Intelligent scheduling system for steel plate hot dipping production process

Abstract

The invention relates to the technical field of steel plate hot-dip production, and discloses an intelligent scheduling system for a steel plate hot-dip production process. The system comprises a working condition sensing module which is used for continuously collecting steel belt inlet temperature, zinc liquid component concentration fluctuation data, air knife pressure time sequence parameters and roller system operation state signals of a galvanizing pot area; the feature extraction module receives the working condition data flow, analyzes the steel strip surface temperature field distribution features and the zinc liquid viscosity change rule, and generates a process feature spectrum; the parameter optimization module dynamically adjusts membership function parameters and rule base weight coefficients of the fuzzy reinforcement learning controller according to the process characteristic spectrum, and outputs an optimized plating layer control parameter set; the self-adaptive plating layer control module generates a plating layer thickness control instruction based on the real-time zinc liquid temperature parameter and the optimized plating layer control rule set; and the dynamic roller system scheduling module realizes global cooperative control of the roller system speed based on the instruction. The system can realize intelligent scheduling of the steel plate hot-dip production process.

Description

Technical Field

[0001] The present invention relates to the technical field of hot-dip production of steel plates, and in particular to an intelligent scheduling system for the hot-dip production process of steel plates. Background Art

[0002] Balancing coating quality and production efficiency remains a key challenge in the hot-dip galvanizing process. Traditionally, operating parameters in the galvanizing pot area have relied heavily on manual adjustments, making them difficult to adapt to complex and changing production environments. Fluctuations in the inlet temperature of the steel strip directly affect the wetting effect of the zinc bath. Excessively high or low temperatures can easily lead to coating defects such as missing coatings and blistering. Unstable zinc bath concentrations can alter the physical properties of the bath, affecting coating uniformity. The control accuracy of air knife pressure plays a key role in coating thickness. Existing technologies often use fixed parameter control methods, which cannot dynamically adjust to real-time operating conditions, resulting in large variations in coating thickness. Inadequate coordination of the roller system's operating states can cause fluctuations in strip tension, impacting production continuity and potentially causing quality issues such as surface scratches. The complex coupling relationships between process parameters make it difficult to improve overall production efficiency by optimizing a single parameter. Traditional scheduling systems lack the ability to coordinate and optimize multiple parameters globally, resulting in high energy consumption during production and difficulty maintaining an ideal product qualification rate, which in turn fails to meet the high efficiency and high quality requirements of modern industrial production. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent scheduling system for the hot-dip coating production process of steel plates to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above-mentioned object, the present invention provides an intelligent scheduling system for a steel plate hot-dip coating production process, the system comprising: The working condition sensing module continuously collects the working condition data stream of the galvanizing pot area, including the steel strip inlet temperature, zinc liquid component concentration fluctuation data, air knife pressure timing parameters, and roller system operation status signals; The feature extraction module receives the working condition data stream, analyzes the surface temperature distribution characteristics of the steel strip and the viscosity change law of the zinc liquid, and generates a process feature map; The parameter optimization module dynamically adjusts the membership function parameters and rule base weight coefficients of the preset fuzzy reinforcement learning controller according to the process characteristic map, and outputs the optimized coating control parameter set; The adaptive coating control module makes a dual-objective collaborative decision on the air knife opening and the amount of zinc ingot added based on the real-time zinc liquid temperature parameters and the optimized coating control rule set, and generates coating thickness control instructions; The dynamic roll system scheduling module solves the speed constraint boundary of the tension roll group based on the coating thickness control instruction, and performs mixed integer programming on the roll gap parameters in combination with the primal dual interior point method to achieve global coordinated control of the roll system speed.

[0005] Preferably, the continuous collection of the working condition data stream of the galvanizing pot area includes: Based on historical coating defect data and zinc liquid temperature fluctuation curve, a correlation mapping table between steel strip running speed and sampling frequency is established; The sampling intervals of the temperature sensor and the component analyzer are dynamically adjusted based on the association mapping table, a sliding window filtering process is performed on the collected original zinc liquid temperature data, and an outlier correction compensation is performed on the component concentration data.

[0006] Preferably, the analyzing the distribution characteristics of the surface temperature field of the steel strip and the variation law of the zinc liquid viscosity to generate a process characteristic map includes: The pre-processed steel strip surface temperature time series data is divided into continuous subsequence blocks; Initialize the weight matrix and learning rate attenuation coefficient, and calculate the local temperature prediction deviation of each subsequence block through the loss function; The weight matrix parameters are updated according to the gradient direction of the temperature prediction deviation, and the iteration is terminated when the prediction deviation change rate of five consecutive subsequence blocks is less than the convergence threshold; The updated weight matrix is ​​used to predict the viscosity change trend of the zinc liquid, and the process characteristic map including the temperature field gradient vector and viscosity change rate is generated in combination with the air knife pressure fluctuation characteristics.

