Automatic control type coating roller painting algorithm

Abstract

The invention discloses an automatic control type coating roller coating algorithm, and relates to the technical field of metal surface treatment and industrial automation control, the algorithm collects real-time state data such as roller gap spacing, coating pressure, coating roller rotating speed and the like through a sensor system, and uses a soft measurement model constructed by a deep belief network to estimate the coating thickness; calculating the current deviation and the deviation change rate of the coating thickness; on the basis of the deviation, the deviation change rate and the prediction model confidence coefficient, distributing control weights of model prediction control and fuzzy adaptive PID through a dynamic weight calculation module; the model prediction control module generates a first control quantity through online parameter identification and rolling optimization, and the fuzzy adaptive PID module dynamically adjusts controller parameters to generate a second control quantity; according to the method, high-precision and stable control over the coating thickness under the complex working condition is achieved, manual intervention is reduced, the product quality is improved, and the production cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal surface treatment and industrial automation control, in particular to an automatic control type coating roller coating algorithm. BACKGROUND

[0002] In the metal plate processing industry, the roller coating technology is widely used for coating the surface of base materials such as steel strips. The uniformity and accuracy of the coating thickness are directly related to the corrosion resistance, appearance quality and production cost of the product. The traditional roller coating machine control relies on the experience of the operator to manually adjust the roll gap and pressure of the coating roller. This method has slow response, low control accuracy, and cannot adapt to the dynamic changes of process parameters such as production line speed and coating viscosity. In order to improve the control accuracy, the existing technology has appeared a scheme using an automatic control system.

[0003] For example, patent document CN103182357A discloses a "method for automatically adjusting the coating thickness of a roller coater", which uses a pressure sensor and a magnetic scale sensor to form a double closed-loop control system of pressure and position, and sets and adjusts the roll gap and pressure through a pre-established process parameter database. This scheme realizes the progress from manual to automatic, but its control logic largely depends on the pre-established static database, and when encountering conditions not recorded in the database or when the production conditions significantly drift, the adaptability and control effect of the system will decrease.

[0004] Another patent document CN118023057B proposes a "self-optimizing method and system applied to continuous roll-to-roll MPL coating production line", which detects the coating effect through an image capture device, and adjusts the coating roller speed and nozzle material spraying amount based on the detection results. This method introduces online detection and parameter self-optimization, but its control strategy is relatively direct, and for the complex process of roller coating with nonlinearity, time-varying and large lag characteristics, a single adjustment strategy is difficult to maintain optimal control performance under various conditions.

[0005] In summary, the existing roller coating automatic control systems based on fixed parameter rules or single control models generally have insufficient self-adaptability when facing complex and dynamically changing industrial production environments. The core defect is that it cannot intelligently adjust its core control strategy according to the real-time running state of the system, resulting in slow response or oscillation of the controller when the working conditions change widely, making it difficult to achieve the best balance between control speed, stability and accuracy, thereby restricting the further improvement of the coating thickness control accuracy.

[0006] Therefore, the field urgently needs an automatic control algorithm that can intelligently adapt to dynamic changes in the production process and has stronger robustness to solve the technical problem of maintaining high-precision coating thickness control under diversified working conditions. SUMMARY

[0007] The purpose of the present application is to make up for the shortcomings of the prior art, and provide an automatic control type coating roller coating algorithm, which collects real-time state data of the roller coating process through a sensor, estimates the coating thickness by means of a soft measurement model, dynamically allocates control weights in combination with coating thickness deviation, deviation change rate and prediction model confidence, and generates control instructions by fusing model predictive control and fuzzy adaptive PID, which can adapt to dynamic changes in working conditions and improve coating thickness control precision and stability.

[0008] To solve the above technical problems, the present application provides the following technical scheme: an automatic control type coating roller coating algorithm, comprising the following steps:

[0009] S1, collecting real-time state data of the roller coating process through a sensor system, wherein the real-time state data includes roll gap distance, coating pressure, roller speed and strip speed;

[0010] S2, calculating the current deviation and the current deviation change rate of the coating thickness based on the real-time state data;

[0011] S3, inputting the current deviation, the current deviation change rate and a prediction model confidence index into a dynamic weight calculation module to output a first weight value of model predictive control (MPC) output;

[0012] S4, using a model predictive control (MPC) module to perform rolling optimization based on an online parameter identification prediction model to generate a first control amount;

[0013] Using a fuzzy adaptive PID module to adjust the PID controller parameters according to the current deviation and the current deviation change rate and generate a second control amount;

[0014] S5, weighting and fusing the first control amount and the second control amount according to the first weight value to generate a final control instruction;

[0015] S6, outputting the final control instruction to an actuator to control the action of the coating roller.

[0016] Further, the dynamic weight calculation module is implemented by a fuzzy logic system, and the process of outputting the first weight value by the fuzzy logic system is characterized by the following mathematical formula:

[0017]

[0018] wherein,​ This represents the fuzzy inference and defuzzification mapping relationship of the dynamic weight calculation module. The current deviation is represented by a closed interval defined based on the maximum permissible deviation. This represents the current rate of change of deviation, and its universe of discourse is a closed interval defined based on the maximum permissible rate of change. This represents the confidence index of the prediction model, with its universe of discourse being a closed interval from 0 to 1. This index is calculated by using the model to predict and control the MPC module in the most recent... The normalized value is obtained by taking the reciprocal of the moving average of the prediction error over each control period.

