A three-roll grinder roll shaft spacing error compensation control method

Abstract

The present application belongs to the technical field of three-roll grinder control, and specifically relates to a three-roll grinder roll shaft spacing error compensation control method, which comprises the following steps: obtaining the working gap between the rolls of the three-roll grinder, the operation parameters of the roll driving motors, and the temperature distribution data of the roll bodies; estimating the elastic deflection amount of each roll under the shearing load in real time, and separating the pure geometric gap component and the thermal drift component from the working gap; constructing an inverse model connecting the driving amount of the micro-displacement actuator and the change of the roll gap; predicting the gap thermal drift increment and the deflection increment of the next control cycle, and feeding the prediction results into the actuator driving instruction in the form of feedforward; in the present application, the micro-displacement sensor measurement values are decoupled by the method, and trend prediction is performed, and the micro-displacement actuator is driven to dynamically adjust the roll shaft spacing; the hysteresis error introduced by thermal inertia and load mutation is overcome, and the working condition adaptability, roll gap control precision and long-term stability of the three-roll grinder are improved.

Description

Technical Field

[0001] This invention belongs to the field of three-roll mill control technology, specifically a method for compensating for the error in the roller spacing of a three-roll mill. Background Technology

[0002] The three-roll mill is a key piece of equipment widely used in the fine dispersion and grinding of high-viscosity materials such as coatings, inks, and electronic pastes. Its performance directly affects the fineness, uniformity, and stability of the final product. Its core working principle is that three horizontally arranged rollers with different speeds rotate relative to each other, forming an adjustable, wedge-shaped, gradual gap between adjacent roller surfaces, ranging from tens of micrometers to submicrometers. The narrow gap between the rollers generates extremely high shear and extrusion forces on the material, causing the material to withstand strong shear, extrusion, and stretching caused by the speed difference between the roller surfaces in a very short time as it passes through the gap. This achieves uniform refinement of the material, meeting the process requirements of fully deagglomerated and uniformly dispersed primary particles.

[0003] Existing grinding mill roller gap error control typically uses post-processing feedback to adjust the roller gap, which cannot decouple thermal drift and elastic deflection in real time, resulting in large thermal inertia, delayed compensation, and a sharp increase in roller gap error when the load changes abruptly. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention proposes a method for compensating and controlling the roller spacing error of a three-roll mill. This method is mainly used to solve the problem that existing roller spacing error control methods for grinding mills typically use post-processing feedback to adjust the roller gap, which cannot decouple thermal drift and elastic deflection in real time.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A method for compensating and controlling the roller spacing error of a three-roll mill includes: The first working gap between the rear roller and the middle roller of the three-roll mill and the second working gap between the middle roller and the front roller are obtained. The operating parameters of the drive motors of each roller and the temperature distribution data of each roller are collected. Based on the operating parameters, the elastic deflection of each roller under shear load is estimated in real time by using an extended Kalman filter, and the pure geometric gap component and thermal drift component are separated from the first working gap and the second working gap to obtain the gap state decomposition quantity group. An inverse model is constructed that connects the driving quantity of the micro-displacement actuator with the change of the roll gap. The difference between the gap state decomposition group and the preset reference gap is used as input. After the inverse model is calculated, the actuator driving command is generated, which directly drives the actuator to perform mechanical domain closed-loop correction of the roll gap. Based on temperature distribution data and real-time torque trends, the gap thermal drift increment and deflection increment for the next control cycle are predicted by thermal expansion prediction model and deflection load prediction model, respectively. The prediction results are then injected into the actuator drive command in a feedforward manner.

[0007] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill includes the following steps for real-time estimation of the elastic deflection of each roller under shear load: The real-time torque, real-time speed, elastic modulus, and moment of inertia of each roller are taken from the operating parameters of each roller drive motor as inputs to the extended Kalman filter. A state-space equation is established with roller deflection as the state variable and motor torque as the observation variable. The process equation is constructed based on the elasticity model of a simply supported beam, and the observation equation is constructed based on the mapping relationship between torque and shear force. Through recursive calculations based on time updates and measurement updates, the estimated value of the elastic deflection of each roller under shear load is output in real time.