[0007] Preferably, the method of dynamically adjusting the membership function parameters and rule base weight coefficients of the preset fuzzy reinforcement learning controller according to the process characteristic map and outputting the optimized coating control parameter set includes: Taking the coating thickness uniformity index as the optimization objective function, the particle swarm position vector and velocity vector are initialized; Decode each particle position vector into the rule triggering threshold and membership function width parameter of the fuzzy reinforcement learning controller; Evaluate the thickness control performance of each particle corresponding to the parameter set in the coating simulation environment and calculate the particle fitness score; The particle velocity vector is iteratively updated according to the global optimal particle position and the individual historical optimal position. When the optimal fitness score does not improve for ten consecutive generations, the parameter set corresponding to the current optimal particle is output.

[0008] Preferably, the dual-objective collaborative decision-making of the air knife opening and the amount of zinc ingot added based on the real-time zinc liquid temperature parameter and the optimized coating control rule set includes: Construct a decision variable space including air knife opening, zinc ingot adding rate, and steel strip running speed; Based on the current zinc liquid temperature parameter, a neighborhood solution set is randomly generated in the decision variable space; Calculate the coating thickness prediction deviation and zinc consumption cost weighted evaluation value of each neighborhood solution; The neighborhood solutions are accepted or discarded according to the annealing temperature curve, and the air knife opening instruction and zinc ingot addition instruction corresponding to the Pareto optimal solution are output when the annealing temperature drops to the termination threshold.

[0009] Preferably, the calculating of the speed constraint boundary of the tension roller group based on the coating thickness control instruction includes: Establish the tension roller group speed constraint equation and the coating thickness response delay constraint inequality group; Calculate the sub-gradient projection vector of the objective function under the current roller speed configuration; The initial solution of the roll gap parameters is updated along the subgradient projection direction, and the new solution is mapped to the speed-constrained feasible region using the projection operator. When the roller speed adjustment amplitude is less than the displacement threshold for three consecutive iterations, the roller system collaborative control parameters are locked.

[0010] Preferably, the system further comprises: The coating quality feedback module uses a Kalman filter to fuse the coating thickness distribution data, and feeds the fused thickness distribution characteristics back to the feature extraction module, parameter optimization module and adaptive coating control module, triggering the dynamic reconfiguration of the coating control strategy.

[0011] Preferably, the operation of the coating quality feedback module includes: Synchronously collect coating thickness distribution data at the air knife outlet and surface quality images at the cooling tower outlet; The Kalman filter is used to fuse the thickness gauge measurement value and the thickness distribution characteristics of image recognition to generate a thickness distribution compensation matrix; Input the thickness distribution compensation matrix into the parameter correction interface of the feature extraction module and update the temperature field gradient vector calculation rule; At the same time, the air knife pressure compensation coefficient is sent to the adaptive coating control module to trigger the real-time recalculation of the air knife opening instruction.

[0012] Preferably, generating a thickness distribution compensation matrix includes: Extract the horizontal fluctuation standard deviation and vertical trend slope of thickness distribution data; Identify the location distribution characteristics of zinc nodule defects in surface quality images; The Kalman gain coefficient is used to fuse the lateral fluctuation standard deviation and the zinc nodule distribution density data to generate the weight distribution vector of the thickness distribution compensation matrix.

[0013] Preferably, the global coordinated control of the roller system speed includes: Analyze the tension roller acceleration constraint conditions in the coating thickness control instructions; The mixed integer programming problem of roll gap parameters is decomposed into speed allocation subproblem and delay compensation subproblem; The primal-dual interior point method is used to alternately solve the integer relaxation solutions of the dual gap and delay compensation subproblems of the velocity allocation subproblem. When the difference between the objective functions of the two sub-problems is less than the collaborative error threshold, the roller system speed collaborative control instruction set is output.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The working condition sensing module continuously collects multiple key working condition data streams in the galvanizing pot area, realizing comprehensive real-time monitoring of the production process. It can timely capture subtle changes in the steel strip inlet temperature, zinc liquid concentration, air knife pressure and roller system operating status, providing comprehensive and accurate basic information for subsequent parameter optimization and control decisions. The feature extraction module conducts in-depth analysis of the collected working condition data stream to generate a process feature map, clearly presenting the distribution characteristics of the temperature field on the steel strip surface and the change law of the zinc liquid viscosity. It helps to deeply understand the intrinsic connection between various parameters and the mechanism of their influence on the coating quality, and provides a scientific basis for parameter optimization.