[0019] Furthermore, S4 specifically includes:

[0020] The prediction model is a parameterized nonlinear state-space model based on the principles of fluid dynamics and mechanical transmission in the roll coating process, with the roll gap and roll speed as inputs and the coating thickness as output.

[0021] The key parameter vector in the prediction model is identified online in real time using the recursive least squares method. The key parameter vector includes at least the equivalent coating viscosity coefficient and the dynamic transfer gain.

[0022] In each control cycle, using the current system state as the initial condition, the updated prediction model is identified using parameters to predict the future. The system output sequence of step size;

[0023] By solving constrained optimization problems, the future can be calculated. Given a control input sequence of step size n, the performance index function of the optimization problem is:

[0024]

[0025] in, Indicates in Always Predicted value of coating thickness at any given time. express Coating thickness reference trajectory at any given time. express Control increment at any time, This represents the control increment weighting coefficient, used to balance the system's tracking performance with control stability.

[0026] The first element in the obtained control input sequence is used as the first control output.

[0027] Furthermore, S4 specifically includes:

[0028] The fuzzy adaptive PID module incorporates a fuzzy inference system, whose input variable is the current deviation. The fuzzy value and the current deviation change rate The fuzzy value, the output variable is the proportional coefficient adjustment amount. Integral coefficient adjustment and differential coefficient adjustment amount ;

[0029] The fuzzy inference system performs inference based on preset fuzzy rules, which define the inference at different deviations. and rate of change of deviation In combination, for , , Adjustment strategy;

[0030] The output of the fuzzy inference system is defuzzified to obtain precise parameter adjustment values. , , ;

[0031] Calculate the real-time PID controller parameters using the following formula:

[0032]

[0033] in, , , These are the preset initial parameters for the PID controller;

[0034] Using the adjusted real-time PID controller parameters, based on the current deviation Calculate the second control quantity .

[0035] Furthermore, S5 is specifically implemented through the following weighted fusion formula:

[0036]

[0037] in, This indicates the final control command. This represents the first control variable. This represents the second control variable. This represents the first weight value.

[0038] Furthermore, S1 specifically includes:

[0039] The roller gap spacing is measured using a magnetic scale sensor;

[0040] The coating pressure is measured by pressure sensors installed on the coating roller bearing housing and the adjusting roller.

[0041] The rotational speed of the coating roller and the speed of the strip are measured by a rotary encoder.

[0042] The measured roll gap, coating pressure, coating roll speed, and strip speed are transmitted to the central processing unit.

[0043] Furthermore, the confidence index of the prediction model Calculate using the following steps:

[0044] In each control cycle Calculate the prediction error of the model Its value is the actual measured value of the current coating thickness. Compared with model predictions The absolute value of the difference, i.e. ;

[0045] Maintain a length of A sliding window that stores the most recent Prediction error per control cycle ;

[0046] Calculate the average prediction error within the sliding window. ;

[0047] Calculate the confidence index of the prediction model. median value ;

[0048] right Linear normalization is performed to ensure that the value falls within the range of 0 to 1, thus obtaining the final confidence index of the prediction model. .

[0049] Furthermore, S2 specifically includes:

[0050] The real-time value of the coating thickness is obtained indirectly through a soft measurement model. The construction and use of this soft measurement model includes:

[0051] The soft measurement model is a deep belief network (DBN), whose input variables include roll gap, coating pressure, coating roll speed, coating flow rate, and strip tension.

[0052] During the offline training phase, the input variables covering different working conditions are collected along with corresponding data of the actual coating thickness obtained through offline measurement or high-precision sensors to form a training sample set, which is then used to perform unsupervised pre-training and supervised fine-tuning of the deep belief network DBN.

[0053] During the online application phase, the input variables collected in real time are input into the pre-trained deep belief network (DBN), and the output of the network is the estimated real-time coating thickness value.

[0054] Compared with existing technologies, this automatic control coating roller coating algorithm has the following advantages:

[0055] I. This invention comprehensively collects real-time status data of the roll coating process through a sensor system, accurately obtains coating thickness information using a soft measurement model constructed with a deep belief network, and dynamically allocates control weights by combining coating thickness deviation, deviation change rate, and prediction model confidence. It integrates the predictive optimization characteristics of model predictive control with the fast response capability of fuzzy adaptive PID to construct a dynamic collaborative control system. This system can adapt to the dynamic changes in production line speed, coating characteristics, and other working conditions in real time without relying on a static process database. This effectively solves the technical problems of insufficient adaptability of existing roll coating control systems based on fixed parameter rules or single control models, resulting in decreased control accuracy and poor stability when working conditions change over a wide range. This ensures the consistency and reliability of coating thickness control under complex working conditions.

