and process requirements of new product specifications. This achieves a leap from static control to dynamic adaptation, effectively ensuring the stability of quality control and production efficiency under long-term operation.

[0022] 4. The present invention effectively mitigates the impact during the setting value adjustment process through the design of the instruction smooth transition unit. Especially in the specification switching and speed adjustment stages, it significantly reduces the generation of defective products in the transition section, improves production continuity and yield, and at the same time protects the actuator and extends the service life of the equipment. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention.

[0024] Figure 2 This is a schematic diagram of the interaction relationship and data flow of the distributed parallel decision-making and multi-objective collaborative execution module of the present invention. Detailed Implementation

[0025] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0026] Please refer to the attached document. Figure 1 This embodiment details an intelligent control system for hot-dip galvanizing, aiming to solve the technical challenges of large coating thickness fluctuations, high zinc consumption, poor quality stability, and insufficient adaptability to production rhythm caused by strong nonlinearity, large time delay, and multivariate coupling in the hot-dip galvanizing coating control process. It mainly achieves precise and stable control of coating quality and a significant improvement in production efficiency by constructing a full-process intelligent control architecture that integrates multimodal perception, dynamic coupling modeling, distributed parallel decision-making, and multi-objective collaborative optimization.

[0027] Specifically, it includes a multimodal sensing acquisition module, a dynamic coupling modeling module, a distributed parallel decision-making module, and a multi-objective collaborative execution module. The output of the multimodal sensing acquisition module is coupled to the dynamic coupling modeling module, the output of the dynamic coupling modeling module is coupled to the distributed parallel decision-making module, and the output of the distributed parallel decision-making module is coupled to the multi-objective collaborative execution module.

[0028] The multimodal sensing acquisition module is used to collect real-time data on strip speed, zinc liquid temperature, air knife pressure, air knife distance, strip tension, and post-plating surface temperature distribution during production line operation. The dynamic coupling modeling module is used to establish a nonlinear state-space model describing the dynamic coupling relationship between air knife pressure, strip speed, zinc liquid temperature, and coating thickness based on the real-time data provided by the multimodal sensing acquisition module. The distributed parallel decision-making module is used to receive the state prediction information output by the dynamic coupling modeling module and perform parallel calculations based on multiple preset performance indicators to generate multiple sets of optimized decision schemes for air knife pressure setpoints, strip speed setpoints, and zinc liquid temperature setpoints. The multi-objective collaborative execution module is used to comprehensively evaluate and integrate the multiple sets of optimized decision schemes output by the distributed parallel decision-making module, generate the final execution command, and send it to the corresponding execution mechanism.

[0029] The multimodal sensing acquisition module is the foundation of the system's perception layer, and it includes a velocity detection unit, a temperature detection unit, a pressure detection unit, a distance detection unit, a tension detection unit, and an area array infrared temperature measurement unit.

[0030] The speed detection unit uses a laser velocimeter to measure the strip speed in real time. The laser velocimeter is installed after the looper at the entrance of the galvanizing line. The laser beam emitted by the velocimeter is irradiated onto the surface of the strip at a specific angle. By analyzing the Doppler frequency shift of the reflected laser, the instantaneous linear velocity of the strip is accurately calculated. The calculated speed data is transmitted to the data acquisition card via industrial Ethernet at a rate of one hundred samples per second. The data acquisition card filters and denoises the raw signal to remove measurement noise introduced by strip jitter or changes in surface reflectivity, and finally generates a real-time digital value of the strip speed.

[0031] The temperature detection unit uses an immersion thermocouple array to measure the temperature of the molten zinc in different areas inside the zinc pot. The thermocouple array consists of twelve K-type armored thermocouples arranged in a matrix of three rows and four columns inside the zinc pot. Six thermocouples are located in the middle area of ​​the zinc pot, three are located at the inlet side, and three are located at the outlet side. The measuring end of each thermocouple is encapsulated in a ceramic protective sleeve resistant to zinc corrosion. Its lead is connected to a multi-channel temperature transmitter through a high-temperature compensation wire. The temperature transmitter amplifies the microvolt-level potential signal generated by each thermocouple and converts it into a standard current signal of four to twenty milliamps. Then, it is sent to the programmable logic controller (PLC) via an analog input module. The PLC has a built-in temperature gradient calculation algorithm. By comparing the readings of thermocouples at different locations, it calculates the temperature gradient of the molten zinc in the length and width directions in real time. This gradient data serves as an important indicator of the thermal uniformity of the molten zinc.