[0015] Based on the process characteristic map, the parameter optimization module dynamically adjusts the membership function parameters and rule base weight coefficients of the fuzzy reinforcement learning controller, enabling the controller to better adapt to complex and changeable production conditions. The output optimized coating control parameter set is more in line with actual production needs, enhancing the system's adaptability to different production conditions. The adaptive coating control module combines the real-time zinc liquid temperature parameters with the optimized coating control rule set to make a dual-objective collaborative decision on the air knife opening and the amount of zinc ingot added. It can ensure the coating quality while taking into account production efficiency and resource consumption, thus achieving a balance between multiple objectives. The dynamic roller scheduling module solves the speed constraint boundaries of the tension roller group based on the coating thickness control instructions, and uses the primal dual interior point method to perform mixed integer programming on the roller gap parameters, realizing global coordinated control of the roller system speed, effectively reducing the tension fluctuation of the steel strip, ensuring the continuity and stability of production, and reducing the probability of quality problems such as scratches on the steel strip surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a timing diagram of the intelligent scheduling system for the hot-dip galvanizing production process of steel plates according to the present invention; Figure 2 Flowcharts generated for process profiles; Figure 3 Flowchart for parameter optimization of fuzzy reinforcement learning controller; Figure 4 Flowchart for solving the speed constraint of the tension roller group; Figure 5 Flowchart of the coating quality feedback operation. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 The present invention provides an intelligent scheduling system for the hot-dip coating production process of steel plates, which includes: a working condition perception module, a feature extraction module, a parameter optimization module, an adaptive coating control module and a dynamic roller system scheduling module.

[0019] The operating condition perception module uses a multi-source sensor network distributed throughout the galvanizing pot area to collect real-time data on the steel strip inlet temperature, zinc bath concentration fluctuations, air knife pressure timing parameters, and roller system operating status signals, generating a continuous operating condition data stream. After receiving this data stream, the feature extraction module uses time series analysis to analyze the distribution characteristics of the steel strip surface temperature field and generates a process characteristic map based on a zinc bath viscosity variation model. The parameter optimization module uses an online learning mechanism to adjust the structural parameters of the fuzzy reinforcement learning controller based on the dynamic changes in the process characteristic map, outputting an optimized set of coating control parameters. The adaptive coating control module integrates real-time zinc bath temperature monitoring data with optimized control rules to perform multi-objective decision-making on air knife opening and zinc ingot addition. The dynamic roller system scheduling module uses a mixed integer programming algorithm to coordinate the operating parameters of the tension roller group based on the coating control instructions, achieving precise control of coating thickness.

[0020] Example 1: See Figure 2 , focusing on the dynamic collection mechanism of working condition data stream and the generation process of process characteristic maps. In the galvanizing pot area, nine groups of infrared temperature sensors are distributed in a staggered layout on the transverse section of the steel strip running channel. The sensor spacing is adaptively adjusted according to the strip width, and the minimum spacing is set to 50mm. A nonlinear correlation mapping is established between the sampling frequency of the temperature measurement array and the running speed of the steel strip. The basic sampling frequency is set to 10Hz. When the steel strip speed exceeds 120m / min, the sampling frequency increases by 1.5Hz for every increase of 10m / min, and the maximum sampling frequency is limited to 25Hz. The zinc liquid composition analyzer uses an immersion probe to complete a full-element spectrum scan every 20 seconds, and performs outlier correction processing on the collected aluminum, lead, and tin concentration data: a historical 30-minute concentration distribution model is established, and data points that exceed the range of three times the standard deviation are replaced with the median value of the three adjacent sampling points in the sliding window.

[0021] After the temperature field analysis unit of the feature extraction module receives the continuous temperature data stream, it first performs data alignment. The surface temperature sequence of the steel strip is divided into continuous blocks at intervals of 500 milliseconds, and each block contains 75 original sampling points. In the initialization phase, a weight matrix with a dimension of 9×9 is constructed. The initial values ​​of the matrix elements are assigned different weights according to the positional relationship of the temperature measurement points. The weight coefficient of the central area is set to 0.15, and the coefficient of the edge area is set to 0.08. The initial value of the learning rate is 0.7 and a step-by-step attenuation strategy is adopted. After processing 20 data blocks, the attenuation coefficient is adjusted to 95% of the original value. When calculating the mean square error, a 1.2-fold weighting is applied to the temperature gradient change rate. When the predicted deviation change of five consecutive blocks is less than 0.15°C / s, the system freezes the current weight matrix parameters and starts the update cycle timer.

[0022] The viscosity characteristic calculation unit synchronously receives the processed zinc liquid composition data and establishes a concentration-viscosity conversion model. This model includes a negative correlation curve between aluminum content and viscosity and a positive correlation compensation function for lead. After the weight matrix is ​​updated, the viscosity prediction program is automatically activated: the component values ​​of the temperature gradient vector are input into the viscosity base value model, and the composition compensation factor acts on the second-order derivative term. Every 5 seconds, the system outputs a temperature field gradient vector containing nine dimensions and a viscosity change characteristic value. The viscosity value is accurate to 0.1mPa·s, and the temperature gradient value is retained to three decimal places. The encoding structure of the process characteristic map is in a two-dimensional matrix format. The first column records the timestamp information, and the subsequent ten columns store the nine components of the temperature gradient vector and the viscosity change rate value respectively. The matrix row index strictly corresponds to the time series.