[0008] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill includes the following steps for obtaining the set of decomposed parameters of the gap state: The first and second working gaps collected by the non-contact micro-displacement sensor are recorded as the measured total gap; Subtract the estimated elastic deflection from the measured total gap to obtain the pure geometric gap component; Based on the temperature distribution data of each roller, the radial thermal expansion of each roller caused by frictional heat generation and material heat exchange is calculated by the thermal expansion decoupling algorithm, and then converted into a thermal drift component. The pure geometric gap component, thermal drift component, and elastic deflection estimate are combined to form a set of gap state decomposition quantities.

[0009] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill is provided, wherein the decoupling calculation formula for each component in the gap state decomposition group is expressed as follows:

[0010] In the formula, Let t be the working interval for sensor data acquisition. It is a pure geometric gap component. For thermal drift components, This refers to the elastic deflection. The observation function of the extended Kalman filter takes into account the motor torque as input. Real-time rotational speed ω, roller shaft elastic modulus E, and moment of inertia I; This is the raw gap value measured directly by a non-contact sensor.

[0011] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill includes the following steps in constructing an inverse model: Offline frequency sweep excitation was performed on the combined system of micro-displacement actuator and roller system flexibility to obtain the frequency response function from the actuator drive amount to the change of roller gap; The frequency response function is discretized and inverted to obtain the initial inverse model; The inverse model is configured as a feedforward filter in the digital controller, with its output connected in parallel with the closed-loop feedback correction path of the actuator.

[0012] According to the present invention, a method for compensating for roller spacing error in a three-roll mill includes the following steps in generating actuator drive commands: The total gap error is obtained by calculating the difference between the sum of each component in the gap state decomposition set and the preset reference gap. The total gap error is input into the inverse model for calculation to obtain the feedforward driving amount; The actual displacement feedback value of the micro-displacement actuator is collected, the feedback error between it and the target displacement is calculated, and the feedback correction amount is generated through the proportional-integral controller. The feedforward drive quantity and the feedback correction quantity are summed to obtain the final actuator drive command.

[0013] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill includes the following steps in constructing a thermal expansion prediction model: Multiple temperature sensors are arranged axially and radially inside the rollers of the three-roll mill to form a temperature monitoring array; Based on data from the temperature monitoring array, a mapping relationship between the roller temperature distribution and radial thermal expansion is established using the lumped parameter thermal network method. Based on the mapping relationship, the radial thermal expansion of each roller in the next control cycle is predicted as the gap thermal drift increment.

[0014] According to the present invention, a method for compensating and controlling the roller spacing error of a three-roll mill includes the following steps in constructing a flexural load prediction model: Based on real-time torque trend data, the material shear load entering the roll gap area in the next control cycle is predicted by the transmission relationship between material shear force and motor torque. Using the elasticity equation of a simply supported beam, the expected bending deformation of the roller is calculated based on the shear load of the material; the expected bending deformation is then used as the deflection increment.

[0015] According to the present invention, a method for compensating for the roller spacing error of a three-roll mill includes the step of injecting the prediction result into the actuator drive command in a feedforward manner, comprising: The gap thermal drift increment and deflection increment are respectively converted into the corresponding equivalent displacement compensation amounts; The equivalent displacement compensation is added to the total clearance error to form a feedforward input with predictive compensation. The inverse model is computed based on feedforward inputs with predictive compensation, thereby enabling the generated actuator drive instructions to include anticipatory suppression of gap drift in the next control cycle.

[0016] According to the present invention, a method for compensating for the roller spacing error of a three-roll mill is provided, in which the calculation formula of the inverse model based on the feedforward input with predictive compensation is expressed as follows:

[0017] In the formula, The predicted compensation driving amount injected into the feedforward path at time t. This is the inverse model of the micro-displacement actuator and the roller system flexibility. This is the feedforward gain coefficient. The coefficient of thermal expansion of the material. The thermal expansion coupling weight is the value for the i-th temperature measurement point. Let τ be the temperature change at the i-th temperature measurement point predicted with a lead time of τ. Where L is the load-deflection conversion factor, L is the roller support span, and I is the moment of inertia of the roller section. τ represents the change in material shear load predicted with a lead time of τ, and N represents the total number of temperature measurement points.