[0032] The pressure detection unit uses high-precision pressure transmitters to monitor the pressure values ​​of the main air supply pipe and the two side branches of the air knife. An absolute pressure transmitter with a range of 0 to 600 kPa and an accuracy of 0.1 percent is installed on the main air supply pipe of the air knife. A gauge pressure transmitter with a range of 0 to 200 kPa and an accuracy of 0.05 percent is installed on the left and right side branches of the air knife. All pressure transmitters are equipped with pulse dampers to suppress reading fluctuations caused by airflow pulsation. The 4 to 20 mA signal output by the pressure transmitters is transmitted to the distributed control system through a shielded cable. The distributed control system compares the pressure of the left and right branches in real time. If the pressure difference between the two sides exceeds 5 Pa, the air knife lateral pressure equalization adjustment mechanism is triggered.

[0033] The distance detection unit uses a laser displacement sensor to detect the distance between the air knife lip and the strip surface in real time. Two laser triangulation displacement sensors are installed on the left and right sides of the air knife device, respectively. Their measuring beams are perpendicular to the strip surface. The sensors emit lasers with a wavelength of 650 nanometers. The displacement of the reflected light spot is detected by the built-in position sensitive device, thereby calculating the absolute distance from the air knife lip to the strip surface. The measurement data is uploaded to the industrial control computer at a frequency of 100 Hz through the RS485 serial interface. The industrial control computer performs an arithmetic average of the distance measurement values ​​on the left and right sides and compares the average value with the set distance value to generate a distance deviation signal for closed-loop control of the air knife height.

[0034] The tension detection unit acquires real-time tension data of the strip steel through a tension sensor attached to the tension roller. The tension sensor adopts the strain gauge force measurement principle, and its elastic body structure is directly integrated into the bearing housing of the tension roller. When the tension of the strip steel acts on the tension roller, the elastic body generates micro-strain. The full-bridge strain gauge attached to it converts the strain into a change in resistance, which is then converted into a millivolt-level voltage signal through a Wheatstone bridge. The millivolt-level voltage signal is amplified by an instrumentation amplifier and digitized by a 24-bit high-resolution analog-to-digital converter. The digitized tension data is periodically transmitted to the main controller via the PROFIBUS DP fieldbus. The main controller performs specification normalization processing on the original tension value according to the material, width, and thickness of the strip steel to obtain the tension value per unit width for subsequent modeling and control.

[0035] An area-array infrared thermography unit is installed at the inlet of the post-plating cooling section to acquire two-dimensional temperature field distribution data of the strip surface. The core component of the area-array infrared thermography unit is a 320x240 pixel uncooled microbolometer infrared focal plane array with a spectral response range of 8 to 14 micrometers, a temperature measurement range of 100 to 600 degrees Celsius, and a thermal sensitivity better than 0.1 degree Celsius. A compressed air purging window is installed in front of the infrared lens to continuously purge and prevent zinc vapor condensation from contaminating the lens. The infrared thermography unit acquires thermal images of the strip surface at a rate of 25 frames per second. Each frame of thermal image contains 76,800 independent temperature pixels. The acquired raw thermal image data is transmitted to a dedicated image processing computer via gigabit Ethernet. The image processing computer first performs non-uniformity correction to eliminate differences in the detector's own pixel response, then performs emissivity compensation, converting the measured radiation value into the true temperature value according to the emissivity curve of the strip surface, and finally removes random noise through a spatial filtering algorithm. The average, maximum, minimum, and standard deviation of the temperature of the entire image are calculated to form a feature dataset of the temperature distribution of the plated surface after plating.