[0023] The processing of the air knife pressure fluctuation characteristics is carried out independently. The pressure sensor collects the original signal at a frequency of 100Hz, and the signal is filtered by a 0.1-15Hz bandpass filter to extract the main frequency component of the pressure oscillation. The energy spectrum density distribution of the pressure fluctuation is calculated every 2 seconds, and the frequency values ​​corresponding to the first three significant peaks are selected as characteristic parameters. In the process of generating the process characteristic map, the pressure feature is used as an independent channel to align with the temperature-viscosity data in time and space: when the temperature field vector and viscosity value matching the time interval are received, the system automatically associates the pressure characteristic value within the time period. The final map data packet contains a process characteristic vector with fourteen dimensions, including timestamp, nine temperature gradient components, viscosity change, and pressure main frequency parameter. The data update cycle is synchronized with the temperature field feature for 5 seconds.

[0024] The abnormal data processing mechanism has an independent process in the system. When the component analyzer detects that the instantaneous jump of the aluminum concentration in the zinc liquid exceeds 1.5%, the re-verification sampling program is automatically started, and three repeated measurements are performed within 200 milliseconds. If the two re-verification results confirm the initial abnormal value, the viscosity calculation module immediately activates the emergency calculation model, replaces the current value with the sliding average of the aluminum concentration in the previous 5 minutes, and adds a data anomaly mark to the process characteristic map. At the same time, when the weight matrix fails to converge in the temperature field feature calculation stage, the system automatically switches to the historical optimal weight configuration library, gives priority to calling the successful parameter combination for producing products of similar specifications, and sends a model reset alarm signal to the main console. This nested exception handling architecture maintains the continuity of the feature extraction process and eliminates the impact of single-point failures on map generation.

[0025] Example 2: See Figure 3 , covering the optimization process of coating control parameters and the specific operation of the dual-objective decision-making mechanism. The parameter optimization module has a built-in particle swarm algorithm framework. 50 individual particles are set in the population initialization stage. The position vector of each particle has 19 dimensions, of which the first 7 dimensions correspond to the control parameters of the Gaussian membership function, and the last 12 dimensions store the weight coefficients of the rule base. Real number encoding format is used for vector encoding. The decoding interval of the membership function width parameter is limited to 0.2 to 1.5, and the rule trigger threshold is mapped to the standardized input variable range. The particle velocity vector is initialized using a uniform random distribution, and the maximum velocity constraint is set to 15% of the search space range.

[0026] The particle fitness assessment phase activates a plating process simulation environment synchronized with real-time production line data. Each assessment run lasts five production cycles (approximately 300 seconds). During the simulation, operating parameters such as the current measured zinc bath temperature and steel strip speed are loaded. The control system imports the measured parameter set into the fuzzy reinforcement learning controller architecture and records the instantaneous fluctuations in the output coating thickness. The data processing unit calculates the root mean square error (RMS) between the actual coating thickness and the target thickness, adds the standard deviation of the rule base trigger frequency as a complexity penalty, and finally weights this to form a fitness evaluation value. The dynamic range of the evaluation value is mapped to a range of 0-100, with lower scores indicating better control performance.

[0027] The particle update mechanism adopts a dynamic balance strategy. Each iteration calculates the distance function between the individual historical optimal solution and the group optimal solution, and dynamically adjusts the ratio of the social learning factor to the individual learning factor based on the distance value. The inertia weight in the velocity update formula adopts a linear decay mechanism, gradually decreasing from an initial value of 0.9 to 0.4. Boundary checks are performed after the position is updated, and mirror reflection processing is performed on dimension values ​​that exceed the parameter range. The optimization process monitoring unit tracks the historical change curve of the optimal fitness. When the fluctuation range of the optimal evaluation value for ten consecutive generations is less than 0.01 minutes, the controller parameter set corresponding to the current optimal particle is output to the decision module. The parameter conversion process retains three decimal places of precision, and binary checksums are used to ensure data transmission integrity.

[0028] After receiving the optimized parameter set, the adaptive coating control module activates the simulated annealing decision process. The decision variable space is constructed as a three-dimensional cube: the air knife opening range is 0-100% with a resolution of 1%; the zinc ingot addition rate is measured in kg / min with a step size of 0.5 kg; and the steel strip speed adjustment range is controlled within ±15% of the baseline value. The initial solution is generated using the current production status as the reference point, and the neighborhood radius is initially set to 15% of the solution space diameter. The annealing temperature starts at 1200°C, with a fixed temperature decay coefficient of 0.92, and the temperature decreases after each neighborhood search.

[0029] The neighborhood solution generation mechanism uses a Gaussian perturbation strategy to randomly generate 20 candidate solutions. Each solution is evaluated in two parallel processes: the coating thickness prediction model loads the current zinc bath temperature parameter and calculates the theoretical coating thickness based on the air knife aperture and steel strip speed; the cost accounting unit calculates the total zinc ingot consumption and converts it into economic costs. In the dual-objective evaluation value calculation formula, the thickness deviation term is weighted to 0.7, and the zinc consumption cost term is weighted to 0.3. The acceptance probability calculation of the new solution takes into account the quality difference between the current annealing temperature and the solution, and uses an exponential probability distribution function to control the random acceptance probability.