[0018] This invention decouples the measurements from the micro-displacement sensor using an extended Kalman filter algorithm, separating the pure geometric gap, thermal drift, and elastic deflection in real time. Based on this, a lumped-parameter thermal network model and material load trend prediction are introduced to proactively calculate the incremental changes in thermally and mechanically induced gaps, generating a predictive feedforward compensation. This compensation works in conjunction with a feedback correction loop to drive the micro-displacement actuator to dynamically adjust the roller gap. This upgrades traditional passive feedback compensation to active closed-loop control based on real-time decoupling, multi-physics modeling, and predictive feedforward, effectively overcoming the hysteresis errors introduced by thermal inertia and sudden load changes. It significantly improves the roller gap control accuracy and long-term stability of the three-roll mill under conditions of high-viscosity materials, speed fluctuations, and long-term continuous grinding, achieving consistent gap spacing. Attached Figure Description

[0019] The invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart of a method for compensating for the roller spacing error of a three-roll mill according to an embodiment of the present invention; Figure 2 This is a flowchart of obtaining the gap state decomposition quantity group in an embodiment of the present invention; Figure 3This is a flowchart in an embodiment of the present invention, showing how the prediction results are injected into the executor driving instructions in a feedforward manner. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0022] Example: like Figures 1 to 3 As shown, this embodiment provides a method for compensating and controlling the roller spacing error of a three-roll mill, including: The first working gap between the rear roller and the middle roller of the three-roll mill and the second working gap between the middle roller and the front roller are obtained. The operating parameters of the drive motors of each roller and the temperature distribution data of each roller are collected. The first working gap refers to the minimum physical distance between the roller surfaces of the rear roller and the middle roller. The second working gap refers to the minimum physical distance between the roller surfaces of the middle roller and the front roller. The "operating parameters" include at least the real-time output torque and real-time speed of the motor. The temperature distribution data refers to the array set that can characterize the axial and radial temperature field distribution of the roller due to frictional heat generation and material heat exchange, collected by multiple temperature sensors distributed inside or on the surface of the roller.

[0023] Based on operating parameters, the elastic deflection of each roller under shear load is estimated in real time using an extended Kalman filter. The steps include: The real-time torque, real-time speed, elastic modulus, and moment of inertia of each roller drive motor are used as inputs to the extended Kalman filter; the state vector of the extended Kalman filter is taken as x=[δ, '] T δ is the mid-span deflection of the roller shaft. 'This represents the rate of change of the roller's mid-span deflection, initially set to x0|0=[0,0] T The initial covariance matrix P0|0 is taken with 10 elements on the main diagonal. -6 The diagonal matrix. The modulus of elasticity is a measure of a material's resistance to elastic deformation, found in roller material mechanical property handbooks or calibrated through bending tests; the moment of inertia is a geometric parameter characterizing the roller's cross-section's resistance to bending deformation, defined as I=πD for a solid circular cross-section. 4 / 64 calculation, where I is the moment of inertia of the section and D is the diameter of the roller shaft; the two together determine the stiffness characteristics of the roller shaft under stress.

[0024] A state-space equation is established with roller deflection as the state variable and motor torque as the observation variable. The process equation is constructed based on the elasticity model of a simply supported beam, which simplifies the roller to an ideal beam model supported by bearings at both ends and subjected to uniform or concentrated loads in the middle, and is used to calculate the bending deformation of the beam. The observation equation is constructed based on the mapping relationship between torque and shear force, and converts the rotational torque output by the motor into a linear shear force acting on the roller surface through the roller radius and mechanical transmission efficiency.

[0025] According to mechanics of materials, the roller under concentrated load... The flexural deformation under action is equivalent to a linear spring-damped-mass system:

[0026] in, The equivalent concentrated shear force acting on the roller surface, and the equivalent mass. (ρ is density, A is cross-sectional area, L is span), equivalent stiffness Where E is the elastic modulus and I is the moment of inertia of the cross section; equivalent damping Obtained from experimental modal analysis, the damping ratio is typically taken as 0.01. 0.03; For equivalent stiffness, Let these be the roller's mid-span deflection, velocity, and acceleration, respectively. Discretizing this equation yields the state transition equation:

[0027]

[0028]

[0029] in, Let be the state vector at time k, and Δt is the sampling period. Here is the state transition matrix. For the input matrix, The shear force at time k-1 Let be the process noise vector, denoted as uncorrelated, zero-mean Gaussian white noise. The observation equation converts the motor torque into flexural observations: Where r is the roller radius, The observed value at time k is obtained by converting the motor torque. Let k be the mid-span deflection of the roller shaft at time k. The noise used for measurement is zero-mean Gaussian white noise.