[0056] Second, this invention uses the recursive least squares method to identify the parameters of the prediction model online, ensuring real-time matching between the control model and the actual working conditions. At the same time, it achieves complementary advantages of the two control strategies through a weighted fusion mechanism, which can significantly improve the control accuracy of coating thickness and the stability of the control process, reduce the frequent and violent movements of the actuator, and reduce equipment wear.

[0057] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of the process steps of the present invention;

[0060] Figure 2 This is a diagram of the control system architecture of the present invention. Detailed Implementation

[0061] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0062] Example 1

[0063] like Figure 1 and Figure 2 As shown in the figure, this embodiment discloses an automatic control algorithm for coating roller coating. This algorithm collects real-time status data of the roller coating process through a sensor system, calculates the current deviation and rate of change of the coating thickness, and dynamically allocates control weights for model predictive control and fuzzy adaptive PID control by combining the confidence index of the predictive model. The weighted fusion then generates the final control command, driving the actuator to adjust the coating roller's movement. This algorithm solves the problem of insufficient adaptive capability in traditional control schemes, achieving high-precision and stable control of coating thickness under complex working conditions, improving product quality and reducing production costs.

[0064] The automatic control coating roller coating algorithm described in this embodiment is based on a roller coating system, which includes a sensor system, a central processing unit, and an actuator. These components work together to automatically control the coating thickness. The following describes the specific working process in detail.

[0065] The system composition is described below:

[0066] Sensor system: Includes magnetic scale sensor, pressure sensor and rotary encoder, used to collect roller gap, coating pressure, coating roller speed and strip speed respectively; at the same time, in order to build a soft measurement model, a paint flow sensor and a strip tension sensor are also configured.

[0067] Central processing unit: It has built-in dynamic weight calculation module, model predictive control module, fuzzy adaptive PID module and soft measurement model, and is responsible for data processing, control quantity calculation and control command generation.

[0068] Actuators include a coating roller drive motor and a roller gap adjustment mechanism, which are used to receive control commands from the central processing unit and adjust the coating roller speed and roller gap.

[0069] The detailed implementation steps are as follows:

[0070] S1: Real-time status data acquisition

[0071] This step involves collecting key state parameters of the roll coating process using a sensor system, providing data support for subsequent control calculations.

[0072] The gap between the rollers is measured by a magnetic scale sensor, which is installed at the corresponding positions of the coating roller and the adjusting roller to directly obtain the gap size data between the two rollers.

[0073] The coating pressure is collected by pressure sensors installed on the coating roller bearing housing and the adjusting roller, reflecting the contact pressure between the coating roller and the strip in real time.

[0074] The coating roller speed and strip speed are measured by a rotary encoder. The encoder is coaxially connected to the coating roller shaft and the strip transfer roller, and the actual operating speed is obtained by converting the speed signal.

[0075] The coating flow rate is collected by a flow sensor, and the strip tension is collected by a tension sensor. Both types of data are used to estimate the coating thickness in the soft measurement model.

[0076] All data collected by the sensors is transmitted to the central processing unit in real time. After data preprocessing, it is used for subsequent calculations. The preprocessing process includes data filtering and outlier removal to ensure data reliability.

[0077] S2: Calculation of coating thickness deviation and deviation change rate

[0078] This step first obtains the real-time value of the coating thickness using a soft measurement model, and then calculates the current deviation and the rate of change of deviation.

[0079] The soft measurement model is constructed using a deep belief network. Its input variables are the roller gap, coating pressure, coating roller speed, coating flow rate and strip tension collected in S1, and the output is the estimated coating thickness.

[0080] The training process of the soft measurement model is divided into two stages: offline training and online application. In the offline training stage, input variable data under different working conditions and corresponding actual coating thickness data are collected to form a training sample set. The actual coating thickness is obtained through offline measurement or high-precision sensors. Unsupervised pre-training is performed on the DBN to initialize the weight parameters of each layer of the network. Then, the parameters are optimized through supervised fine-tuning to ensure that the error between the model output and the actual coating thickness meets the control requirements. In the online application stage, the input variables collected in real time are input into the trained DBN, and the network output is the real-time estimated value of the coating thickness.

[0081] Current deviation of coating thickness The calculation method is the difference between the reference value and the real-time estimated value of the coating thickness, and the reference value is set according to the production process requirements.

[0082] Current deviation change rate It is the ratio of the deviation difference between two adjacent control cycles to the duration of the control cycle, reflecting the trend of deviation change.

[0083] S3: Dynamic Weight Calculation

[0084] This step outputs the first weight value of the MPC module through the dynamic weight calculation module. This module is implemented using a fuzzy logic system, and its core formula is:

[0085]

[0086] In the formula, This represents the fuzzy reasoning and defuzzification mapping relationship of the dynamic weight calculation module, which is essentially a logical reasoning process based on preset fuzzy rules.

[0087] The current deviation of the coating thickness is defined as a closed interval based on the maximum allowable deviation in production. The range of the universe of discourse must cover the extreme deviations that may occur in actual production to ensure the comprehensiveness of fuzzy inference.

[0088] The current rate of change of deviation is defined as a closed interval based on the maximum allowable rate of change of deviation, in order to avoid imbalance in weight allocation due to excessively rapid changes in deviation.