[0030] During the search process, an elite retention strategy is set. The optimal solution is recorded in each temperature stage and compared horizontally with the optimal solution of the previous temperature stage. The annealing termination condition adopts a composite judgment mechanism: when the temperature drops to the 50°C threshold or the optimal solution of three consecutive temperature stages is not updated, the search process is terminated. In the output stage, a Pareto front analysis is performed, and the solution with the smallest thickness deviation and zinc consumption cost below the set threshold is selected as the final decision result. The output data packet contains the percentage value of the air knife opening instruction (accuracy 0.1%), the kg / min value of the zinc ingot addition rate (accuracy 0.1kg), and a status identifier of whether the steel strip speed needs to be adjusted. The decision cycle is controlled within 40 seconds, and the operation instruction is sent to the actuator synchronously before the system times out.

[0031] Example 3: See Figure 4, involving the solution of the speed constraint boundary of the tension roller group and the generation process of the roller system coordinated control strategy. The constraint modeling unit of the dynamic roller system scheduling module establishes a set of dynamic equations for 12 tension rollers, each of which contains the tension balance condition between the rollers and the coating response delay limit. The construction of the roller system speed constraint equation is based on the elastic deformation theory of strip steel, considering the relationship between the speed difference between the front and rear tension rollers and the elongation of the strip steel. The delay constraint inequality group quantifies the response time of the coating thickness to the change of roller speed, and converts the thickness control accuracy required by the process into the time delay upper limit. The mathematical expression of the constraint conditions is as follows:

[0032] in: Indicates the Constraints, For the Linear speed of tension roller No. It is The response delay time of the strip steel segment, is the speed difference influence coefficient, is the roller spacing, is the delay sensitivity factor, The maximum constraint value allowed by the system.

[0033] The sub-gradient projection calculation uses the Barzilai-Borwein step size selection method, and the initial step size is Set to 0.05. Each iteration calculates the sub-gradient vector of the objective function , the projection direction is determined as the direction of the bisector of the angle between the negative direction of the subgradient and the tangent plane of the feasible region. Projection operator Tolerance threshold for handling equality constraints when mapping new solutions to the velocity-constrained feasible region Set to 0.001mm / s. Monitor the change of the solution vector during the iteration process , when three times in a row Value less than the displacement threshold mm, lock the current roller speed configuration parameters.

[0034] The mixed integer programming solver discretizes the roll gap adjustment into integer multiples of 0.1 mm, using a branch-and-bound approach to handle integer variables. During the relaxation phase, the speed variables of nine of the 12 tension rollers are made continuous, while retaining the discrete characteristics of the three key rollers (the entrance section, the middle control section, and the main rollers of the exit section). The barrier parameter of the primal dual interior point method is initially set to 1.0 and decays by a factor of 0.5 after every five iterations. The calculation includes two parts: primal feasibility and dual feasibility. The convergence criterion is And integer solution deviation .

[0035] The delay compensation subproblem is solved using a prediction-correction strategy. The prediction step calculates the theoretical delay time under the ideal roller speed configuration. , the correction step adjusts the delay compensation amount according to the actual coating thickness feedback The discretization of the compensation amount adopts the principle of rounding up to ensure that there is a margin for adjustment. In the collaborative solution process of the speed distribution subproblem and the delay compensation subproblem, the original variables and the dual variables are updated alternately, and the difference between the objective functions of the two subproblems is checked after each iteration. .when Less than the collaborative error threshold three times in a row When , the system is judged to have reached a coordinated equilibrium state.

[0036] The generation of roller control instructions adopts a layered coding structure. The base layer contains the reference speed value of each tension roller. , with an accuracy of 0.01m / s; the adjustment layer stores the acceleration limit curve , sampling at 0.1s intervals; the synchronization layer records the start timestamp of each roller The time synchronization accuracy is controlled within ±5ms. The control cycle is fixed at 100ms, and the ready state of the actuator is checked at the beginning of each cycle. The abnormality handling mechanism monitors the deviation between the actual speed of each roller and the command value in real time. ,when Exceeding tolerance value When it lasts for 3 cycles, the speed recalibration process is triggered.

[0037] The system maintains a dynamically updated roll status database, recording historically optimized parameter combinations. After each successful coating control task, the current roll speed configuration parameters are stored in the database, indexed by the product specification code. When processing products of similar specifications, historical parameters are prioritized as the initial solution, shortening solution time. The database utilizes a binary tree structure to organize data, ensuring query response times of less than 50ms. A CRC-16 checksum is used during parameter transmission to ensure data integrity.

[0038] Multiple protection mechanisms are set for online adjustment of roll gap parameters. Both mechanical limit constraints and strip plastic deformation limits must be met. Mechanical limit constraints check the relationship between the current roll gap and the maximum allowable adjustment of the equipment, while deformation limits verify that the strip elongation does not exceed 70% of the material's yield point. The protection mechanism operates on a 10ms cycle and issues an emergency stop signal within 2ms when a potential risk is detected. During normal adjustment, the roll gap change rate is limited to less than 0.5mm / s to avoid shock loads.

