Intelligent thickness control system for calendering high-gloss edge-lit sheeting

Abstract

The application discloses an intelligent thickness control system for calendering high-brightness light edge sealing sheets, and relates to the technical field of intelligent manufacturing and industrial process control, and comprises six steps of multimodal online acquisition, multi-source fusion drift compensation, digital twin real-time synchronization, model prediction closed-loop control, high-speed execution fault self-diagnosis and reinforcement learning adaptive optimization in sequence; thickness data is synchronously acquired through laser, terahertz, ultrasound and environmental quantity, a reliable thickness flow is output by using a drift model and a confidence matrix, a digital twin rolls to predict future thickness to give uncertainty, a secondary planning generates a first control amount, is executed in millisecond level through a distributed clock and is self-healed under shadow driving hot standby and cloud diagnosis, a reinforcement learning agent is online iterated to improve energy saving and stability performance, and full life cycle closed-loop control of thickness precision, low energy consumption and long service life of equipment under complex working conditions is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and industrial process control, in particular to an intelligent thickness control system for calendering high-brightness edge sealing sheet. BACKGROUND

[0002] High-brightness PVC edge sealing sheet produced by high-speed calendering is usually continuously formed by four-roller or five-roller vertical or inclined large-scale calender at a line speed of more than 100 meters per minute, and the workshop environment is kept at a roller temperature of more than 200 degrees Celsius and high humidity for a long time to ensure the melt flowability and surface mirrorization. The existing production line is usually equipped with a laser displacement thickness gauge or a beta / gamma ray gauge at the exit of the roll gap, and the single-path thickness data is sent to the upper computer through an industrial Ethernet, and then the average thickness is maintained through open-loop roll gap or roll bending adjustment. For layer thickness distribution or local depression detection of edge sealing strips, some manufacturers have begun to introduce terahertz or ultrasonic single-point probes for spot-checking scanning, but multi-source synchronous online monitoring has not yet been achieved.

[0003] At the control level, traditional PID and piecewise model predictive control have been used for thickness and flatness adjustment of plastic, rubber and electrode sheet calendering, which can maintain millimeter-level error under stable working conditions. At the same time, the steel strip and aluminum strip industry is exploring the use of reinforcement learning for rolling flatness and thickness control to improve the adaptability to large disturbances. In the field of equipment maintenance, foreign calendering machines have deployed vibration and temperature rise diagnosis packages to predict the failure window of key bearings and servo screws using residual life algorithms. Overall, existing technologies rely on single physical quantity thickness measurement at the sensing end, are based on fixed mechanism models at the control end, and emphasize offline diagnosis at the operation and maintenance end, and there is still a lack of integrated chain from multi-modal sensing to real-time self-learning adjustment and health closed loop.

[0004] The most prominent technical problem at this stage is:

[0005] Millisecond-level stable control of the full-width thickness of mirror-level edge sealing sheet cannot be achieved under the high-temperature, high-humidity and high-speed calendering scene. When the sheet surface glossiness increases, the laser thickness gauge is prone to signal clipping due to spot saturation, the terahertz pulse is attenuated due to water vapor absorption, and the ultrasonic sound path drifts due to temperature-humidity coupling, resulting in instantaneous off-target of single-channel thickness measurement when the working condition changes suddenly; at the same time, the actual opening of the roll gap deviates from the set value due to the thermal expansion of the roller and the fluctuation of the traction tension, and the model predictive control with fixed parameters cannot compensate in time; if the thickness deviation accumulates to the upper limit of the tolerance within a few seconds, it will directly cause the edge sealing strip to fail to fit, the edge to crack or a large number of scrap, causing raw material waste and delivery date violation risks for furniture board manufacturers.

[0006] Due to the lack of multi-source thickness fusion, dynamic confidence evaluation and online self-learning control chain in the existing system, it is difficult to maintain continuous, high-precision and low-energy consumption of thickness closed loop under mirror reflection saturation, high humidity sound speed drift, power grid pressure drop or raw material viscosity jump, and an overall solution is urgently needed to realize real-time accurate thickness adjustment of high-brightness edge sealing sheet under extreme working conditions and long-period safe operation of equipment.

[0007] To this end, the present application provides an intelligent thickness control system for calendering high-brightness edge sealing sheet. SUMMARY

[0008] (I) Technical problems solved

[0009] In view of the deficiencies of the prior art, the present application provides an intelligent thickness control system for calendering high-brightness edge sealing sheet, which synchronously acquires thickness data by laser, terahertz, ultrasound and environmental quantities, outputs reliable thickness flow by drift model and confidence matrix, gives uncertainty by digital twin rolling prediction, generates first-step control quantity by quadratic programming, executes millisecond-level by distributed clock, self-heals under shadow drive hot standby and cloud diagnosis, online iteration improves energy saving and stability performance by reinforcement learning agent, realizes accurate thickness, low energy consumption and long service life of equipment in complex working conditions, and realizes full life cycle closed loop control; The technical problems described in the background art are solved.

[0010] (II) Technical solutions

[0011] To achieve the above purpose, the present application is realized by the following technical solutions:

[0012] The intelligent thickness control system for calendering high-brightness edge sealing sheet comprises,

[0013] A multi-modal acquisition module synchronously acquires a network using multiple physical principles, combines mirror reflection suppression and dynamic range splicing, performs same frequency and same phase acquisition of laser, terahertz and ultrasound data, and synchronously records environmental disturbance signals;

[0014] A drift compensation fusion module models the drift mechanism driven by environmental disturbance, extracts drift vectors and noise distribution of each channel, adaptively adjusts the gain matrix through Bayesian-Kalman fusion algorithm, and outputs a fusion thickness vector with confidence;

[0015] A digital twin synchronization module builds a digital twin model containing roll system elasticity, material rheology and thermal field coupling, synchronizes through Kalman-particle hybridization algorithm and measured data, and realizes millisecond-level prediction of sheet thickness evolution;

[0016] A closed loop execution module generates control instructions that meet the speed, energy consumption and safety constraints of the production line based on the rolling prediction of digital twin, adopts model predictive control algorithm with confidence constraints, and sends the control instructions to the actuator through high-speed scheduling.

[0017] a health assessment module, which establishes a multi-domain health matrix, performs real-time health assessment on the collection, fusion and control link, adopts exponential entropy weight scoring and Markov residual life prediction, and triggers shadow drive hot standby and self-healing strategy;

[0018] a strategy fusion module, which fuses reinforcement learning strategy and model predictive control, optimizes energy consumption and control accuracy under the premise of meeting hard constraints through online fine-tuning and safety monitoring, and executes adaptive intelligent decision-making.

[0019] Further, the multi-source synchronous collection includes synchronous collection of the network by using multiple physical principles, real-time capture of thickness signals on the same space profile and time granularity by a laser displacement sensor, a terahertz pulse probe and an air-coupled ultrasonic probe through upper and lower double-axis reflection, and improvement of signal quality through mirror reflection suppression and control and double-redundancy calibration mechanism.