[0036] The dynamic coupling modeling module is the core analysis engine of the entire system. Its function is to establish and maintain a nonlinear state-space model that can accurately describe the dynamic coupling relationship between air knife pressure, strip speed, zinc liquid temperature and coating thickness based on the real-time data provided by the multimodal sensing acquisition module.

[0037] The core of the dynamic coupling modeling module lies in constructing a nonlinear state-space model containing four state variables, three control inputs, and two main outputs. The four state variables are coating thickness deviation, strip speed change rate, zinc bath temperature gradient, and air knife pressure fluctuation; the three control inputs are air knife pressure setpoint, strip speed setpoint, and zinc bath temperature setpoint; and the two main outputs are predicted coating thickness and predicted coating uniformity index.

[0038] The establishment and updating of the dynamic coupling modeling module is a continuous and dynamic process. First, the dynamic coupling modeling module retrieves historical data from the production of the most recent 1,000 coils of strip steel from the process knowledge base. This data includes the set values ​​of each process parameter, the actual measured values, and the coating thickness values ​​obtained by the final thickness gauge.

[0039] Simultaneously, the dynamic coupling modeling module receives real-time data from the multimodal sensing acquisition module, including strip speed, zinc bath temperature, air knife pressure, air knife distance, strip tension, and post-plating surface temperature distribution. Based on this historical and real-time data, the recursive least squares method is used to identify model parameters online. The core recursive formula of the recursive least squares method is as follows: ; Where k represents the current sampling time, P is the covariance matrix of the parameter estimation error, φ is the data vector containing the state variables and control input, and λ is the forgetting factor.

[0040] The core recursive formula is executed once in each control cycle, typically one hundred milliseconds. By introducing an adaptive mechanism with a forgetting factor, the dynamic coupling modeling module can ensure that it can track the slow time-varying characteristics of process conditions, such as the gradual change in the composition of zinc liquid in the zinc pot or the slight wear of the air knife nozzle. The typical value of the forgetting factor λ is between 0.98 and 0.995. The system dynamically fine-tunes the prediction error based on the adaptive covariance of the prediction error of the dynamic coupling modeling module. If the prediction error continues to increase, the value of λ is automatically reduced to improve the model update speed; if the prediction error stabilizes at a low level, the value of λ is increased to enhance the model's noise resistance.

[0041] Finally, the dynamic coupling modeling module uses Kalman filtering to make an optimal estimate of the model state in order to suppress the interference of measurement noise. The Kalman filter includes two steps: prediction and update.

[0042] In the prediction step, the state at the current moment is predicted using the state-space model based on the state estimate from the previous moment and the current control input. In the update step, the predicted state is compared with the current actual measurement value, and the predicted state is corrected by calculating the Kalman gain matrix to obtain the optimal state estimate.

[0043] This process runs continuously to ensure that the model state is always an unbiased estimate that is closest to the actual process state. The predicted coating thickness output by the model is a scalar value in grams per square meter, which represents the final weight of the coating formed after the strip leaves the zinc pot under the current process parameters. The predicted coating uniformity index is a dimensionless number between zero and one. Its calculation is based on the distribution of the predicted coating thickness in the width direction of the strip. The closer the value is to one, the better the coating uniformity. These two output values ​​are used as key quality prediction indicators and are directly fed into the distributed parallel decision module.

[0044] The distributed parallel decision-making module is responsible for generating multiple sets of control decision schemes for different optimization objectives based on the state prediction information provided by the dynamic coupling modeling module.

[0045] Please refer to the attached document. Figure 2 The distributed parallel decision-making module consists of three parallel decision-making sub-modules: the energy efficiency optimal decision-making sub-module, the quality optimal decision-making sub-module, and the stability optimal decision-making sub-module.

[0046] These three decision submodules can be deployed on different computing cores of the same industrial server or distributed in three independent embedded industrial control computers. They exchange and synchronize data through shared memory or high-speed industrial networks. Each decision submodule adopts a model predictive control framework, which generates an optimized sequence of air knife pressure setpoints, strip speed setpoints, and zinc liquid temperature setpoints by solving a rolling optimization problem in a finite time domain.