[0039] Example 4: See Figure 5, covering the acquisition and fusion of coating quality data and the dynamic adjustment process of closed-loop control strategy. The coating quality feedback module is configured with an X-ray thickness gauge model XR-5000, which is uniformly arranged with 21 measurement probes in the width direction of the strip steel. The probe spacing is adaptively adjusted according to the width of the strip steel: when the width of the strip steel is greater than 1500mm, an equal interval of 80mm layout is adopted, and for narrow strip steel, it is set according to the width of 5%. The sampling frequency of the thickness gauge is automatically matched with the production line speed, and the reference sampling rate is 5 times per second. When the production line speed is accelerated to more than 120m / min, the high-speed mode is triggered to increase to 10 times per second. The high-speed industrial camera system model CVS-8000H runs synchronously, which is installed 3.5 meters above the outlet side of the cooling tower, and the lens pitch angle is adjusted to 15 degrees. The camera is configured with a ring-shaped LED fill light with a color temperature of 5600K. It collects 2048x1536 pixel RGB images at a frame rate of 200fps, and the exposure time is fixed at 2ms.

[0040] Table 1: Coating thickness data fusion processing (unit: μm).

[0041]

[0042] The surface defect recognition algorithm first performs image preprocessing: convert the original RGB image to HSV color space, and perform histogram equalization enhancement in the saturation channel to enhance the contrast of the zinc tumor feature. The adaptive threshold segmentation method is used to extract the suspected defect area, and the morphological opening operation is used to eliminate noise. The zinc tumor judgment standard considers both the area factor (minimum 20 pixels) and the shape factor (aspect ratio greater than 0.6). When generating the final defect distribution map, the width direction of the strip steel is equally divided into 100 sections, and the defect density value (unit: piece / cm²) of each section is calculated. The density value is quantified as a normalized index between 0 and 1.

[0043] The data fusion core uses a discrete Kalman filter architecture. The state vector is defined as the strip steel transverse thickness distribution matrix , and the observation matrix comes from the direct measurement data of the thickness gauge and the image feature conversion data . In the prediction stage, the thickness distribution change is extrapolated according to the strip steel speed, and the material flow coefficient in the state transition matrix is fixed at 0.98. In the observation update stage, a dual-channel fusion mechanism is used: the thickness gauge measurement noise covariance matrix is set to a diagonal matrix with diagonal element value 0.25, and the image recognition observation noise is dynamically adjusted according to the image clarity. When the clarity is lower than 80, the noise coefficient is increased by 30%.

[0044] The thickness distribution compensation matrix generation process includes three parallel calculation threads: the transverse fluctuation calculation thread is based on the thickness gauge data and adopts the sliding window standard deviation algorithm. The window width is 15% of the current strip length, and the standard deviation calculation result is mapped to the 0-10 level fluctuation index. The longitudinal trend analysis thread performs linear regression calculation, and the slope value is retained to three decimal places. The defect feature quantification thread integrates the zinc nodule density distribution of the current and previous two frames of images to generate a spatial position weighted index. Finally, the three types of feature quantities are fused through the Kalman gain coefficient to output the weight distribution vector of the compensation matrix. , each element corresponds to the correction coefficient at a specific position in the bandwidth direction.

[0045] The control strategy reconfiguration mechanism sets dual trigger conditions: when it is detected that the standard deviation of the transverse thickness exceeds the target value by 20%, or when the zinc nodule defect density exceeds 0.8 in the critical area (15% area on both sides of the bandwidth), the system immediately activates the parameter correction process. The interface of the feature extraction module supports the parameter hot update function: after the temperature field gradient vector calculator receives the compensation matrix, it multiplies it with the original weight matrix element by element, and the new matrix overwrites the original configuration within 2 seconds. The adaptive coating control module receives the air knife pressure compensation coefficient, which is calculated and generated by the central controller according to the direction of the thickness deviation: when the thickness deviation on the left side of the strip is greater than that on the right side, the compensation coefficient is distributed with a gradient of higher left and lower right, and the deviation value is scaled by 0.5% of the difference.

[0046] The system maintains a three-tier fault handling mechanism to ensure data continuity. If a single thickness gauge probe fails, Lagrange interpolation of adjacent probe data is used to supplement the data. If image recognition fails, the system automatically switches to historical thickness fluctuation trend prediction using the previous five seconds of data. In extreme cases (such as the simultaneous loss of both thickness gauge and image data), the system generates a virtual compensation matrix based on the zinc bath temperature variation model and triggers an alarm. All data channels are timestamp-checked, with a maximum permissible time difference of 50ms. Expired data is automatically discarded and replaced by the prediction module. Compensation command transmission utilizes a dual-channel redundant design: the primary channel transmits the compensation matrix via the CAN bus, while the backup channel independently sends control coefficient change commands via Industrial Ethernet.

[0047] Example 5 covers the implementation of a global coordinated control strategy for the roll system, specifically parsing the tension roller motion constraints in the coating thickness control instructions and performing optimization calculations. After receiving the coating thickness control instructions, the system's parsing module identifies the implicit tension roller acceleration constraint parameters within the instructions. These constraints are expressed as differential acceleration limit curves for the strip entry section, mid-control section, and exit section. The central controller automatically decomposes the mixed integer programming model for the roll gap parameters into two interrelated subproblems: a speed allocation subproblem responsible for coordinating the relative speeds between the tension rollers, and a delay compensation subproblem addressing the response time differences to coating thickness changes.