[0020] Further, the multi-source synchronous collection further includes ensuring millisecond-level phase synchronization of different sensing channels by using atomic clock and optical fiber synchronization technology, realizing time reference unification through master-slave clock structure and least squares deviation correction, and realizing a priori alignment of thickness data and coordinate system through full-amplitude air running self-checking and mirror reflection target calibration.

[0021] Further, the drift compensation fusion includes using the internal correlation between environmental disturbance and historical thickness sequence, constructing a drift prediction function through linear-nonlinear hybrid kernel regression, real-time extracting slow drift vectors and instantaneous noise distribution of each channel, and adaptively adjusting the gain matrix of the Bayesian-Kalman fusion through the confidence matrix.

[0022] Further, the drift compensation fusion further includes triggering an abnormal replacement mechanism through Kalman residual and Mahalanobis distance threshold judgment when a single channel is abnormal, seamlessly replacing the abnormal channel with the predicted value of the digital twin model, and maintaining the continuity and stability of the fused thickness vector.

[0023] Further, the digital twin synchronization includes constructing a digital twin body including three-layer mechanism models of roll system elasticity-geometry, material rheology-stress and thermal field-solidification, performing state assimilation with measured data through Kalman-particle hybrid assimilation algorithm, and real-time ingesting process parameter flow to update model boundary conditions.

[0024] Further, the digital twin synchronization further includes using a four-order explicit Runge-Kutta-Cash-Karp variable step strategy for rolling prediction, outputting a thickness prediction curve and an uncertainty band, and encapsulating and forwarding the prediction data to the model predictive controller through the message bus.

[0025] Further, the closed-loop execution control comprises a rolling prediction sequence using digital twinning, a rolling quadratic programming model predictive control algorithm with confidence constraints, solving the control quantity sequence that minimizes the weighted square sum of thickness deviation, and issuing instructions to the actuator through a high-speed scheduling mechanism.

[0026] Further, the closed-loop execution control further comprises translating the control instructions into servo pulses, roll temperature valve pulse width and traction drive reference frequency within a millisecond network cycle, and ensuring the accuracy and continuity of the execution response through real-time feedback nesting and redundant switching mechanism.

[0027] Further, the health assessment and self-recovery comprises millisecond-level health assessment of the acquisition-fusion-control whole link, construction of the electric-mechanical-environment three-domain health matrix, and real-time discrimination of device and algorithm performance degradation using exponential entropy weight scoring and Markov residual life prediction.

[0028] Further, the health assessment and self-recovery further comprises triggering shadow drive hot standby, cloud operation and maintenance and self-recovery strategy when the health degree is lower than the threshold, and realizing long-period high availability and low downtime risk of the system through soft and hardware mixed redundancy preemption.

[0029] Further, the strategy fusion optimization comprises designing a reinforcement learning agent architecture compatible with the edge sheet calendering scene, pre-training a deep reinforcement learning through digital twinning offline simulation to generate a high-performance initial strategy, and providing a safe and controllable starting point for online fine-tuning.

[0030] Further, the strategy fusion optimization further comprises iterative reinforcement learning strategy in real production in a controlled manner.

[0031] Through adjustable weight gating and model predictive control output fusion, and using safety monitoring and versioned verification to realize robust adaptive intelligent decision-making.

[0032] (Three) beneficial effects

[0033] The application provides an intelligent thickness control system for calendering high-brightness edge sealing sheet, which has the following beneficial effects:

[0034] The thickness signals of three physical principles of laser, terahertz and ultrasonic are cross-mapped with temperature, humidity, vibration and other environmental quantities in the acquisition layer, and clock unification, range splicing and mirror surface suppression are used to make the basic data have the properties of redundancy, traceability and comparability, avoid misjudgment caused by single source distortion, and realize less noise, less drift and less blind area in the perception chain from the source.

[0035] Through the modeling of the drift mechanism, the confidence adaptive weighting and the Kalman Bayesian fusion, the multi-source data is compressed into a high-confidence thickness flow with confidence in the millisecond scale, and the subsequent digital twin only needs a small amount of correction to synchronize the physical production line, greatly reducing the model flux pressure, and forming a positive loop of high-quality input driving high-fidelity models.

[0036] The three-layer mechanism digital twin and the Kalman particle assimilation are real-time mutual feedback, the thickness trend and the uncertainty band are output by rolling prediction, and the optimal control amount is generated by dynamic weight and confidence constraint quadratic programming, so that the planning layer, the execution layer and the health layer make collaborative decisions in the same time domain, and the adjustment overshoot and the steady-state error are significantly compressed.

[0037] Through the distributed synchronous clock, the double-buffer incremental interpolation and the shadow drive hot standby, it is ensured that the roll gap, the roll temperature and the traction speed are started synchronously in the physical layer and are switched in seconds in case of failure, combined with the multi-domain health vector, the remaining life prediction and the remote diagnosis, the equipment maintenance, the quality control and the energy saving and consumption reduction are unified in the same control cycle, forming a safer control. The information flow and the energy flow form a positive feedback gain at each stage, and finally the edge sheet still maintains the thickness precision, the energy consumption economy and the equipment long life under the conditions of high temperature, high humidity and high speed.

[0038] After the multi-dimensional state, the composite reward and the safety projection are introduced into the reinforcement learning agent, the model predictive control is fused according to the health gate weight, the robustness of the deterministic optimal is inherited, and the evolution of the self-learning is obtained; the online fine-tuning, the experience priority playback and the gray verification mechanism ensure that the innovative strategy is continuously optimized under the premise of not damaging the production safety. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is an intelligent thickness control system structure schematic diagram for calendering high-brightness light edge sheet. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0041] Please refer to Figure 1 The present application provides an intelligent thickness control system for calendering high-brightness light edge sheet, comprising,

[0042] In the continuous production line of high-speed calendering high-brightness edge sealing sheet, the surface of the sheet shows mirror-level gloss, and the production workshop is often in a high-temperature and high-humidity environment. In this scenario, a single sensor is prone to problems such as reflection saturation, temperature drift, and insufficient real-time performance, resulting in dramatic fluctuations in thickness data and making it difficult to provide high-credibility support for downstream control systems.

[0043] If real-time and reliable thickness information cannot be accurately and stably captured at the source, all subsequent model synchronization, prediction control, and execution compensation will be amplified due to input errors, ultimately resulting in negative consequences such as uneven sheet thickness, edge sealing failure, material waste, and production downtime.

[0044] Step one, achieve fast, synchronous, and high-density collection of multi-modal thickness and environmental data on the production line, provide high-resolution and disturbance-labeled original information for subsequent modeling, fusion, and control, and ensure millisecond-level phase synchronization of different sensing channels through atomic clock and optical fiber synchronization to avoid data misalignment caused by time drift. At the same time, use mirror reflection suppression control and double-redundancy calibration mechanism to improve signal quality in harsh working conditions.