[0089] The confidence index for the prediction model has a universe of discourse of [0,1]. This index directly reflects the reliability of the prediction model in the MPC module. The closer the value is to 1, the higher the prediction accuracy of the model, and vice versa.

[0090] Confidence index of prediction model The calculation process is as follows:

[0091] In each control cycle Calculate the prediction error of the model Its value is the actual measured value of the current coating thickness. Compared with model predictions The absolute value of the difference, i.e. .

[0092] Maintain a length of Sliding window, The value is set according to the dynamic characteristics of the production conditions to ensure that the window can cover a sufficient number of control cycles to reflect the changing trend of the model's predictive performance. The window stores the most recent... Prediction error per control cycle .

[0093] Calculate the average prediction error within the sliding window. This average value reflects the overall error level of recent model predictions.

[0094] Calculate the confidence median The median value is negatively correlated with the average prediction error; the smaller the average error, the larger the median value.

[0095] right Linear normalization is performed to ensure that the value falls within the interval [0,1], thus obtaining the final confidence index of the prediction model. .

[0096] The fuzzy inference process of the dynamic weight calculation module follows preset fuzzy rules, such as when Smaller Smaller and When it is large, Take the larger value to highlight the control role of MPC; when Larger Larger and When smaller, The smaller value is selected to enhance the control effect of the fuzzy adaptive PID. The fuzzy rule is designed based on the control characteristics of the roll coating process and engineering practice experience. After defuzzification, the accurate first weight value is output. .

[0097] S4: Control quantity generation

[0098] This step generates the first and second control variables using the MPC module and the fuzzy adaptive PID module, respectively. The specific implementation process is as follows:

[0099] The MPC module generates the first control input:

[0100] Predictive Model Construction: The predictive model is a parametric nonlinear state-space model established based on the fluid dynamics and mechanical transmission principles of the roll coating process. The inputs are the roll gap and the roll speed, and the output is the coating thickness. This model fully considers the influence of coating fluid characteristics and the mechanical transmission law of roll coating on the coating thickness, and can accurately describe the dynamic characteristics of the roll coating process.

[0101] Online parameter identification: The recursive least squares method is used to identify the key parameter vectors in the prediction model in real time. The key parameter vectors include at least the equivalent coating viscosity coefficient and dynamic transfer gain. The recursive least squares method updates the parameter estimates iteratively to minimize the sum of squared errors between the model output and the actual output, ensuring that the model can adapt to changes in coating characteristics, equipment status, and other factors in real time.

[0102] Rolling optimization: In each control cycle, using the current system state as the initial condition, the updated prediction model is identified using parameters to predict the future. The system output sequence of step size, The step size is set according to the system response speed and control accuracy requirements.

[0103] Optimization problem solving: Determining the future by solving constrained optimization problems. Given a control input sequence of length n, the performance index function for the optimization problem is:

[0104]

[0105] In the formula, for The coating thickness reference trajectory at any given time is set according to the production process requirements, and can be either a step trajectory or a smooth transition trajectory. for The control increment at any given moment is the difference between the current control input and the control input at the previous moment. To control the incremental weighting coefficient, which is used to balance the system's tracking performance and control stability, The larger the value, the smoother the change in control input, avoiding frequent and drastic movements of the actuator. The smaller the value, the faster the system tracks the reference trajectory. It needs to be set reasonably according to the control accuracy and equipment wear requirements in actual production.

[0106] Control output: Obtain the future value after solving the optimization problem. A control input sequence of step size is used, and the first element of the sequence is taken as the first control quantity. The output is then sent to the subsequent weighted fusion module.

[0107] The fuzzy adaptive PID module generates the second control input:

[0108] Fuzzy Inference System Configuration: The fuzzy adaptive PID module has a built-in fuzzy inference system, with the current deviation as the input variable. The fuzzy value and the current rate of change of deviation The fuzzy value, the output variable is the proportional coefficient adjustment amount. Integral coefficient adjustment and differential coefficient adjustment amount .

[0109] Blurring: and The precise value is converted into a fuzzy value. The fuzzification process uses a triangular membership function to convert the value into a fuzzy value. and The domain of discourse is divided into multiple fuzzy subsets, each of which corresponds to a specific range of membership degrees.

[0110] Fuzzy rule-based reasoning: A fuzzy reasoning system performs reasoning based on preset fuzzy rules, which define different... and Combination , , Adjustment strategies. For example, when For "upright" and When the negative is small, Take "upright and righteous" Take "small" By adopting the "smallest negative" approach, the response speed is accelerated by enhancing the proportional action, the integral action is appropriately introduced to eliminate steady-state error, and the derivative action is weakened to avoid system oscillation. The fuzzy rules are formulated based on the PID control principle and experience in roller coating process control to ensure the rationality and comprehensiveness of the rules.

[0111] Defuzzification: The output of the fuzzy inference system is defuzzified, and the centroid method is used to convert the fuzzy output into precise parameter adjustment values. , , This ensures the accuracy of parameter adjustments.

[0112] PID parameter update: Calculate the real-time PID controller parameters according to the following formula:

[0113]

[0114]

[0115]

[0116] In the formula, , , These are the preset initial parameters for the PID controller. The initial parameters are determined through offline debugging and optimized based on the control effect under typical operating conditions. , , These are the updated real-time PID controller parameters.