[0030] Through recursive calculations of time updates and measurement updates, the estimated elastic deflection of each roller under shear load is output in real time. The time update predicts the prior state value and its covariance matrix at the current moment based on the optimal estimate from the previous moment and the system model; the time update equation is:

[0031]

[0032] in, This is a priori estimation of the state at time k based on information from the previous k-1 steps. This is the optimal posterior estimate at time k-1. To estimate the error covariance matrix a priori, For the posterior estimation of the error covariance matrix, The process noise covariance matrix is ​​set as a constant diagonal matrix based on the power spectral density of the roller vibration acceleration. .

[0033] Measurement update involves calculating the Kalman gain after acquiring new sensor observation data, and then using this gain to weight and fuse the prior estimate with the observed values ​​to obtain the optimal posterior estimate for the current time step. This process is repeated cyclically to achieve real-time output. The measurement update formula is:

[0034] in, Here is the Kalman gain matrix. For the observation matrix, To measure the noise variance, the resolution of the displacement sensor is typically determined, and is usually set to... Magnitude; ,in, This is the optimal posterior estimate at time k. These are actual observed values. For the posterior estimation of the error covariance matrix, It is an identity matrix; the final deflection estimate is... .

[0035] The pure geometric gap component and thermal drift component are separated from the first working gap and the second working gap to obtain the gap state decomposition set. The steps include: The first working gap and the second working gap collected by the non-contact micro-displacement sensor (such as an eddy current sensor or a laser displacement sensor) are recorded as the measured total gap; Subtracting the estimated elastic deflection from the measured total clearance yields the pure geometric clearance component, which is the static clearance portion determined entirely by the mechanical installation position of the rollers after eliminating the effects of dynamic elastic deformation and thermal expansion. This component reflects the initial assembly accuracy of the roller system. This subtraction operation is performed in the controller according to the sampling period. ,in, It is a pure geometric gap component. The original gap value is measured directly by a non-contact sensor, and the low-pass cutoff frequency is set to 10Hz to suppress high-frequency noise.

[0036] Based on the temperature distribution data of each roller, the radial thermal expansion of each roller caused by frictional heat generation and material heat exchange is calculated by the thermal expansion decoupling algorithm, and then converted into a thermal drift component. The thermal expansion decoupling algorithm refers to establishing a mapping matrix between the temperature field and structural deformation, distinguishing and calculating the increase in size along the radial direction of the roller caused by different heat sources, and converting it into the influence on the roller gap, i.e., the thermal drift component. This component represents the slow time-varying characteristics of the gap caused by temperature changes.

[0037] The formula for calculating the thermal drift component is:

[0038] in, Let be the thermal drift component at time t, and N be the total number of roller temperature measurement points involved in the calculation. The thermal-gap coupling weight for the i-th measuring point is determined through finite element thermal-structural coupling simulation or step temperature rise experiment, representing the change in roll gap caused by a unit temperature rise at that point; α is the linear expansion coefficient of the roll material, taken as... for steel rolls... ; Let be the nominal value of the radius of the cross section where the i-th measuring point is located; Let t be the temperature at that point; The cold-state reference temperature is typically taken as the steady-state temperature under constant temperature conditions before the equipment is started. When the roller body temperature distribution is uneven, this summation form can accurately reflect the cumulative effect of axial thermal crown on the clearance.

[0039] By combining the pure geometric gap components, thermal drift components, and elastic deflection estimates, a gap state decomposition set is formed, which is the multidimensional state vector after the total gap is decomposed according to physical causes (geometric, thermal, and mechanical).

[0040] The decoupling calculation formula for each component in the interstitial state decomposition set is expressed as follows:

[0041] In the formula, The gap state decomposition set at time t is the comprehensive state vector of the roll gap at the current time. The observation function of the extended Kalman filter represents the overall input-output mapping relationship of the EKF algorithm; the input includes motor torque. Real-time rotational speed ω, roller elastic modulus E, and cross-sectional moment of inertia I.