[0045] Step 101, construction of multi-physical principle synchronous acquisition network

[0046] In the same spatial profile and the same time granularity, real-time capture of laser thickness signal, terahertz thickness signal, ultrasonic thickness signal, and environmental disturbance signal provides cross-source, same-frequency, and same-phase data base for thickness fusion-compensation;

[0047] To suppress mirror saturation, the laser displacement sensor uses upper and lower dual-axis transmission and installs a polarization filter cover. When the normal reflectivity of the sheet surface increases and the receiver current approaches the saturation threshold , the transmission power is adjusted adaptively according to the following formula:

[0048]

[0049] Where: transmission power : current output power of the laser; : transmission power issued by the laser in the previous control period as a reference quantity;

[0050] Power adjustment coefficient : a dimensionless proportional coefficient between zero and one, used to limit the rapid decrease of power; receiving current : real-time photodiode output; saturation threshold : upper limit current obtained according to sensor range calibration.

[0051] By employing a linear proportional voltage drop mechanism, the emission power is rapidly reduced upon detecting a saturation trend, thereby avoiding signal shearing distortion. Therefore, the original laser thickness signal... It maintains a linear response even in mirror-like scenarios, enabling the measurement of the true thickness across the entire surface.

[0052] Based on the complementarity of different sensing principles for materials and operating conditions, terahertz pulse probes and air-coupled ultrasonic probes are spaced apart on the same measurement cross-section. To ensure alignment of the two channels on the time axis, dual PLL clocks are used to synchronously correct trigger delays. Due to the physical differences in propagation speed between ultrasonic ranging and terahertz time-of-flight, an equivalent conversion factor is introduced in real time to make their output thicknesses comparable. :

[0053]

[0054] Where: Conversion factor : Obtained by scanning a standard sample during the self-inspection phase, it is used to eliminate the coupling effect of the difference between the elastic modulus and strain rate of the medium on the sound velocity; The converted ultrasonic thickness sequence is consistent with the terahertz thickness in terms of dimensions, reference coordinate system and value range, and is used for subsequent fusion and compensation.

[0055] Ultrasonic equivalent thickness The thickness sequence, after coefficient correction, has the same dimensions and reference coordinate system as the terahertz thickness. This conversion ensures the uniformity of the two signals in terms of dimensions, coordinate system, and value range, laying the foundation for prior consistency in subsequent Kalman fusion.

[0056] High temperature and humidity fluctuations in the production environment have a significant impact on the performance of various sensors and the physical properties of the sheets. A thermocouple array and a MEMS humidity-vibration composite are arranged around the measuring frame to output the ambient temperature in real time. relative humidity With roller surface vibration acceleration The aforementioned disturbance signal and the original thickness signal are timestamped. Packing into quadruplets :

[0057]

[0058] In the formula: quadruplet vector A six-dimensional feature vector is formed by the combined thickness, three-source data, and three-source environmental data, and bound to a unified time label. Ring network clock drift compensation technology ensures that the sampling error of each dimension is below the microsecond level, thus allowing it to be considered as a slice in the same time frame during multi-source fusion. A real-time mapping between thickness and environmental disturbances is established to provide a lateral correlation basis for subsequent drift compensation, avoiding misjudging operating condition fluctuations as thickness anomalies.

[0059] The thickness fluctuation range of the high-speed production line may exceed the single sensor range. To avoid clipping distortion, both the laser and the terahertz sensor are configured with double ADC links: one with high gain and small range for capturing small fluctuations, and one with low gain and large range for covering extreme over-differences. A range splicing algorithm is used:

[0060]

[0061] Wherein: spliced thickness : thickness value selected after dual-link selection for subsequent calculation; high-gain thickness : high-sensitivity ADC reading; low-gain thickness : wide-range ADC reading; baseline thickness : current batch target thickness calibration value; threshold value : range switching criterion, used to prevent frequent jitter.

[0062] Through dynamic range splicing, the system balances high resolution and wide dynamic range, and can still maintain linearity without distortion during thickness mutation or fluctuation periods. The above mechanism ensures that even if the sheet thickness is temporarily over-difference, the system can still capture the signal completely without saturation cutting, and reserves data integrity for subsequent filtering-compensation.

[0063] Step 101 establishes a multi-source thickness raw data network across physical principles, ranges, and working conditions through four technical features: mirror surface suppression of laser dual-axis arrangement, terahertz-ultrasonic redundancy cooperation, environmental disturbance synchronous recording, and dynamic range expansion, providing time and space consistent, dimension consistent, and error describable data input for the next step of signal drift compensation and weighted fusion.

[0064] The construction of the multi-physical principle acquisition network makes the thickness perception independent of the dependence on a single optical echo: the laser path avoids high-energy beams through adaptive polarization in real time when the mirror surface reflects intense fluctuations; the terahertz channel provides independent ranging capability for different layer interfaces, and the ultrasonic channel compensates for the blind area of electromagnetic wave signal attenuation in extreme high-temperature and high-humidity scenarios. The three-way complementation makes the inevitable defects of a single channel be orderly covered by the remaining channels. Dynamic range splicing ensures that the system still maintains complete data link during the thickness step caused by process speed sudden increase or raw material batch switching, without interruption due to saturation cutting. Environmental disturbance synchronous recording not only provides a basis for subsequent algorithms in drift compensation, but also provides a continuous reference for the simultaneous optimization of heat, humidity, and vibration in the production process, realizing the first connection between the perception link and the process link.

[0065] Step 102, time reference unification and prior calibration link

[0066] On the basis of ensuring that the thickness data from different sources are in the same frequency and phase, we further utilize hardware clock-fiber time reference, full-width no-load self-test and specular reflection target calibration to achieve prior alignment between the thickness data and the coordinate system, so as to provide a measurable original thickness signal for subsequent Kalman filtering-confidence adaptive fusion output.

[0067] It adopts a master-slave clock structure, with the master clock being an atomic time base. The clock is the embedded crystal oscillator in the sensor. The PPS signal is distributed through a ring fiber optic network, and the deviation of each slave clock is calculated using a least-squares method. Correction:

[0068]

[0069] Where: after calibration from the clock : No. The calibration time axis of each sensor; For the first The original measurement time (delay) sequence of frames (sampling points); Reference (main) channel in the 1st Frame time (delay);

[0070] Deviation The minimum offset to be determined is the optimal constant offset relative to the main channel.

[0071] Discrete sampling points : The sampling sequence number within the calibration time window; The variables in the solution process (the offset to be optimized);

[0072] Sample size Total number of data points used for bias estimation.

[0073] By calculating the deviation using least squares, we ensure that the synchronization deviation between the timelines of each sensor and the master clock is less than the microsecond level, thereby ensuring that all signals are in the same time domain during Kalman filtering-dynamic fusion.

[0074] Before starting the machine each day, run the calender empty and place a target sheet of standard thickness inside. The measuring frame is used; simultaneously, automatic scanning from three sources—laser, terahertz, and ultrasound—is activated and measured values ​​are recorded. A one-dimensional polynomial fit is then performed on the scanning trajectory.

[0075]

[0076] in: For the first The polynomial model output of the channel; For channel identification;

[0077] Scan location : raw measurement input, spatial coordinate along the sheet width direction

[0078] polynomial coefficient : the channel the order coefficient, the polynomial coefficient is mapped to the real thickness;

[0079] polynomial order : the highest order determined according to the device resolution, controlling the balance between approximation accuracy and overfitting risk.

[0080] If for any input, , there is , then the zero point is consistent; otherwise, update the zero point compensation table according to the deviation curve .