[0117] Second control input calculation: Using the updated real-time PID controller parameters, based on the current deviation... The second control quantity is calculated using the PID control algorithm. The PID control algorithm uses a positional control formula to ensure the real-time performance and accuracy of the control quantity.

[0118] S5: Weighted fusion generates final control command

[0119] This step is based on the first weight value. For the first control quantity With the second control quantity Weighted fusion is performed, and the fusion formula is as follows:

[0120]

[0121] In the formula, For the final control command, The first weight value obtained in S3, These are the weight values ​​for the fuzzy adaptive PID module. This formula enables the dynamic combination of two control strategies; when the prediction model has high confidence and the system operating conditions are stable, The value is relatively large. It accounts for a higher proportion in the final control commands, fully leveraging the predictive optimization advantages of MPC; when the system operating conditions change drastically and the confidence of the predictive model decreases, The value decreases. The proportion is increased, and the fast response characteristics of fuzzy adaptive PID are used to ensure control stability.

[0122] S6: Control command execution

[0123] The final control command generated by S5 The output is sent to the actuator, which adjusts the motion parameters of the coating roller according to the control command. If the control command is for the coating roller speed, the drive motor adjusts its speed output accordingly; if the control command is for the roller gap, the roller gap adjustment mechanism adjusts the gap between the coating roller and the adjusting roller through mechanical transmission. After the actuator operates, the sensor system continuously collects new real-time status data, enters the next control cycle, and forms a closed-loop control to ensure that the coating thickness is always maintained within the set range.

[0124] In summary, this embodiment comprehensively collects process status data of the roll coating process through a sensor system, accurately estimates the coating thickness using a soft sensor model, and dynamically allocates control weights by combining deviation information and prediction model confidence. It integrates the predictive optimization capabilities of MPC with the fast response characteristics of fuzzy adaptive PID to construct a complete automatic control scheme. This scheme does not rely on a static process database, can adapt to changes in operating conditions in real time, overcomes the shortcomings of insufficient adaptive capabilities in existing technologies, effectively improves the accuracy and stability of coating thickness control, reduces the need for manual intervention, and is suitable for various roll coating production scenarios for metal sheets.

[0125] Example 2

[0126] like Figure 1 and Figure 2 As shown, this embodiment uses a cold-rolled strip steel epoxy coating roll coating production line as an application scenario. Addressing the practical needs of implementing an automated control-type coating roll coating algorithm, it details the specific operational methods for key aspects such as system hardware deployment, soft measurement model implementation, control logic application, and actuator coordination. By combining the equipment characteristics of the production line with the epoxy coating process requirements, the operational details of each module of the algorithm in the industrial scenario are clarified, verifying the operability and stability of the technical solution and ensuring that the coating thickness control accuracy meets the production standards for cold-rolled strip steel anti-corrosion coatings.

[0127] This embodiment establishes a control system based on a cold-rolled strip epoxy coating roll coating production line. This production line mainly produces cold-rolled strip steel with a thickness of 0.3-2.0mm, a target coating thickness of 5-30μm, and a strip running speed range of 10-100m / min. The following details the implementation process of the automatic control type coating roll coating algorithm in conjunction with the actual operation requirements of the production line.

[0128] The sensor system is adapted and deployed on the production line as follows:

[0129] In the production of epoxy coating for cold-rolled strip steel, the coating is prone to splashing, and the strip is subject to slight vibration during operation. Therefore, sensor deployment must balance measurement accuracy and anti-interference capabilities. The specific deployment method is as follows:

[0130] Roll gap measurement: Magnetic scale sensors are installed on the bearing supports at both ends of the coating roller and adjusting roller, using a symmetrical arrangement. The distance between the sensor probe and the measuring surface is set according to the recommended value in the equipment manual, and a stainless steel protective cover to prevent paint splashing is installed on the outside of the probe. During operation, the average value of the sensor readings on both sides is used as the final value of the roll gap, eliminating unilateral measurement deviation caused by slight bending of the coating roller.

[0131] Coating pressure measurement: Strain gauge pressure sensors are installed at the pressure transmission nodes of the bearing seats on both sides of the coating roller and at the pressure detection port of the adjusting roller's pressure cylinder. The sensors on the bearing seats are used to monitor the pressure balance at both ends of the coating roller, while the sensors on the pressure cylinder are used to directly obtain the pressure applied by the adjusting roller to the coating roller. The sensor signal cables are dedicated cables with metal shielding, fixed along the production line equipment support to avoid parallel laying with motor cables and reduce electromagnetic interference.

[0132] Rotational speed and strip speed measurement: The incremental rotary encoder is connected to the coating roller drive shaft and the strip conveying tension roller shaft via flexible couplings. The couplings are selected to match the shaft diameter to ensure that the coaxiality error is controlled within the allowable range of the equipment. The pulse signal output by the encoder is directly connected to the high-speed counting module of the central processing unit. Through the conversion relationship between the pulse number and the shaft diameter, the rotational speed of the coating roller and the strip speed are obtained in real time. To avoid speed errors caused by strip slippage, the encoder installation position at the tension roller should avoid the strip overlap section and be selected in an area where the strip is conveyed flat.