[0042] Constructing an inverse model connecting the driving amount of the micro-displacement actuator with the change in the roll gap involves the following steps: Offline frequency sweep excitation was performed on the combined system of micro-displacement actuator and roller system flexibility to obtain the frequency response function from the actuator drive amount to the change of roller gap; the frequency response function is a function that describes the ratio of the output response to the input excitation amplitude and the phase difference at different frequencies, reflecting the dynamic stiffness and damping characteristics of the system.

[0043] The frequency sweep excitation signal uses a logarithmic sweep sine wave ranging from 0.1Hz to 80Hz, with the amplitude limited to within 10% of the actuator's full stroke to avoid impact. The actuator drive voltage and roll gap response are collected, and the frequency response function is calculated using the H1 robust estimation method. ,in This is an estimate of the frequency response function. For the input-output cross power spectral density, The power spectral density is used as the input, and it is averaged multiple times to reduce noise interference. The measured frequency response function often exhibits the characteristics of a type II system with multiple resonance peaks. The low-frequency gain is approximately the transmission coefficient of the actuator reduction ratio relative to the gap.

[0044] The frequency response function is discretized and inverted to obtain the initial inverse model. Discretization transforms the continuous frequency response function into a computer-processable digital difference equation through mathematical transformation. Inversion is the process of taking the reciprocal of the frequency response function, which aims to ensure that the driving quantity output by the controller can offset the dynamic lag and inertia of the system.

[0045] Discretization is achieved through bilinear transformation Complete, the transfer function H(z) is obtained in the z-domain, where s is a complex variable in the continuous-time domain. Let z be the sampling period, and z be a complex variable in the discrete time domain. The unit delay operator represents delaying the signal by one sampling period. Direct inversion often yields non-causal systems; therefore, a zero-phase digital tracking method is used to construct the inverse model as follows: Where d is the system's pure time delay step, determined by the phase slope of the swept frequency data; combined with a low-pass filter F(z) to process high-frequency noise amplification, the final inverse model is: ,in, This is the discrete transfer function of the feedforward inverse model. For frequency sweep The system transfer function is obtained by discretization using bilinear transformation. This is a pure time-delay compensation term that leads by d steps, where d is determined by the phase slope of the swept frequency data; It is a low-pass filter used to suppress high-frequency noise amplification in the inverse model; the cutoff frequency of the low-pass filter is set to twice the closed-loop bandwidth of the actuator.

[0046] The inverse model is configured as a feedforward filter in the digital controller, with its output connected in parallel with the closed-loop feedback correction path of the actuator.

[0047] The difference between the gap state decomposition set and the preset reference gap is used as input. After inverse model calculation, an actuator drive command is generated to directly drive the actuator to perform mechanical domain closed-loop correction of the roll gap. The steps include: The total gap error is obtained by calculating the difference between the sum of all components in the gap state decomposition set and the preset reference gap. The formula is as follows:

[0048] in, This represents the total clearance error. This is the preset reference gap value.

[0049] The total gap error is input into the inverse model for calculation to obtain the feedforward driving amount, which is expressed by the formula:

[0050] in, The basic feedforward driving quantity at sampling time k is the basic feedforward driving quantity. The coefficients of the denominator polynomial in the inverse model are... The coefficients of the numerator polynomial of the inverse model are given by... The numerator and denominator polynomials determine the value, where k is the discrete sampling time. Let be the total gap error at time k.

[0051] The actual displacement feedback value of the micro-displacement actuator is acquired. This actual displacement feedback value is the true position measured by the displacement sensor built into the actuator. The feedback error between this value and the target displacement is calculated, and a feedback correction value is generated through a proportional-integral (PI) controller. The PI controller is a classic feedback control algorithm, and its calculation formula is as follows: ,in For feedback correction output, This is the proportional gain, used for fast response to errors. The integral gain is used to eliminate steady-state residual error. e(t) is the difference between the target displacement and the actual displacement. Through integral action, long-standing small deviations can be completely eliminated.

[0052] The target displacement is determined by the sum of the total clearance error and the reference clearance, i.e. ,in The target displacement of the micro-displacement actuator. The actuator gain coefficient (m / V) For the preset reference gap, Basic feedforward driving quantity; feedback error Updated every Δt, where The proportional gain is the actual displacement of the actuator measured by the built-in displacement sensor. and integral gain The Ziegler-Nichols critical proportional method is used for tuning, and an integral saturation limit is added to prevent actuator saturation.