[0081] In the formula: is the target thickness (target value); the zero point consistency threshold : the upper limit of the static error allowed by the device.

[0082] Eliminate static mechanical assembly errors and sensor zero point offsets through full-width empty running self-test, align each sensor output in spatial coordinates, and provide a mechanical-optical balanced priori basis for subsequent fine filtering and fusion.

[0083] Arrange mirror reflection targets at both ends of the production line, when the sheet head and tail pass through the measurement frame, the laser sensor will scan the target surface with ultra-high reflectivity, thereby triggering fast linearization calibration:

[0084]

[0085] In the formula: linearization coefficient : real-time adjustment of laser gain factor; is the real-time laser thickness value; the initial thickness : the first frame thickness of the sheet entering the measurement interval; the mirror thickness : the theoretical extreme thickness corresponding to the mirror target reflection.

[0086] By using the target reflection as a dynamic slope calibration point, the gain-linearity of the laser signal can be corrected without stopping, ensuring that the laser data is in the linear working area throughout the entire production cycle. Mirror target calibration is an online linearity maintenance method that ensures that the laser thickness remains calibrated consistently without the need for manual calibration over a long period of time, significantly improving the system's online continuous operation capability.

[0087] In order to achieve parameter consistency in subsequent steps, use a unified coordinate identifier The thickness multi-source data and the environment data in the same collection cycle are packaged:

[0088]

[0089] Frame package : complete data frame object;

[0090] Identifier : unique number generated based on the master clock and the space index mixing.

[0091] Through the binding of the unique identifier, the same identifier is used when all subsequent steps refer to the thickness data, and there is no ambiguity of the same object with multiple names or multiple objects with the same name, providing strong consistency for the vertical running of the process.

[0092] Step 102 takes optical fiber synchronization-inlaid clock calibration, empty running self-check-zero point fitting, mirror target-dynamic linearization, and cross-source coordinate unique identification as the core, and completes the four-dimensional alignment of multi-source thickness signals in time, space, gain, and naming system. Through this series of processing, the thickness original signal is transformed into error measurable and frame level unique basic data, providing a completely isomorphic input matrix for the Kalman fusion-drift compensation of step two.

[0093] The design of the time reference unification and the prior calibration link realizes the dual solidification of the measurement time domain and the space domain. The optical fiber-inlaid clock scheme avoids the common time drift in the high electromagnetic noise environment of the production workshop, so that the multi-channel data frame has a natural consistent time label; the empty running self-check completes the static space curve fitting and thermal expansion compensation in advance, so that the thickness measurement has excluded most of the mechanical assembly errors before the real production; the mirror target provides a dynamic linearization path that is continuously connected to production, so that the laser gain is always in a controllable interval from start to stop. The cross-source coordinate unique identification establishes an unambiguous primary key for all thickness-environment-time multi-dimensional data, so that the data fusion, twin modeling, and predictive control do not occur misplacement when calling data objects.

[0094] As the first step of the intelligent thickness control method, multi-modal online collection realizes the reliable landing of thickness data in high light, high temperature and humidity, and high speed rolling scenes through steps 101 and 102. The former focuses on the sensing layer and simultaneously captures thickness from laser, terahertz and ultrasonic channels and records environmental disturbances, ensuring the space-time co-frame of multi-source information; the latter shifts the focus to clock alignment, zero calibration and gain linearization, so that each signal is completely consistent in sampling dimension and coordinate dimension.

[0095] The rolling production site of high-brightness light edge sealing sheet continuously outputs hundreds of frames The multi-source thickness raw data stream includes laser thickness signals that are dynamically attenuated by specular reflection, as well as terahertz pulse penetration echoes, air-coupled ultrasonic time delay calculations, and six-dimensional environmental disturbance vectors. Although step one ensures that all signals are consistent across time and space axes, the instantaneous errors inherent in a single frame and the slow drift in multi-frame sequences continue to erode the reliability of thickness measurements due to the long-term cumulative effects of sensor element aging, thermal drift, humidification nonlinearity, and vibration coupling noise. If structured, traceable multi-source fusion and drift compensation are not implemented immediately after acquisition, the thickness data will be amplified by asynchronous noise and drift terms, leading to distortion of the digital twin model input, misjudgment of operating conditions by the model predictive controller, and even reverse adjustment.

[0096] Step 2: By modeling the drift mechanism and using a Bayesian-Kalman fusion algorithm, the multi-source thickness measurement flow is compressed into a single, high-confidence state vector with confidence. This step estimates the drift trend and measurement noise in real time and adaptively adjusts the gain matrix. Even in the event of single-channel failure or noise bursts, it can still output stable and quantifiable uncertainty limits, providing reliable predictive input for digital twins and back-end control.

[0097] Step 201: Thickness drift mechanism modeling and adaptive confidence measurement

[0098] By leveraging the intrinsic correlation between environmental disturbances and historical thickness sequences, we extract the slow-varying drift vectors and instantaneous noise distributions of each channel, and calculate the confidence matrix in real time to provide quantization weights for subsequent fusion.

[0099] Laser thickness signals are particularly sensitive to thermal drift, while terahertz signals are more susceptible to absorption and attenuation due to humidity; ultrasonic channels experience path shifts due to the temperature-viscosity coupling of the material. First, in the scrolling window... Internal convergence , , A drift prediction function is constructed using three thickness signals and a hybrid linear-nonlinear kernel regression. For each thickness signal Represented as instantaneous true thickness With drift item The sum, therefore, can be minimized by:

[0100]

[0101] In the formula: the drift vector is obtained by fitting. And it is updated in real time.

[0102] For the first Real-time frame thickness; ambient temperature relative humidity Roller surface vibration acceleration ;

[0103] Drift vector Description of the first Slowly varying errors of road sensors under environmental disturbances;

[0104] Regression function A nonlinear kernel model that maps three-dimensional environmental disturbances to thickness drift space;

[0105] Window Scale : Matches the time it takes for the sheet banner to pass through the sampling rack once.

[0106] By using drift estimation driven by environmental variables, the error trend of each channel as the operating conditions change is obtained, providing a quantitative basis for subsequent weight reduction.

[0107] The instantaneous residual obtained after removing the drift term Including high-frequency noise and random jump points, to characterize its statistical distribution, a window is used... Upward sliding estimation of residual covariance matrix For this matrix, a Shannon entropy threshold is introduced to determine the abundance of information content, thereby dynamically adjusting the upper limit of noise variance to adapt to changes in sampling density caused by changes in production cycle time.

[0108] Real-time updates to the residual noise covariance matrix allow the Kalman filter to have a noise prior synchronized with the current operating conditions, preventing the filter gain from being weakened by outdated statistics.

[0109] The drift vector and noise covariance are jointly projected onto the confidence space to generate a diagonal confidence matrix. :

[0110]

[0111] Where: confidence matrix It comprehensively characterizes the confidence level of each channel at the current moment. The confidence level automatically decreases when the drift is greater or the noise is stronger. , that is , , , No. Dynamic compensation amount / deviation of the channel;

[0112] The environmental disturbance / measurement noise covariance matrix characterizes the statistical properties of three-dimensional disturbances (or three-sensor errors). This is its precision matrix, used for whitening or unifying dimensions.