[0133] Coating flow rate and strip tension measurement: A turbine flow sensor is installed in the vertical section of the coating supply pipeline to collect the real-time supply of epoxy coating; a strip tension sensor is installed between the inlet tension roller and the outlet tension roller of the strip conveyor to directly measure the actual tension of the strip during operation, avoiding changes in the contact pressure between the strip and the coating roller due to tension fluctuations, which would affect the coating thickness.

[0134] Data preprocessing: After receiving sensor data, the central processing unit uses a moving average filter to remove instantaneous pulse interference. At the same time, it sets up data validity judgment logic. When a sensor data exceeds the process allowable extreme value range, the system automatically calls the average value of the previous 3 control cycles as temporary data to ensure that the control process is not interrupted. Real-time data acquisition is automatically restored after the sensor fault is cleared.

[0135] The training and application of the soft measurement model on the production line are as follows:

[0136] The viscosity of the epoxy coating on cold-rolled strip steel fluctuates with changes in ambient temperature. The soft measurement model needs to be trained and updated regularly to ensure the accuracy of coating thickness estimation. The specific operation is as follows:

[0137] Training sample collection: The sample collection covers typical operating conditions of the production line, including 3 epoxy coating viscosity grades, 5 strip speed ranges, and 6 target coating thicknesses. 50 sets of data are continuously collected under each operating condition. Each set of data includes real-time values ​​of roller gap, coating pressure, coating roller speed, coating flow rate, strip tension, and the synchronous actual coating thickness value.

[0138] Sample preprocessing: Outlier samples were removed using the 3σ criterion, and then all input variables were normalized to eliminate model training bias caused by differences in variable magnitude.

[0139] DBN Model Training: In the offline pre-training phase, the DBN is configured with two Restricted Boltzmann Machine (RBM) layers. The first RBM has the same number of visible layer nodes as the number of input variables, and 10 hidden layer nodes. The second RBM has 10 visible layer nodes and 8 hidden layer nodes. A contrastive divergence algorithm is used for layer-by-layer training with an initial learning rate of 0.05. Training stops when the reconstruction error of each RBM layer changes less than 0.001 for 100 consecutive iterations. In the supervised fine-tuning phase, an output layer is added to the top layer of the DBN. Gradient descent is used to optimize the weights of the entire network. The training, validation, and test sets are divided in a 7:2:1 ratio, with mean squared error used as the loss function. Fine-tuning stops when the validation set error does not decrease for 50 consecutive iterations, ultimately ensuring that the thickness estimation error of the test set is less than 1 μm.

[0140] Online model updates: During production line operation, for every 2000 sets of online data accumulated, 100 sets of samples that can be verified for actual thickness through offline retesting are selected as model update samples. Incremental learning is adopted to fine-tune only the weights of the top output layer of the DBN and the second RBM layer, avoiding production line downtime caused by full training. The update frequency is adjusted according to the season: once every two weeks when the ambient temperature changes greatly in summer, and once a month when the environment is stable in winter, to ensure that the model always adapts to changes in coating characteristics.

[0141] The dynamic weight calculation module is adapted to the following operating conditions:

[0142] In the production of epoxy coating for cold-rolled strip steel, changes in operating conditions mainly manifest in two scenarios: strip speed adjustment and coating viscosity fluctuation. The fuzzy rules of the dynamic weight calculation module need to be adapted to the control requirements of these two scenarios. The specific application method is as follows:

[0143] In scenarios with stable strip speed: When the strip speed is maintained within a certain fixed range and the epoxy coating viscosity is stable, the system deviation is small and changes gradually. If the deviation is "very small", the deviation change rate is "very slow", and the model confidence is "very high", the dynamic weight calculation module outputs the first weight value as "maximum". At this time, the MPC module takes the lead in control, reducing the adjustment range of the coating roller speed and roller gap through prediction optimization, avoiding small fluctuations in coating thickness, and ensuring the uniformity of the coating on the strip surface.

[0144] Scenarios of sudden changes in coating viscosity: When the ambient temperature drops sharply, causing the viscosity of the epoxy coating to increase, the coating thickness is prone to deviation and changes rapidly. If the deviation is "large", the rate of change of deviation is "fast", and the model confidence is "low", the dynamic weight calculation module outputs the first weight value as "very small". At this time, the fuzzy adaptive PID module dominates the control, quickly adjusting the roller gap to reduce the coating pressure and avoid the coating from becoming too thick due to the increase in coating viscosity. After the deviation is reduced to below "medium" and the model confidence recovers, the first weight value is gradually increased to restore the optimization effect of MPC.

[0145] Strip speed switching scenario: When the strip speed increases, the coating thickness deviation and the rate of change of deviation are both at a "medium" level, and the model confidence level temporarily drops to a "medium" level due to the change in operating conditions. At this time, the dynamic weight calculation module outputs a first weight value of "medium". MPC and fuzzy adaptive PID collaborative control: MPC plans the adjustment trend of the coating roll speed based on the updated prediction model, and fuzzy adaptive PID finely adjusts the roll gap in real time to compensate for the pressure fluctuations caused by the speed change, ensuring that the coating thickness deviation is controlled within the allowable range of the process during the speed switching process.