[0047] The energy efficiency optimization decision submodule performs optimization calculations with the objective function of minimizing zinc consumption per unit area. The objective function is mathematically expressed as minimizing the total amount of zinc consumed per unit area of ​​the strip within the prediction time domain. This optimization calculation not only considers the zinc liquid blown away by the air knife, but also comprehensively considers the evaporation loss of the zinc pot and the efficiency of the strip carrying away the zinc liquid. The optimization process is subject to a series of constraints, including that the coating thickness must meet the lower and upper limits of the product specifications, the air knife pressure must not exceed the equipment safety limit, the strip speed must be within the maximum design speed range of the production line, and the zinc liquid temperature must be maintained above the zinc liquid solidification point and excessive oxidation must be avoided.

[0048] The optimal quality decision submodule optimizes the calculation by minimizing the weighted sum of coating thickness deviation and uniformity index. The coating thickness deviation is defined as the absolute value of the difference between the predicted coating thickness and the target thickness. The uniformity index is directly adopted from the output of the dynamic coupling modeling module. The optimization focus of the optimal quality decision submodule is to ensure that the coating thickness accurately reaches the set value and is distributed as uniformly as possible across the entire board surface. In addition to the physical limits of the equipment, its constraints particularly emphasize that the coating uniformity index must not be lower than 0.85, which is the red line for process quality.

[0049] The stability-optimal decision submodule performs optimization calculations with the objective function of minimizing the sum of squared rates of change of key process parameters, including air knife pressure setpoint, strip speed setpoint, and zinc liquid temperature setpoint. The objective function aims to minimize the variation of these setpoints in adjacent control cycles, thereby ensuring the smoothness of the production process and reducing the impact on mechanical actuators and thermal systems. It is particularly suitable for the transitional stage of strip specification switching or production speed adjustment. Its constraints ensure that the changes in process parameters are smooth and continuous, avoiding step-like jumps in setpoints.

[0050] Each decision submodule's model predictive control algorithm operates on a 250-millisecond control cycle, with a prediction time domain length of 20 cycles (5 seconds). At the start of each control cycle, each decision submodule obtains the latest state prediction from the dynamic coupling modeling module and then independently solves its own optimization problem. The optimization problem is solved using the interior-point method, which transforms the constrained optimization problem into a series of unconstrained optimization problems by introducing a barrier function and then rapidly converges to the optimal solution using Newton's iteration method. After solving, each decision submodule outputs an optimized sequence of air knife pressure setpoints, strip speed setpoints, and zinc liquid temperature setpoints for the next five sampling cycles. However, only the current setpoint is submitted to the multi-objective collaborative execution module for fusion.

[0051] The core function of the multi-objective collaborative execution module is to comprehensively evaluate and fuse the three sets of optimized decision schemes output by the distributed parallel decision module to generate the final execution instruction. The decision fusion mechanism of the multi-objective collaborative execution module is as follows: First, a dynamic weight coefficient is assigned to the decision schemes output by the energy efficiency optimal decision submodule, the quality optimal decision submodule, and the stability optimal decision submodule. This dynamic weight coefficient is not fixed, but is adaptively adjusted according to the current production stage and target priority. The system divides the production stage into a normal steady-state production stage and a specification switching or acceleration / deceleration transition stage.

[0052] During the normal steady-state production phase, that is, after the strip steel specifications and speed have been running stably for more than 30 seconds, the weight coefficient of the quality optimal decision submodule is set to 0.5, and the weight coefficients of the energy efficiency optimal decision submodule and the stability optimal decision submodule are each set to 0.25.

[0053] During the specification switching or acceleration / deceleration transition phase, that is, within a 60-second time window after the system detects a change in target specification or speed command, the weight coefficient of the stability-optimal decision submodule is increased to 0.5, the weight coefficient of the quality-optimal decision submodule is adjusted to 0.3, and the weight coefficient of the energy-optimal decision submodule is adjusted to 0.2. The adjustment process of the weight coefficients adopts a first-order inertial element for smooth transition to avoid oscillation of the final set value caused by sudden weight changes.