[0113] Confidence matrix To provide adaptive weights for subsequent Bayesian-Kalman fusion, slowly varying drift and instantaneous noise are included in a unified evaluation framework. Through environmental coupling drift factor extraction, instantaneous noise covariance estimation and confidence matrix normalization, the error source identification-quantification-weight mapping closed loop is completed, providing an updateable and iterative prior information flow for the multi-source thickness fusion of step 202.

[0114] By closing the three links of environmental coupling drift, instantaneous noise statistics and confidence quantification in the same rolling window, the system forms a layered capture capability for slow varying errors and fast varying noises. The drift model introduces the cross-check of time derivative and prediction residual, so that the system not only corrects errors, but also gives an early warning in the error acceleration stage; noise covariance adaptive update gives the filter flexibility to follow the sudden changes in working conditions; the confidence matrix maps drift and noise to a unified evaluation scale, so that the weight distribution is theoretically interpretable.

[0115] Step 202, Bayesian-Kalman dynamic fusion and abnormal replacement

[0116] With the online update of Bayesian-Kalman gain driven by the confidence matrix, the three thickness signals are fused into a high-confidence thickness vector, and the model prediction value is seamlessly enabled to replace the single-channel abnormality, ensuring the stability and continuity of the thickness data flow.

[0117] At the starting time of the rolling window, the state of the last time of the digital twin is called and the confidence matrix , the thickness prior distribution is generated by the Bayesian formula . This distribution is used as the input of the prediction step of the Kalman filter, so that the filter has a shape-accurate prior before observing new data. The prior distribution is modulated by the confidence matrix, and the channel with high confidence occupies a larger weight in the prior, achieving good and bad source differentiation in prediction.

[0118] After obtaining the observation vector (measurement (observation) thickness vector, corresponding to the real-time fusion before output of the laser, terahertz and ultrasonic three channels), the Kalman gain calculation uses the improved formula:

[0119]

[0120] Kalman gain : the matrix that balances the prediction covariance and the observation noise covariance , the fusion weight matrix, the proportional coefficient matrix within the range ; prediction covariance : generated by the prior distribution, indicating the prediction error intensity, positive definite matrix, interval for positive real number domain; observation noise covariance : directly using step 201 real-time updated , representing the observation noise intensity, positive definite matrix;

[0121] By introducing the confidence matrix into the gain formula, the system dynamically suppresses the channels with large drift or noise, ensuring that the fusion result is more dependent on high-confidence sources.

[0122] Update the thickness estimate using Kalman gain:

[0123]

[0124] In the formula: fused thickness : final output thickness vector;

[0125] , prior predicted thickness (predicted from the previous state by physical / data model to the current time);

[0126] Kalman residual : observation and fusion difference;

[0127] If the Kalman residual If a three times squared Mahalanobis distance jump occurs in any channel, mark the channel as abnormal and remove the corresponding row and column from the confidence matrix and the observation noise covariance , and fill in the gap with the predicted , until the abnormal signal returns to within the drift-noise threshold. The abnormal replacement strategy ensures that any single-source failure will not interrupt the thickness data stream, and the fused thickness remains smooth and continuous. The Mahalanobis distance threshold is three times the square value, which is the abnormal decision criterion.

[0128] Output thickness vector frame packaging and database landing:

[0129] The final fused thickness and the updated balanced prediction covariance are packaged into the corresponding data packet , and written into the special timing database along with the residual statistics. This database provides an API interface for the digital twin step to call, realizing cross-step parameter penetration.

[0130] The thickness fusion result and quality metric are synchronized and persisted, providing direct key-value queries for the twin synchronization algorithm in step three, improving the overall throughput of the system.

[0131] Bayesian-Kalman dynamic fusion uses the confidence matrix to map multiple physical source thickness signals to a high-confidence thickness vector. The abnormal replacement mechanism ensures that any single-source failure can be completed by the remaining channels and prediction, and the fused thickness meets the digital twin model requirements in terms of timing and quality, realizing real-time closed loop between the perception layer and the model layer.

[0132] Step two successfully realizes the real-time conversion from multi-source thickness raw signals to high-confidence thickness vectors by modeling the drift mechanism + confidence quantification and Bayesian-Kalman dynamic fusion + two-level structure of abnormal replacement.

[0133] The confidence matrix output in step 201 and the residual covariance provide accurate priors for the Kalman gain iteration in step 202, enabling the fusion process to adaptively allocate weights in fluctuating operating conditions; the thickness vector completed in step 202 encapsulates Not only does it meet the dual requirements of real-time and continuity, but it also synchronously writes the covariance and residual into the database, providing complete state correction basis for digital twins.

[0134] The cleverness of Bayesian-Kalman dynamic fusion lies in injecting real-time confidence into the Kalman gain, while the abnormal replacement strategy ensures that any single-source mutation does not break the data chain. In this framework, the prediction covariance, observation noise covariance, and confidence matrix form a prediction-observation-quality three-element closed loop, and the system adjusts the trust level in real time according to the quality of the signal, avoiding good signals being drowned or bad signals being amplified. Multi-frame confirmation of abnormal processing not only highlights the tolerance of instantaneous spikes in high-speed scenarios, but also takes into account the stability and safety of persistent faults. The positive definite check of covariance before disk writing eliminates the risk of mathematical anomalies to the twin model, controlling data integrity to the secondary verification level.

[0135] In the production process of calendering high-gloss edge sealing sheet, the digital twin model not only needs to be similar to the real production line, but also must be synchronized with the production line in milliseconds on the time scale, otherwise the future thickness trajectory obtained by the model predictive controller will be out of phase with the physical process, directly leading to inaccurate timing of roll gap adjustment. The core task of step three is to make the virtual model continuously fit the physical world under dynamic environment and equipment aging conditions: on the one hand, the fusion thickness vector and covariance output by step two are assimilated to the model state; on the other hand, real-time process parameters are ingested, including roll temperature sequence, traction tension sequence, melt viscosity curve, etc., which are used as driving boundary condition updates to the model kernel, forming a virtual-real interactive closed loop.

[0136] Step three, build a digital twin that is real-time synchronized with the production line, through Kalman-particle hybrid assimilation with measured data, realize millisecond-level prediction of plate thickness evolution and equipment state through three-layer mechanism model of roll system elasticity-geometry, material rheology-stress, thermal field-curing. The model boundary conditions are refreshed with new data, so that the virtual production line still maintains physical consistency under extreme operating conditions, providing dynamic constraints and priors for MPC optimization.

[0137] Step 301, physical-data assimilation framework construction

[0138] A multi-field digital twin model is constructed, which contains the roll-melt coupling mechanism, the heat-flow-stress multi-field mutual feedback and the real-time data interface. The state vector initialization and parameter implantation are realized by the Kalman-particle hybrid assimilation algorithm.