[0146] The process constraint configuration for the MPC module is as follows:

[0147] Considering the equipment limitations and process requirements of the cold-rolled strip epoxy coating production line, the following constraints are set for the optimization problem of the MPC module to ensure the safety and feasibility of the control quantities:

[0148] Control input constraints:

[0149] Coating roller speed constraint: The upper limit is the rated speed of the coating roller drive motor, and the lower limit is the minimum speed allowed by the process;

[0150] Roll gap spacing constraint: The upper limit is the maximum spacing allowed by the mechanical limit of the equipment, and the lower limit is the minimum safe spacing.

[0151] Control Incremental Constraints:

[0152] The single change in coating roller speed shall not exceed 10% of the rated speed;

[0153] The single change in roll gap shall not exceed 50 μm.

[0154] Output constraints: The predicted coating thickness must be controlled within ±2μm of the target thickness. If the predicted value exceeds this range during the optimization process, the MPC module will prioritize adjusting the roll gap spacing, which has a more significant impact on the coating thickness, and then fine-tune the coating roll speed to ensure that the predicted output returns to the qualified range.

[0155] The production line coordinated control of the actuators is as follows:

[0156] In the production of epoxy coating for cold-rolled strip steel, the coordinated adjustment of the coating roll drive motor and the roll gap adjustment mechanism directly affects the stability of the coating thickness. The specific linkage method is as follows:

[0157] Scenarios where the coating thickness is relatively thin:

[0158] Prioritize adjusting the roller gap: The roller gap adjustment mechanism gradually reduces the gap according to the control command, increases the coating pressure, and improves the transfer of epoxy coating.

[0159] Synchronous fine-tuning of coating roller speed: If the coating thickness still does not reach the target value after three control cycles of spacing adjustment, the coating roller speed is appropriately reduced to prolong the contact time between the coating and the strip and help improve the coating thickness.

[0160] Scenarios where the coating thickness is relatively thick:

[0161] Prioritize adjusting the coating roller speed: The coating roller drive motor gradually increases its speed according to the control command to reduce the amount of coating transferred;

[0162] Assist in adjusting the roll gap: If the coating thickness is still out of tolerance after adjusting the speed for 3 control cycles, increase the roll gap appropriately and reduce the coating pressure to avoid insufficient contact between the strip and the coating roller due to a direct and significant increase in the gap.

[0163] Scenarios requiring increased strip speed:

[0164] Synchronous adjustment of coating roller speed: The coating roller drive motor increases its speed synchronously with the increase of the strip speed, maintaining a stable ratio of coating roller speed to strip speed;

[0165] Compensatory adjustment of roll gap: When the strip speed is increased, the coating pressure will decrease slightly. The roll gap adjustment mechanism appropriately reduces the gap to compensate for the pressure loss and maintain a stable coating transfer volume.

[0166] The system debugging and trial operation of the production line are as follows:

[0167] To ensure the algorithm runs stably on the production line, the following debugging and trial operation steps must be completed before initial use:

[0168] Sensor calibration: The magnetic scale sensor is calibrated using a standard gauge block. Five calibration points are selected within the measurement range, and the sensor zero point and gain are adjusted to ensure that the reading error is less than 1μm. The pressure sensor is calibrated using a standard pressure source with three pressure points of 5kN, 15kN, and 25kN. The calibration curve is fitted to ensure that the measurement error is less than 1% of the full scale. The encoder is verified using a standard speed source, and the counting coefficient is adjusted to ensure that the speed measurement error is less than 0.5%.

[0169] Model Validation: Three typical working conditions were selected. After the production line was running stably, sensor data and the actual thickness measured by offline laser thickness measurement were collected simultaneously. The estimated value of the soft measurement model was compared with the actual value. The error was controlled within 2μm, and the model validation was qualified.

[0170] Control parameter tuning: The initial parameters of the PID were tuned using the step response method. A step input of 50μm was applied to the roll gap, and the response curve of the coating thickness was recorded. The initial parameters were adjusted so that the overshoot was less than 5% and the settling time was less than 2 seconds. The prediction step size of the MPC was set to 8 control cycles and the control step size was set to 3 control cycles. The control increment weight coefficient was determined through simulation verification to ensure a balance between tracking performance and control stability.

[0171] Trial run: After debugging, a 48-hour continuous trial run was conducted. During the trial run, three strip speeds and two target coating thicknesses were randomly switched. The coating thickness deviation was monitored in real time and controlled within ±2μm. There were no defects such as coating sagging or missing coating. The trial run was successful and the system was officially put into production.

[0172] In summary, this embodiment, through practical adaptation to a cold-rolled strip epoxy coating roll coating production line, clarifies the hardware deployment, model training, control application, and debugging process of the automatic control coating roll coating algorithm in an industrial scenario. It verifies the algorithm's control accuracy and stability under dynamic operating conditions, meeting the production requirements of cold-rolled strip anti-corrosion coatings.