[0053] The feedforward drive quantity and the feedback correction quantity are summed to obtain the final actuator drive command. This achieves a combination of the speed of feedforward and the robustness of feedback, ensuring that the actuator can respond quickly and position accurately.

[0054] Based on temperature distribution data and real-time torque trends, the gap thermal drift increment and deflection increment for the next control cycle are predicted by thermal expansion prediction model and deflection load prediction model, respectively. The prediction results are then injected into the actuator drive command in a feedforward manner.

[0055] The steps for constructing the thermal expansion prediction model include: Multiple temperature sensors are arranged axially and radially inside the three-roll mill rollers to form a temperature monitoring array. Specifically, temperature sensors are placed at multiple points in key locations such as the bearing positions, the middle section of the roller body, the feeding area, and the discharge area to obtain complete two-dimensional temperature field information. One contact thermocouple is installed at each bearing position. The working section of the roller body is divided into three segments along the axial direction, with four embedded thermocouples evenly distributed circumferentially in each segment, totaling approximately 13 measuring points. Simultaneously, an infrared temperature probe is placed on the non-working area of ​​the outer surface of the roller body as a reference. All temperature signals are acquired at a frequency of 1Hz and stored in a circular buffer.

[0056] Based on data from a temperature monitoring array, a mapping relationship between the temperature distribution of the roller and the radial thermal expansion is established using the lumped parameter thermal network method. The lumped parameter thermal network method discretizes the continuous temperature field into several nodes, each with a concentrated temperature and heat capacity. The nodes are connected by branches representing thermal resistance, forming a network similar to a circuit. The overall thermal deformation is predicted by solving the thermal balance equation of this network.

[0057] The roller body is divided into 5 isothermal sections along the axial direction and two layers of internal core nodes and surface nodes radially, for a total of 10 nodes. The heat balance equation is written for each node i: ,in For node heat capacity, For the equivalent volume of the node, Specific heat; The thermal resistance between the two nodes is determined by the thermal conductivity. The contact thermal resistance is synthesized and identified online by recursive least squares analysis using finite element calibration or experimental data. For the density of the node material, Let i be the temperature of node i. Let j be the temperature of the adjacent node j. The internal heat source of node i comes from frictional heat and material shear heat, estimated based on motor torque, and is expressed as follows: , The internal heat source power generated by friction is represented by μ, the coefficient of friction, which is proportionally distributed to the nodes in the contact area. It is discretized using the backward Euler method, based on the current temperature. Recursively predict the temperature at time τ in the future. Predicted thermal drift:

[0058] in, This represents the gap thermal drift increment predicted with a lead time of τ.

[0059] Based on the mapping relationship, the radial thermal expansion of each roller in the next control cycle is predicted. As the gap thermal drift increment, it directly reflects the increase in roller diameter caused by the increase in temperature in the future, and thus the decrease in gap.

[0060] The steps for constructing the flexural load prediction model include: Based on real-time torque trend data, the material shear load entering the roll gap region in the next control cycle is predicted by analyzing the transmission relationship between material shear force and motor torque. In specific implementation, the torque signal... After applying the moving average filter, its first and second differences are calculated, and linear trend extrapolation is used:

[0061] In the formula, This represents the predicted change in material shear load. Here, η is the steady-state torque-to-force conversion coefficient, and η is the overall transmission efficiency of the reducer and bearings (typically 0.85). 0.95), where r is the roller radius. This is the no-load torque; The dynamic compensation coefficient is calibrated through a step load experiment, and its value range is [not specified]. ; The current torque change rate is predicted, and the prediction step size τ is set to 20~50ms in the controller.

[0062] Using the elasticity equations of a simply supported beam, the expected bending deformation of the roller is calculated based on the shear load of the material; the expected bending deformation is taken as the deflection increment, and the formula is expressed as:

[0063] in, The predicted elastic deflection increment is given by β, which is the load distribution correction factor. When the material is uniformly distributed across the entire width of the roll gap, β = 5 / 8; when the material passes through in a concentrated manner, β = 1, and this value is set by the operating process. L is the center distance between the two bearings of the roll shaft. E and I are the material parameters of the roll shaft. This represents the predicted change in shear load.