[0054] Finally, a linear weighted average is calculated on the three weighted decision schemes to determine the final air knife pressure setpoint, strip speed setpoint, and zinc liquid temperature setpoint. The linear weighted average calculation is performed independently for each control parameter. For example, the final air knife pressure setpoint equals the air knife pressure suggested by the energy efficiency optimization decision submodule multiplied by the energy efficiency weight, plus the air knife pressure suggested by the quality optimization decision submodule multiplied by the quality weight, and the air knife pressure suggested by the stability optimization decision submodule multiplied by the stability weight. The calculated final setpoints must be processed by the instruction smoothing transition unit before being sent to the underlying actuator.

[0055] The instruction smoothing transition unit is a key interface connecting high-level control decisions and low-level actuators. Its core function is to process step changes in setpoints with a ramp effect to ensure smooth operation of the actuators. The instruction smoothing transition unit receives the final air knife pressure setpoint, strip speed setpoint, and zinc liquid temperature setpoint from the multi-target collaborative execution module.

[0056] For each setpoint, the instruction smoothing transition unit maintains a current actual output value. When a new setpoint instruction arrives, the instruction smoothing transition unit compares the difference between the new setpoint and the current actual output value. If the difference is less than a preset dead zone threshold, the new setpoint is output directly. If the difference exceeds the dead zone threshold, the ramp function is activated. The ramp rate of the ramp function is not fixed but is adaptively calculated based on the strip speed and specification changes. For the strip speed setpoint, its ramp rate is proportional to the current strip speed, with a base rate of 0.5 meters per second, but a maximum of 2 meters per second. For the air knife pressure setpoint, its ramp rate is calculated based on the strip width change. For every 100 millimeters increase in width, the ramp rate increases by 1,000 Pascals per second. For the zinc liquid temperature setpoint, due to its high thermal inertia, the ramp rate is fixed at 0.5 degrees Celsius per minute.

[0057] This adaptive ramp treatment effectively avoids the impact of sudden changes in set values ​​on the stability of strip operation and the coating formation process, significantly reduces the number of products with excessive thickness during specification switching transitions, and protects key actuators such as air knife regulating valves, main drive motors, and zinc pot heaters, extending their service life.

[0058] The system also includes a process knowledge base and an adaptive learning engine, which together constitute the system's self-optimization and evolution capabilities. The process knowledge base is a structured database that stores three main types of data.

[0059] The first category is the historical optimal process parameter combination. For each steel grade and specification combination that has been produced, such as low-carbon steel with a thickness of 1.2 mm and a width of 1250 mm, it stores multiple sets of process parameter settings that historically achieved the best coating quality, lowest zinc consumption, or most stable production. These settings include air knife pressure, strip speed, zinc bath temperature, and air knife distance. The second category is the process window limitations for different steel grades and specifications. For example, for a certain high-strength steel, the upper limit of the zinc bath temperature is 465 degrees Celsius, and the lower limit of the air knife pressure is 20 kPa. The third category is the characteristic data and coping strategies for various defects. For example, when the coating has an excessively thick edge, the characteristic data is that the edge temperature detected by the area array infrared temperature measurement unit is more than 10 degrees Celsius higher than that of the center. The coping strategy is to appropriately increase the edge air knife pressure setting by 5 to 10 Pa. The process knowledge base provides data query and services to other modules in the system through a standard structured query language interface.

[0060] The adaptive learning engine is the intelligent core of the system. Its function is to iteratively optimize and update the parameters of the dynamically coupled modeling module, the objective function weights of the distributed parallel decision-making module, and the dynamic weight coefficients of the multi-objective collaborative execution module based on the actual results of each production task. The iterative optimization and update strategy adopts a reinforcement learning algorithm based on gradient descent. After the production of each coil of strip steel is completed, the system will obtain the average actual coating thickness, uniformity standard deviation, and actual zinc consumption data of the coil of strip steel from the downstream quality inspection station.