[0173] 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. An automatic control algorithm for roller coating of paint, characterized in that, Includes the following steps: S1. Real-time status data of the roll coating process is collected through a sensor system. The real-time status data includes the roll gap, coating pressure, roll speed and strip speed. S2. Based on the real-time status data, calculate the current deviation of the coating thickness and the rate of change of the current deviation; S3. Taking the current deviation, the rate of change of the current deviation, and a prediction model confidence index as inputs, the first weight value of the model prediction control MPC output is output through a dynamic weight calculation module. S4. Using a Model Predictive Control (MPC) module, a rolling optimization is performed based on an online parameter identification predictive model to generate the first control quantity. A fuzzy adaptive PID module is used to adjust the PID controller parameters and generate a second control quantity based on the current deviation and the rate of change of the current deviation. S5. Based on the first weight value, the first control quantity and the second control quantity are weighted and fused to generate the final control command; S6. Output the final control command to the actuator to control the action of the coating roller.

2. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, The dynamic weight calculation module is implemented using a fuzzy logic system, and it outputs the first weight value. The process is represented by the following mathematical formula: in, This represents the fuzzy inference and defuzzification mapping relationship of the dynamic weight calculation module. The current deviation is represented by a closed interval defined based on the maximum permissible deviation. This represents the current rate of change of deviation, and its universe of discourse is a closed interval defined based on the maximum permissible rate of change. This represents the confidence index of the prediction model, with its universe of discourse being a closed interval from 0 to 1. This index is calculated by using the model to predict and control the MPC module in the most recent... The normalized value is obtained by taking the reciprocal of the moving average of the prediction error over each control period.

3. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, S4 specifically includes: The prediction model is a parameterized nonlinear state-space model based on the principles of fluid dynamics and mechanical transmission in the roll coating process, with the roll gap and roll speed as inputs and the coating thickness as output. The key parameter vector in the prediction model is identified online in real time using the recursive least squares method. The key parameter vector includes at least the equivalent coating viscosity coefficient and the dynamic transfer gain. In each control cycle, using the current system state as the initial condition, the updated prediction model is identified using parameters to predict the future. The system output sequence of step size; By solving constrained optimization problems, the future can be calculated. Given a control input sequence of step size n, the performance index function of the optimization problem is: in, Indicates in Always Predicted value of coating thickness at any given time. express Coating thickness reference trajectory at any given time. express Control increment at any time, This represents the control increment weighting coefficient, used to balance the system's tracking performance with control stability. The first element in the obtained control input sequence is used as the first control output.

4. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, S4 further specifically includes: The fuzzy adaptive PID module incorporates a fuzzy inference system, whose input variable is the current deviation. The fuzzy value and the current deviation change rate The fuzzy value, the output variable is the proportional coefficient adjustment amount. Integral coefficient adjustment and differential coefficient adjustment amount ; The fuzzy inference system performs inference based on preset fuzzy rules, which define the inference at different deviations. and rate of change of deviation In combination, for , , Adjustment strategy; The output of the fuzzy inference system is defuzzified to obtain precise parameter adjustment values. , , ; Calculate the real-time PID controller parameters using the following formula: in, , , These are the preset initial parameters for the PID controller; Using the adjusted real-time PID controller parameters, based on the current deviation Calculate the second control quantity .

5. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, S5 is specifically implemented through the following weighted fusion formula: in, This indicates the final control command. This represents the first control variable. This represents the second control variable. This represents the first weight value.

6. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, S1 specifically includes: The roller gap spacing is measured using a magnetic scale sensor; The coating pressure is measured by pressure sensors installed on the coating roller bearing housing and the adjusting roller. The rotational speed of the coating roller and the speed of the strip are measured by a rotary encoder. The measured roll gap, coating pressure, coating roll speed, and strip speed are transmitted to the central processing unit.

7. The automatic control type coating roller coating algorithm according to claim 2, characterized in that, The confidence index of the prediction model Calculate using the following steps: In each control cycle Calculate the prediction error of the model Its value is the actual measured value of the current coating thickness. Compared with model predictions The absolute value of the difference, i.e. ; Maintain a length of A sliding window that stores the most recent Prediction error per control cycle ; Calculate the average prediction error within the sliding window. ; Calculate the confidence index of the prediction model. median value ; right Linear normalization is performed to ensure that the value falls within the range of 0 to 1, thus obtaining the final confidence index of the prediction model. .

8. The automatic control type coating roller coating algorithm according to claim 1, characterized in that, S2 specifically includes: The real-time value of the coating thickness is obtained indirectly through a soft measurement model. The construction and use of this soft measurement model includes: The soft measurement model is a deep belief network (DBN), whose input variables include roll gap, coating pressure, coating roll speed, coating flow rate, and strip tension. During the offline training phase, the input variables covering different working conditions are collected along with corresponding data of the actual coating thickness obtained through offline measurement or high-precision sensors to form a training sample set, which is then used to perform unsupervised pre-training and supervised fine-tuning of the deep belief network DBN. During the online application phase, the input variables collected in real time are input into the pre-trained deep belief network (DBN), and the output of the network is the estimated real-time coating thickness value.

Images