[0139] The digital twin model adopts a three-layer nested structure: the outermost layer is the roll system elastic-geometric sub-model, which describes the force-deformation-gap evolution of the upper and lower rolls; the middle layer is the material rheological-stress sub-model, which describes the viscoelastic-viscosity coupling flow of the melt in the roll gap; the innermost layer is the thermal field-solidification sub-model, which simulates the process of polymer from thermal state to solid state. The three layers share the core variables, including roll gap displacement, melt shear rate, polymer specific heat, etc., so that the heat-flow-force information is synchronized in time domain. This structure ensures scalability while maintaining details. If cooling air curtain or roll texture friction is added later, it can be mounted as an attached physical block without destroying the overall coupling framework.

[0140] The three-layer physical variables are compressed into the state vector , and then mapped to the measurement domain through the observation projection matrix to obtain the thickness prediction:

[0141]

[0142] The observation projection matrix : the matrix that linearly maps multi-field variables to the thickness space, the dimension is determined by the number of mechanism layer variables and the dimension of thickness observation. In the formula: the state vector : contains roll gap displacement, polymer temperature field average, melt density, etc., the value changes continuously with time; the predicted thickness : the thickness estimate output by the digital twin model, the dimension is consistent with the fusion thickness vector .

[0143] To consider the uncertainty of linear high-dimensional state and nonlinear parameters, the system uses the Kalman-particle hybrid assimilation algorithm: the state vector is updated using the unscented Kalman filter, and the particle filter weight resampling is used for material parameters that are difficult to linearize (such as yield modulus, temperature sensitivity). The state assimilation update formula:

[0144]

[0145] In the formula: the predicted state : the prior state obtained after the model time is advanced, which is derived from the predicted state of by the state equation / physical model. Typically contains: thickness field state quantity, material / device parameter drift term, environmental compensation coefficient, etc.

[0146] ​​corrected state : assimilated state, final state after fusing current observation, used for output and next step prediction;

[0147] unscented kalman gain : update weight calculated from state-observation covariance;

[0148] particle set: updated by residual weight resampling. represents the particle material parameter vector.

[0149] is the weighted / robust processed observation thickness vector of three channels (laser, terahertz, ultrasound), or the measurement output of the kalman observation model in the last section; is the predicted measurement (prior measurement) obtained by state prediction;

[0150] Thus, linear and nonlinear variables are separated and high-dimensional state converges quickly without sacrificing nonlinear parameter adaptation.

[0151] Roll temperature sequence , traction tension sequence , melt viscosity curve Real-time push to twin model, update boundary vector . Immediately after assimilation, refresh the boundary vector , and use the corrected state and the boundary vector as input for the next prediction step, forming an external driving-internal state two-way closed loop. Process parameter embedding maps real operation to the model, keeping the prediction consistent with the actual working condition, and avoiding false response due to boundary condition lag.

[0152] Through three-layer mechanism nesting, state projection matrix, kalman-particle hybrid assimilation and real-time refresh of boundary conditions, the digital twin model obtains the dynamic heartbeat with the same frequency as the physical production line, laying a solid data-model integrated foundation for the next step of prediction-uncertainty quantification-output broadcast.

[0153] Through the deep coupling of three-layer mechanism sub-model and data assimilation, the digital twin model breaks away from the traditional offline modeling and online correction mode, and evolves online in real time. The kalman-particle hybrid strategy separates linear high-dimensional state and nonlinear material parameters, so that computing resources are allocated according to weight, so as to realize accurate tracking of complex coupled processes within the hard real-time computing budget. At the same time, the embedding of process parameters makes the model respond immediately to roll temperature, tension and raw material batch disturbance, and no longer appears the situation that the model gradually deviates from the physical production line.

[0154] Step 302, Rolling Prediction-Uncertainty Quantification-Data Broadcasting

[0155] A multi-step prediction is made in the rolling time domain using the assimilated state, and a thickness prediction curve and an uncertainty band are outputted, which are encapsulated and forwarded to the model predictive controller.

[0156] The model time advancement adopts a fourth-order explicit Runge-Kutta-Cash-Karp variable step strategy, dynamically adjusts the step size according to the covariance estimation at the previous time, shrinks the step size when the state disturbance accelerates to prevent numerical explosion, and relaxes the step size at the stable stage to improve the calculation throughput and keep pace with the production rhythm.

[0157] The prediction step size The thickness prediction and the extrapolated covariance are outputted at each step. The covariance extrapolation uses the Lyapunov equation to ensure positive definiteness, and if the extrapolated value shows a divergence trend, step 301 is triggered to re-assimilate, forming a self-converging closed loop.

[0158] The thickness prediction and the extrapolated covariance are mapped into upper and lower confidence bounds in the credibility channel space . If the confidence bounds touch the edge of the thickness tolerance band, an early warning flag is attached before data broadcasting, so that the model predictive controller can tighten the optimization constraints in advance according to the risk.

[0159] Finally, the prediction thickness sequence, uncertainty band, and warning flag are packaged into an uncertainty matrix and pushed to the real-time message bus. The interface description follows the same data mode as step two, ensuring that the controller on the subscription side can be called without secondary parsing, completing the parameter cross-mapping. Rolling prediction-uncertainty quantification-data broadcasting converts the achievements of the assimilated digital twin model into a thickness-confidence-warning triplet that can directly drive the model predictive controller, realizing seamless transfer of the virtual model to the decision logic.

[0160] Through the physical-data assimilation framework of step 301, the state of the digital twin model is successfully aligned with the actual working condition in real time, and then in step 302, the model cognition is converted into decision-making information through rolling prediction and uncertainty quantification. The former is responsible for alignment, and the latter is responsible for forward-looking, both of which provide the model predictive closed-loop controller in step four with a high-fidelity, high-real-time, and risk-marked thickness prediction sequence. The injection of the digital twin model's self-learning, self-converging, and self-warning capabilities marks the transition of the intelligent thickness control system from passive correction to active prevention.

[0161] The rolling prediction module not only provides future thickness values, but also gives covariance bands and warning signs simultaneously, visualizing and quantifying the uncertainty of future, for the controller to choose a robust strategy; variable step-size integration and covariance positive definiteness guardianship guarantee the numerical stability of the model in long-time running, avoiding the distortion of traditional fixed step-size methods when encountering high gradient areas. The dual-topic message broadcasting framework enables the control system to simultaneously obtain the current thickness and future trend at the same interface, reducing the system integration complexity.

[0162] The tolerance of high-gloss edge sealing sheet to thickness tolerance is extremely narrow, and any millisecond-level control lag may push the sheet to the edge of the out-of-tolerance. Step three has output a rolling thickness prediction sequence with uncertainty, a warning sign and a frame-level consistent process parameter flow, providing a future-oriented process condition preview for the decision layer. The mission of step four is to: integrate this prediction information with the physical limits of the equipment, energy consumption targets and product quality weights into an integrated optimization framework, and calculate the roll gap displacement, roll temperature adjustment and traction speed three types of execution instructions in real time; then use high-speed scheduling mechanism to accurately broadcast the instructions to each execution mechanism within milliseconds, realizing thickness feedforward-feedback coupled adjustment. The entire closed loop needs to remain robust in the environment where the production line runs at high speed, uncertainty changes dynamically and hardware constraints coexist, ensuring that the control output will not produce sharp overshoot due to model error or delay, and leaving enough margin for energy consumption and mechanical life in multi-objective trade-off.