[0048] During the implementation of the primal-dual interior point method, an independent solution environment is set up for each subproblem during the initialization phase. The velocity allocation subproblem solver uses the current roll speed configuration as a reference point when constructing the initial feasible solution, with the objective function defined as minimizing strip tension fluctuations. The dual gap calculation phase employs a dual iteration mechanism: an inner loop handles the Lagrange multiplier updates for the equality constraints, while an outer loop adjusts the obstacle parameters to achieve path tracking within the feasible domain. The calculation process records the primal and dual objective values ​​for each iteration, and the subproblem is considered converged when the difference between the two converges within a set threshold.

[0049] The delay compensation subproblem solver incorporates an integer relaxation mechanism, converting discrete time delay variables into continuous constraints. The integer characteristics of key rollers (entry looper roller, process section main roller, and exit tension roller) are retained, while the parameters of the remaining auxiliary rollers are made continuous. Upper and lower bounds are set on the relaxation variables to prevent excessive deviation from physical reality. The compensation calculation utilizes a predictive correction model: the coating thickness trend is predicted based on the current roller speed, and a correction vector is generated by comparing it with the actual measured value. The delay compensation subproblem solver outputs a hybrid solution consisting of continuous adjustments and discrete step changes.

[0050] The collaborative optimization controller establishes a feedback connection mechanism between the two subproblems. After each iteration, the intermediate solution output by the speed allocation subproblem is automatically input into the delay compensation subproblem as a constraint boundary; the time parameters generated by the delay compensation subproblem are fed back to modify the equality constraints of the speed allocation subproblem. The dual-loop structure exchanges intermediate variables through a shared memory area, with the exchange period fixed to synchronize every three minutes. A tolerance monitoring mechanism is set during the alternating solution process. When the difference between the objective functions of the two subproblems remains stable within the set fluctuation range for six consecutive iterations, the system automatically generates a convergence status indicator.

[0051] After receiving the optimization results, the roller control command generation unit constructs a hierarchical command structure. The basic control layer contains baseline speed command values ​​for the twelve tension rollers, with command accuracy reaching the thousandth of a meter per second level. The dynamic adjustment layer generates a speed adjustment curve distributed by timestamp, defining a sequence of acceleration changes in 0.1 second intervals. The safety protection layer sets speed deviation tolerance parameters, with the upper limit of the dynamic error between the actual speed of each roller and the command value set to a specific ratio of the theoretical speed. The control cycle is fixed at 100 milliseconds, and the command issuance unit refreshes the control queue at the beginning of each cycle.

[0052] The system is equipped with an exception handling module to prevent control failures. It monitors the load fluctuation data of the tension roller drive motor in real time. When a sudden sudden change in a roller's load reaches the warning threshold, it automatically activates the independent PID control mode for that roller and reduces the speed gradient of the adjacent roller system. In the event of an abnormal coating thickness response, the safety mechanism initiates a parameter rollback operation, recalculates the historical optimized parameters from the previous five minutes, and simultaneously sends an abnormal operating condition code to the main console. After each parameter adjustment, the system automatically executes a simulation verification procedure to virtually verify the execution of the control instructions. The verification cycle covers the entire strip processing stage.

[0053] The roll system collaborative control database utilizes a hierarchical storage architecture. The base layer maintains detailed optimization records for the past 72 hours, including boundary parameters, intermediate variables, and output instructions for each solution. The index layer maps strip specifications to control parameters, using thirteen dimensions such as material thickness and zinc layer target value as classification labels. The strategy layer stores the feature vectors of optimal solutions under typical operating conditions. When the matching degree between the currently processed strip specifications and database records reaches a certain threshold, the system prioritizes the historical optimization solution as the initial value, reducing optimization startup time to 40% of conventional processes.

[0054] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent scheduling system for hot-dip galvanizing production process of steel plates, characterized in that: include: The working condition sensing module continuously collects the working condition data stream of the galvanizing pot area, including the steel strip inlet temperature, zinc liquid component concentration fluctuation data, air knife pressure timing parameters, and roller system operation status signals; The feature extraction module receives the working condition data stream, analyzes the surface temperature distribution characteristics of the steel strip and the viscosity change law of the zinc liquid, and generates a process feature map; The parameter optimization module dynamically adjusts the membership function parameters and rule base weight coefficients of the preset fuzzy reinforcement learning controller according to the process characteristic map, and outputs the optimized coating control parameter set; The adaptive coating control module makes a dual-objective collaborative decision on the air knife opening and the amount of zinc ingot added based on the real-time zinc liquid temperature parameters and the optimized coating control rule set, and generates coating thickness control instructions; The dynamic roll system scheduling module solves the speed constraint boundary of the tension roll group based on the coating thickness control instruction, and performs mixed integer programming on the roll gap parameters in combination with the primal dual interior point method to achieve global coordinated control of the roll system speed.

2. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The continuous collection of working condition data streams in the galvanizing pot area includes: Based on historical coating defect data and zinc liquid temperature fluctuation curve, a correlation mapping table between steel strip running speed and sampling frequency is established; The sampling intervals of the temperature sensor and the component analyzer are dynamically adjusted based on the association mapping table, a sliding window filtering process is performed on the collected original zinc liquid temperature data, and an outlier correction compensation is performed on the component concentration data.

3. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The analysis of the distribution characteristics of the steel strip surface temperature field and the viscosity variation of the zinc liquid to generate a process characteristic map includes: The pre-processed steel strip surface temperature time series data is divided into continuous subsequence blocks; Initialize the weight matrix and learning rate attenuation coefficient, and calculate the local temperature prediction deviation of each subsequence block through the loss function; The weight matrix parameters are updated according to the gradient direction of the temperature prediction deviation, and the iteration is terminated when the prediction deviation change rate of five consecutive subsequence blocks is less than the convergence threshold; The updated weight matrix is ​​used to predict the viscosity change trend of the zinc liquid, and the process characteristic map including the temperature field gradient vector and viscosity change rate is generated in combination with the air knife pressure fluctuation characteristics.

4. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The method dynamically adjusts the membership function parameters and rule base weight coefficients of the preset fuzzy reinforcement learning controller according to the process characteristic map, and outputs the optimized coating control parameter set, including: Taking the coating thickness uniformity index as the optimization objective function, the particle swarm position vector and velocity vector are initialized; Decode each particle position vector into the rule triggering threshold and membership function width parameter of the fuzzy reinforcement learning controller; Evaluate the thickness control performance of each particle corresponding to the parameter set in the coating simulation environment and calculate the particle fitness score; The particle velocity vector is iteratively updated according to the global optimal particle position and the individual historical optimal position. When the optimal fitness score does not improve for ten consecutive generations, the parameter set corresponding to the current optimal particle is output.

5. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The dual-objective collaborative decision-making of air knife opening and zinc ingot addition amount based on the real-time zinc liquid temperature parameter and the optimized coating control rule set includes: Construct a decision variable space including air knife opening, zinc ingot adding rate, and steel strip running speed; Based on the current zinc liquid temperature parameter, a neighborhood solution set is randomly generated in the decision variable space; Calculate the coating thickness prediction deviation and zinc consumption cost weighted evaluation value of each neighborhood solution; The neighborhood solutions are accepted or discarded according to the annealing temperature curve, and the air knife opening instruction and zinc ingot addition instruction corresponding to the Pareto optimal solution are output when the annealing temperature drops to the termination threshold.

6. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The method of solving the speed constraint boundary of the tension roller group based on the coating thickness control instruction includes: Establish the tension roller group speed constraint equation and the coating thickness response delay constraint inequality group; Calculate the sub-gradient projection vector of the objective function under the current roller speed configuration; The initial solution of the roll gap parameters is updated along the subgradient projection direction, and the new solution is mapped to the speed-constrained feasible region using the projection operator. When the roller speed adjustment amplitude is less than the displacement threshold for three consecutive iterations, the roller system collaborative control parameters are locked.

7. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: Also includes: The coating quality feedback module uses a Kalman filter to fuse the coating thickness distribution data, and feeds the fused thickness distribution characteristics back to the feature extraction module, parameter optimization module and adaptive coating control module, triggering the dynamic reconfiguration of the coating control strategy.

8. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 7, characterized in that: The operation of the coating quality feedback module includes: Synchronously collect coating thickness distribution data at the air knife outlet and surface quality images at the cooling tower outlet; The Kalman filter is used to fuse the thickness gauge measurement value and the thickness distribution characteristics of image recognition to generate a thickness distribution compensation matrix; Input the thickness distribution compensation matrix into the parameter correction interface of the feature extraction module and update the temperature field gradient vector calculation rule; At the same time, the air knife pressure compensation coefficient is sent to the adaptive coating control module to trigger the real-time recalculation of the air knife opening instruction.

9. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 8, characterized in that: Generating a thickness distribution compensation matrix includes: Extract the horizontal fluctuation standard deviation and vertical trend slope of thickness distribution data; Identify the location distribution characteristics of zinc nodule defects in surface quality images; The Kalman gain coefficient is used to fuse the lateral fluctuation standard deviation and the zinc nodule distribution density data to generate the weight distribution vector of the thickness distribution compensation matrix.

10. The intelligent scheduling system for hot-dip galvanizing production process of steel plates according to claim 1, characterized in that: The global coordinated control of the roller system speed is achieved by: Analyze the tension roller acceleration constraint conditions in the coating thickness control instructions; The mixed integer programming problem of roll gap parameters is decomposed into speed allocation subproblem and delay compensation subproblem; The primal-dual interior point method is used to alternately solve the integer relaxation solutions of the dual gap and delay compensation subproblems of the velocity allocation subproblem. When the difference between the objective functions of the two sub-problems is less than the collaborative error threshold, the roller system speed collaborative control instruction set is output.

Images