[0064] The steps of injecting the prediction results into the actuator drive instructions in a feedforward form include: The gap thermal drift increment and deflection increment are respectively converted into the corresponding equivalent displacement compensation amounts; The equivalent displacement compensation is added to the total clearance error to form a feedforward input with predictive compensation; the resulting composite error signal is:

[0065] in, The signal is fed directly into the inverse model to generate feedforward, which is a feedforward input error with prediction compensation.

[0066] The inverse model is computed based on feedforward inputs with predictive compensation, thereby enabling the generated actuator drive commands to include anticipatory suppression of drift in the next control cycle, as expressed by the formula:

[0067] In the formula, The predicted compensation driving amount injected into the feedforward path at time t. This is the inverse model of the micro-displacement actuator and the roller system flexibility. This is the feedforward gain coefficient, which is usually set to 1. When there is a known deviation in the system model, it is slightly less than 1 to enhance robustness. The coefficient of thermal expansion of the material. The thermal expansion coupling weight is the value for the i-th temperature measurement point. Let τ be the temperature change at the i-th temperature measurement point predicted with a lead time of τ. Where L is the load-deflection conversion factor, L is the roller support span, and I is the moment of inertia of the roller section. This represents the material shear load change predicted with a lead time of τ. All parameters maintain accuracy through offline identification and online adaptive correction. The feedforward compensation for thermal expansion and deflection is calculated during the command update cycle. The superposition calculation is completed internally to ensure that the feedforward signal is injected into the actuator without phase delay.

[0068] In summary, the present invention provides a roller gap error compensation control method for a three-roll mill. This method decouples the measurements from micro-displacement sensors using an extended Kalman filter algorithm, separating the pure geometric gap, thermal drift, and elastic deflection in real time. Based on this, a lumped-parameter thermal network model and material load trend prediction are introduced to calculate the incremental changes in thermally and mechanically induced gaps in advance, generating a predictive feedforward compensation amount. This compensation amount works in conjunction with a feedback correction loop to drive the micro-displacement actuator to dynamically adjust the roller gap. This method upgrades traditional passive feedback compensation to an active closed-loop control of "real-time decoupling - multi-physics modeling - predictive feedforward," effectively overcoming the hysteresis errors introduced by thermal inertia and sudden load changes. It significantly improves the roller gap control accuracy and long-term stability of the three-roll mill under conditions of high-viscosity materials, speed fluctuations, and long-term continuous grinding, achieving consistent gap spacing.

[0069] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for compensating and controlling the roller spacing error of a three-roll mill, characterized in that, include: The first working gap between the rear roller and the middle roller of the three-roll mill and the second working gap between the middle roller and the front roller are obtained. The operating parameters of the drive motors of each roller and the temperature distribution data of each roller are collected. Based on the operating parameters, the elastic deflection of each roller under shear load is estimated in real time by using an extended Kalman filter, and the pure geometric gap component and thermal drift component are separated from the first working gap and the second working gap to obtain the gap state decomposition quantity group. An inverse model is constructed that connects the driving amount of the micro-displacement actuator with the change of the roll gap. The difference between the decomposed set of the gap state and the preset reference gap is used as input. After the inverse model is calculated, the actuator driving command is generated, which directly drives the actuator to perform mechanical domain closed-loop correction of the roll gap. Based on the temperature distribution data and real-time torque trend, the gap thermal drift increment and deflection increment for the next control cycle are predicted by the thermal expansion prediction model and the deflection load prediction model, respectively, and the prediction results are injected into the actuator drive command in a feedforward form.

2. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 1, characterized in that, The steps for real-time estimation of the elastic deflection of each roller under shear load include: The real-time torque, real-time speed, elastic modulus, and moment of inertia of each roller are taken from the operating parameters of each roller drive motor as inputs to the extended Kalman filter. A state-space equation is established with roller deflection as the state variable and motor torque as the observation variable. The process equation is constructed based on the elasticity model of a simply supported beam, and the observation equation is constructed based on the mapping relationship between torque and shear force. Through recursive calculations based on time updates and measurement updates, the estimated value of the elastic deflection of each roller under shear load is output in real time.

3. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 2, characterized in that, The steps for obtaining the set of decomposed interstitial states include: The first and second working gaps collected by the non-contact micro-displacement sensor are recorded as the measured total gap; Subtract the estimated elastic deflection from the measured total gap to obtain the pure geometric gap component; Based on the temperature distribution data of each roller, the radial thermal expansion of each roller caused by frictional heat generation and material heat exchange is calculated by the thermal expansion decoupling algorithm, and then converted into the thermal drift component. The pure geometric gap component, the thermal drift component, and the estimated elastic deflection are combined to form the gap state decomposition set.

4. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 3, characterized in that, The decoupling calculation formula for each component in the interstitial state decomposition set is expressed as follows: In the formula, Let be the set of interstitial state decomposition quantities at time t. It is a pure geometric gap component. For thermal drift components, This refers to the elastic deflection. The observation function of the extended Kalman filter takes into account the motor torque as input. Real-time rotational speed ω, roller shaft elastic modulus E, and moment of inertia I; This is the raw gap value measured directly by a non-contact sensor.

5. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 1, characterized in that, The steps for constructing the inverse model include: Offline frequency sweep excitation was performed on the combined system of micro-displacement actuator and roller system flexibility to obtain the frequency response function from the actuator drive amount to the change of roller gap; The frequency response function is discretized and inverted to obtain the initial inverse model; The inverse model is configured as a feedforward filter in a digital controller, with its output connected in parallel with the closed-loop feedback correction path of the actuator.

6. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 1, characterized in that, The steps for generating actuator driver instructions include: The total gap error is obtained by calculating the difference between the sum of each component in the gap state decomposition set and the preset reference gap. The total gap error is input into the inverse model for calculation to obtain the feedforward driving amount; The actual displacement feedback value of the micro-displacement actuator is collected, the feedback error between it and the target displacement is calculated, and the feedback correction amount is generated through the proportional-integral controller. The feedforward drive quantity and the feedback correction quantity are summed to obtain the final actuator drive command.

7. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 1, characterized in that, The steps for constructing the thermal expansion prediction model include: Multiple temperature sensors are arranged axially and radially inside the rollers of the three-roll mill to form a temperature monitoring array; Based on the data from the temperature monitoring array, a mapping relationship between the roller temperature distribution and radial thermal expansion is established using the lumped parameter thermal network method. Based on the mapping relationship, the radial thermal expansion of each roller in the next control cycle is predicted and used as the gap thermal drift increment.

8. The method for compensating and controlling the roller spacing error of a three-roll mill according to claim 1, characterized in that, The steps for constructing the flexural load prediction model include: Based on real-time torque trend data, the material shear load entering the roll gap area in the next control cycle is predicted by the transmission relationship between material shear force and motor torque. Using the elasticity equation of a simply supported beam, the expected bending deformation of the roller shaft is calculated based on the shear load of the material; the expected bending deformation is then used as the deflection increment.

9. A method for compensating and controlling the roller spacing error of a three-roll mill according to claim 6, characterized in that, The steps of injecting the prediction results into the actuator drive instructions in a feedforward form include: The gap thermal drift increment and the deflection increment are respectively converted into corresponding equivalent displacement compensation amounts; The equivalent displacement compensation is added to the total clearance error to form a feedforward input with predictive compensation. The inverse model is computed based on the feedforward input with predictive compensation, thereby enabling the generated actuator drive instructions to include advance suppression of gap drift in the next control cycle.

10. A method for compensating and controlling the roller spacing error of a three-roll mill according to claim 9, characterized in that, The calculation formula for enabling the inverse model to operate based on the feedforward input with prediction compensation is expressed as follows: In the formula, The predicted compensation driving amount injected into the feedforward path at time t. This is the inverse model of the micro-displacement actuator and the roller system flexibility. This is the feedforward gain coefficient. The coefficient of thermal expansion of the material. The thermal expansion coupling weight is the value for the i-th temperature measurement point. Let τ be the temperature change at the i-th temperature measurement point predicted with a lead time of τ. Where L is the load-deflection conversion factor, L is the roller support span, and I is the moment of inertia of the roller section. τ represents the change in material shear load predicted with a lead time of τ, and N represents the total number of temperature measurement points.

Images