[0163] Step four, after getting predictable thickness and uncertainty information, use the rolling quadratic programming model predictive control (MPC) algorithm with confidence constraints to generate optimal control instructions that meet the line speed, energy consumption and safety constraints in real time, and through instruction quantization-pulse mapping and IEEE-1588 nanosecond-level network synchronization, send the instructions to the driving execution mechanism within milliseconds.

[0164] Step 401, solving the rolling time domain optimal control quantity

[0165] Using the digital twin rolling prediction sequence and uncertainty information, solve the control quantity sequence that minimizes the weighted sum of thickness deviation under given physical and process constraints, and output the first-step control vector of the current period.

[0166] The future provided by digital twin Step thickness prediction sequence and target thickness constitute the deviation vector. The system defines the rolling cost function:

[0167]

[0168] In the formula: the objective function : the total cost in the rolling window; is the thickness prediction value of the future step.

[0169] weight coefficient : thickness accuracy weight, value range ;

[0170] control input column vector : contains roll gap displacement increment , roll temperature adjustment , pulling speed increment ;

[0171] input weight matrix : diagonal positive definite matrix, parameters calibrated by energy consumption and mechanical wear index

[0172] weight coefficient : energy consumption weight, value range , control input column vector

[0173] The thickness deviation square term ensures that the control strategy focuses on size accuracy; the input weighting term suppresses drastic adjustment, controls energy consumption and equipment impact. The prediction covariance band provides the thickness future fluctuation range, to avoid optimal solution falling in high-risk area, introduces confidence constraint, let the optimization solution search near the center of the prediction band, reduce the risk of out-of-bound:

[0174]

[0175] In the formula: thickness safety margin : loose tolerance margin defined according to process standard is the minimum predicted thickness allowed in the step is the maximum predicted thickness allowed in the step; confidence parameter : value , the smaller the more conservative

[0176] Roll gap displacement, roll temperature and pulling speed are limited by mechanical limit and thermal inertia, to ensure that the optimization will not output the instructions that the equipment cannot execute or cause material instability:

[0177]

[0178] In the formula: roll gap lower limit and roll gap upper limit converted from roll contact safety distance is the current roll gap displacement is the roll gap increment in the step

[0179] Roll temperature upper and lower limits , Determined by the polymer thermal degradation threshold and the temperature resistance limit of the roller material; This is the current roller temperature;

[0180] For the first Step roller temperature increment;

[0181] Speed ​​upper and lower limits , Matching traction drive capability with material cooling cycle, The current traction speed, For the first Step speed increment, For the prediction step index;

[0182] Using a sequential quadratic planarizer with predictive covariance linearization, in hard real-time budgeting The optimal control vector sequence is obtained internally. Only the first step is captured. The data is sent to the scheduling layer, and the remaining steps are used as a warm backup trajectory cache. If the solution is not completed in the next cycle, the cached trajectory can be used temporarily to ensure uninterrupted control. The rolling time-domain optimal control quantity solution utilizes multi-objective quadratic programming with confidence constraints to provide the lower-level scheduler with a risk-assessed first-step control vector. The first-step-warm backup dual-caching design ensures that the solution delay will not lead to a control gap.

[0183] By introducing a dynamic energy-accuracy allocator, the objective function automatically shifts its focus during periods of fluctuating and stable operating conditions, ensuring neither the sacrifice of thickness quality nor unlimited high-energy-consumption operation. Simultaneously, by linking the principal eigenvalue with a safety margin, the optimizer can preemptively reduce the amount of operations when the predicted mean is safe but volatility is high, avoiding the blind spot of traditional confidence region concepts that ignore fluctuation trends. Dynamic physical constraints allow the safety boundary to slowly adjust with the equipment's health status and ambient temperature, eliminating the need for manual hard coding. FPGA coprocessor pipeline budgets significantly reduce computation time, providing algorithmic support for millisecond-level scheduling.

[0184] Step 402: High-speed instruction scheduling and real-time feedback nesting

[0185] Within a millisecond-level network cycle, the initial control vector is translated into servo pulses, roller temperature valve pulse widths, and traction drive reference frequencies, and the execution response is monitored in real time, with feedback embedded into the next cycle optimization.

[0186] Control vector element roll gap increment Roller temperature increment Speed ​​increment The calibration curve of the equipment is converted into an executable pulse: the roll gap displacement quantization table converts the roll gap increment. Mapped to the number of stepper motor pulses with an acceleration ramp; the roller temperature valve uses PWM to modulate the roller temperature increment. Convert to duty cycle; traction speed increment Convert directly to VFD frequency increment. All quantization curves are linearly interpolated to ensure high-resolution output.

[0187] To ensure that the three-way action is synchronized to reach the effect layer, a zero-point alignment pulse is distributed using the IEEE 1588-based distributed time protocol on the real-time industrial Ethernet. After static calibration of the network, the clock deviation is controlled at the microsecond level, so that the roll gap adjustment, roll temperature regulation, and traction speed variable frequency are all started at the same physical time, avoiding control cross-coupling-induced thickness secondary fluctuations.

[0188] During the execution of the actuator, the grating ruler reads the roll gap displacement in real time, the infrared probe reads the roll temperature in real time, and the new magneto-electric encoder reads the traction speed. These feedbacks are returned to the scheduling layer within half a control period. The remaining time period that has not been executed in the current period is calculated using an interpolation algorithm to continue sending incremental instructions, forming an incremental-correction microcycle, and ensuring that the deviation between the expected value and the execution value seen by the controller is compressed within one adjustment period.

[0189] If the read value deviates from the target by more than the threshold within two periods, the system quickly locates the fault according to the failure mode diagnosis table: roll gap channel deviation triggers hydraulic redundancy reduction, roll temperature channel deviation triggers parallel electric heating block, and traction speed channel deviation triggers standby servo driver. The redundant devices have the same instruction interface, and the switching process is completed within the network cycle level, avoiding thickness control interruption.

[0190] High-speed instruction scheduling uses three safety measures: synchronized clock, incremental interpolation, and redundant switching, to convert the optimal control vector into stable, traceable, and uninterrupted physical action, and write the response value back to the next optimization period, forming an optimization-execution-feedback trinity closed loop.

[0191] The instruction-effect mapping sampling uses cubic spline interpolation to eliminate linear splicing jitter and improve the smoothness of high-precision roll gap and roll temperature regulation. Double-layer synchronous grid locking solves the old difficult problem of network synchronization but asynchronous endpoints, eliminating weak links for high-speed production. The double-loop incremental interpolation strategy divides the passband into two levels of servo and main control, allowing local devices to be transient and stable, and giving the optimizer enough time to analyze trends, avoiding mutual interference between servo response and main control sluggishness in reality. Bayesian networks help diagnose complex composite faults, allowing the redundant switching to select the most targeted backup path.