[0061] These actual data, along with the system's own predicted values ​​and control objectives, constitute a reward signal. The adaptive learning engine aims to maximize cumulative production benefits in the long term. Production benefits are a comprehensive indicator, and their calculation takes into account coating quality score, zinc consumption cost, and production stability score. The engine calculates the gradient by calculating the difference between the actual benefits and expected benefits under the current strategy and using the backpropagation algorithm. Then, it fine-tunes the model parameter matrix in the dynamic coupling modeling module, the weight coefficients within each objective function in the distributed parallel decision module, and the threshold parameters used for stage judgment in the multi-objective collaborative execution module along the gradient direction. The step size of each update is strictly limited, usually not exceeding one percent of the original parameter value, to ensure the stability of the learning process and avoid drastic fluctuations in system performance due to a single bad data.

[0062] Through continuous small-step iterative learning, the system can gradually adapt to the slow changes in the characteristics of production line equipment, such as the wear of air knife nozzles, the drift of temperature sensing elements, and the process requirements of new product specifications, thereby achieving a leap from static program control to dynamic self-adaptation.

[0063] The entire hot-dip galvanizing coating intelligent control system relies on a robust industrial computing and communication platform. The main controller can be a high-performance programmable automation controller equipped with a multi-core processor, each core with a frequency of no less than 1.8 GHz and a memory capacity of no less than 4 gigabytes.

[0064] The computational tasks of the distributed parallel decision-making module can be deployed on an additional industrial server equipped with a graphics processing unit to accelerate the solution of model predictive control optimization problems. All modules of the system communicate with each other through a deterministic industrial Ethernet network. The network protocol adopts the IEEE 802.1AS time-sensitive networking standard to ensure that control commands and feedback data are transmitted within a strictly defined time window, with the worst-case communication latency being less than one millisecond.

[0065] The system software architecture adopts a modular design, follows the IEC 61131-3 standard to write control logic, and is supplemented by a high-performance algorithm library written in C++. The human-machine interface provides complete system status monitoring, parameter setting, trend recording and alarm management functions. Operators can clearly observe the real-time data of the multimodal sensing acquisition module, the predictive output of the dynamic coupling modeling module, the optimization results of the three decision sub-modules, and the final decision and weight allocation of the multi-objective collaborative execution module, thereby realizing transparent monitoring and intelligent intervention of the production process.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart control system for hot-dip galvanizing, characterized in that, include: The multimodal sensing and acquisition module is used to collect data in real time on strip speed, zinc liquid temperature, air knife pressure, air knife distance, strip tension, and temperature distribution of the plated surface after plating during the production line operation. The dynamic coupling modeling module is coupled to the output of the multimodal sensing and acquisition module. It is used to establish a nonlinear state-space model describing the dynamic coupling relationship between air knife pressure, strip speed, zinc liquid temperature and coating thickness based on the real-time data provided by the multimodal sensing and acquisition module. The distributed parallel decision-making module is coupled to the output of the dynamic coupling modeling module. It is used to receive the state prediction information output by the dynamic coupling modeling module and perform parallel calculations based on multiple preset performance indicators to generate multiple sets of optimized decision schemes for the air knife pressure setpoint, strip speed setpoint, and zinc liquid temperature setpoint. The multi-objective collaborative execution module is coupled to the output of the distributed parallel decision-making module. It is used to comprehensively evaluate and integrate multiple sets of optimized decision schemes output by the distributed parallel decision-making module, generate the final execution instructions, and send them to the corresponding execution mechanism.

2. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, The multimodal sensing acquisition module includes a velocity detection unit, a temperature detection unit, a pressure detection unit, a distance detection unit, a tension detection unit, and an area array infrared temperature measurement unit. The speed detection unit uses a laser velocimeter to measure the running speed of the strip steel in real time. The temperature detection unit uses an immersion thermocouple array to measure the temperature of the molten zinc in different areas inside the zinc pot. The pressure detection unit uses a high-precision pressure transmitter to monitor the pressure value of the main air supply pipe and the two side branch pipes of the air knife. The distance detection unit uses a laser displacement sensor to detect the distance between the air knife lip and the strip surface in real time; The tension detection unit acquires real-time tension data of the strip steel through a tension sensor attached to the tension roller. The area array infrared temperature measurement unit is installed at the inlet of the post-plating cooling section to acquire two-dimensional temperature field distribution data of the strip surface.

3. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, The nonlinear state-space model constructed by the dynamic coupling modeling module includes four state variables, three control inputs, and two main outputs. The four state variables are coating thickness deviation, strip speed change rate, zinc bath temperature gradient, and air knife pressure fluctuation. The three control inputs are the air knife pressure setting value, the strip speed setting value, and the zinc liquid temperature setting value; The two main outputs are the predicted coating thickness and the predicted coating uniformity index.

4. The intelligent control system for hot-dip galvanizing according to claim 3, characterized in that, The model building process of the dynamic coupling modeling module includes: Based on historical production data and real-time sensor data, the recursive least squares method is used to identify model parameters online. By introducing an adaptive mechanism with a forgetting factor, the model is able to track the slow time-varying characteristics of process conditions. In addition, Kalman filtering is used to make optimal estimates of the model state in order to suppress the interference of measurement noise.

5. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, The distributed parallel decision-making module consists of three parallel decision sub-modules: the energy efficiency optimal decision sub-module, the quality optimal decision sub-module, and the stability optimal decision sub-module. The energy efficiency optimal decision submodule performs optimization calculations with the objective function of minimizing zinc consumption per unit area. The optimal quality decision submodule performs optimization calculations with the objective function of minimizing the weighted sum of coating thickness deviation and uniformity index; The stability-optimal decision submodule performs optimization calculations with the objective function of minimizing the sum of squared rates of change of key process parameters.

6. The intelligent control system for hot-dip galvanizing according to claim 5, characterized in that, Each decision submodule adopts a model predictive control framework, which generates an optimized sequence of air knife pressure setpoints, strip speed setpoints, and zinc liquid temperature setpoints by solving a rolling optimization problem in a finite time domain.

7. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, The decision fusion mechanism of the multi-objective collaborative execution module includes: A dynamic weighting coefficient is assigned to the decision schemes output by the energy efficiency optimal decision submodule, the quality optimal decision submodule, and the stability optimal decision submodule; the dynamic weighting coefficient is adaptively adjusted according to the current production stage and target priority. During the normal steady-state production phase, the weight coefficient of the quality-optimal decision-making submodule is set to 0.5, and the weight coefficients of the energy-optimal decision-making submodule and the stability-optimal decision-making submodule are each set to 0.

25. During specification switching or acceleration / deceleration transition phases, the weight coefficient of the stability-optimal decision submodule is increased to 0.5, the weight coefficient of the quality-optimal decision submodule is adjusted to 0.3, and the weight coefficient of the energy-optimal decision submodule is adjusted to 0.

2. Finally, a linear weighted average was performed on the weighted decision schemes to calculate the final air knife pressure setting, strip speed setting, and zinc liquid temperature setting.

8. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, It also includes a process knowledge base and an adaptive learning engine; The process knowledge base is used to store historical optimal process parameter combinations, process window limitations for different steel grades and specifications, and characteristic data and coping strategies for various defects. The adaptive learning engine is used to iteratively optimize and update the parameters of the dynamically coupled modeling module, the objective function weights of the distributed parallel decision-making module, and the dynamic weight coefficients of the multi-objective collaborative execution module based on the actual results of each production task.

9. The intelligent control system for hot-dip galvanizing according to claim 8, characterized in that, The adaptive learning engine's update strategy employs a gradient descent-based reinforcement learning algorithm, with the long-term optimization goal of maximizing cumulative production benefits.

10. The intelligent control system for hot-dip galvanizing according to claim 1, characterized in that, A smooth instruction transition unit is provided between the multi-target collaborative execution module and the underlying execution mechanism; The instruction smooth transition unit is used to perform ramp processing on step changes of set values. Its ramp rate is adaptively calculated based on the strip speed and specification changes to ensure smooth operation of the actuator and avoid impact on the strip running and coating formation process.

Images