[0192] ​Step four uses the rolling horizon optimal control amount to solve the high-speed command scheduling and real-time feedback nested two-level mechanism, which maps the digital twin prediction to physical action, realizes the last hop from data to execution. Through the confidence constrained quadratic programming, the system takes the digital twin uncertainty into the control decision, avoiding overlooking the risk; through the synchronous clock and incremental interpolation, the system converts the calculation results into stable output on the millisecond scale, filling the model-execution gap; through failure diagnosis and redundancy switching, the system takes potential execution layer faults into closed-loop control, extending the intelligent tentacle to the edge of hardware.

[0193] High-speed calendering production puts forward strict requirements on the thickness control execution chain: on the one hand, the first control vector has been optimized in step four, but if any link in the servo, hydraulic, temperature control or frequency conversion chain is delayed, drifted or stuck, the thickness error will be immediately amplified; on the other hand, long-term operation in high temperature and humidity will accelerate the aging of the actuator, and if there is no real-time health assessment and self-healing closed loop, the gradually accumulated micro-faults will eventually evolve into a sudden shutdown.

[0194] Step five, millisecond-level health assessment is performed on the collection-fusion-control whole chain, a health matrix of the electrical-mechanical-environmental three domains is constructed, and exponential entropy weight scoring and Markov residual life prediction are used to real-time judge the performance degradation of devices and algorithms; if the health degree is lower than the threshold, the shadow drive hot standby, cloud operation and maintenance and self-healing strategies are triggered, realizing the mixed redundancy occupation of software and hardware, and ensuring the system to maintain high availability and low downtime risk in long-term operation.

[0195] Step 501, real-time health assessment and early warning identification

[0196] In each control cycle, the health index vector of the key nodes of the execution chain is extracted, the drift trend and dynamic reliability are quantified, and the decision basis for fault prediction and redundancy switching is provided.

[0197] The execution chain is divided into four nodes: roll gap servo, electro-hydraulic pressing, roll temperature loop and traction frequency conversion, and three-domain signals are collected for each node: electrical domain signals (current waveform, power factor), mechanical domain signals (displacement residual, speed ripple), and ring domain signals, (oil temperature, bearing vibration), through the normalized mapping function :

[0198]

[0199] Form a node health vector , and then splice to form a whole chain health matrix , the health matrix provides a real-time health quantification basis for the whole chain in a unified scale:

[0200]

[0201] Normalized mapping function , the first Dimension compression and dimension-unification linear-nonlinear hybrid mapping of node three-domain signals; node health vector , The value is 1 to 4, including power distortion factor, motor temperature rise rate, residual mean square root and other elements.

[0202] In the sliding window The first-order difference is performed on the full-chain health matrix to calculate the drift matrix , and then the exponential entropy weight method is used to calculate the node health score of each row and column of the drift matrix :

[0203]

[0204] Node health score : Output interval , the value closer to 1 indicates higher health;

[0205] Exponential decay factor : According to the long-term reliability index, take a positive real number;

[0206] Weight matrix : Equal to the entropy weight vector Diagonalization and then left multiplication of the inverse of the node Mahalanobis covariance matrix, to ensure double weighting of high information index and covariance compression dimension;

[0207] First-order drift vector : First-order difference of node health vector; second-order drift vector : First-order difference of , used to capture drift acceleration; second-order weighting coefficient : The value is , which is adaptively increased with production speed to strengthen the early sensitivity to high-speed misalignment; two-norm : Euclidean length after matrix multiplication, used to quantify the overall drift energy.

[0208] By assigning higher weights to drift indicators with high information content, the score is sensitive to real failure precursors but not excessively responsive to noise.

[0209] Node health score sequence is discretized into several health level states to construct a Markov transition matrix The Chapman-Kolmogorov equation is used to predict the first arrival probability of the fault absorbing state, and the residual life of the node is calculated When the health score is lower than the preset threshold, the system generates a preventive maintenance work order in the cloud.

[0210] health score synchronize the input weight matrix of step four Let the high health node bear more control amplitude, and the low health node automatically reduce the load to prolong the total chain life cycle.

[0211] Through multi-domain health vector, exponential entropy weight drift score, Markov residual life, and health-control collaborative weighting, the system diagnoses the execution chain life trend in milliseconds, and embeds the health information closed loop into the optimizer.

[0212] The real-time health assessment system is based on multi-domain signal fusion, making the previous black box executor become a transparent white box. The second-order drift and entropy weight amplification strategy makes the system extremely sensitive to early micro-defects; the Markov RUL mechanism gives the remaining available time on the probability level, so that the decision no longer depends on the head. Health-control collaborative weighting decouples the operation and maintenance information from production optimization, realizing the dynamic symbiosis of equipment and process.

[0213] Step 502, redundancy switching and remote operation and maintenance collaboration

[0214] When the node health score drops sharply or the actual response continues to be inaccurate, the same type of redundant device is triggered in milliseconds and the operation and maintenance data is uploaded to the cloud and remote diagnosis is completed in the background.

[0215] The redundant executor is in shadow driving mode in normal times: the driver receives the same pulse, PWM or frequency instructions as the main channel, but the output is optically coupled and does not act on the mechanical end, only for health monitoring. In this way, the shadow executor is always in synchronization with the main channel, and only the soft relay switching output path is needed to complete seamless takeover.

[0216] The switching threshold uses the health score and the response residual Double criteria: when the health score drops below the soft threshold and the response residual exceeds the residual band for three consecutive periods, trigger the shadow drive to preempt, the soft threshold, that is, the preset buffer ensures early replacement to avoid hard failure, and the residual band avoids false positives.

[0217] where, is the real-time observation value of the th index / channel (such as thickness, temperature, power, etc.);

[0218] is the real-time observation value of the The instruction / target / reference value corresponding to the index; the meaning of residual band is the safety interval that allows the residual to float around zero, and beyond it is considered abnormal or requires strong compensation;

[0219] After preemption, the local controller uploads the latest ten-thousand-frame health vector slice, drift matrix segment, and event log of the fault node to the cloud operation platform; the platform calls the deep residual network and case library for comparison to generate a diagnosis report, which is pushed to the mobile terminal. The maintenance personnel will resolve tasks according to the report in the next shift.

[0220] If the health of the secondary node in the same chain is rapidly deteriorating after shadow drive preemption, enter the safe shutdown state: reduce the traction speed to the technical lower limit, increase the roll temperature safety margin, and continuously output thick tolerance products until manual intervention is confirmed to ensure that the main equipment is not damaged due to cascading failures.

[0221] Shadow drive hot standby, health-response dual threshold triggering, edge-cloud collaborative operation and maintenance, and safe shutdown bottom-up together build the execution layer fault precursor-hot standby preemption-remote diagnosis-risk downshift self-healing closed loop.

[0222] The shadow drive hot standby design breaks the traditional cold standby charging or manual machine replacement lag mode, achieving seamless second-level plugging; the dual threshold triggering allows switching decisions to be checked in both fault signal and quality signal dimensions, further reducing the risk of misplacement; edge-cloud collaboration upgrades empirical diagnosis to data-driven expert systems, shortening fault location time; the safe shutdown bottom-up strategy replaces shutdown by downgrading production, gaining critical buffer for order delivery.

[0223] Step five realizes millisecond monitoring-millisecond preemption-hour repair-safe downshift full-time execution chain guardianship with real-time health assessment as the vanguard and redundant switching and remote maintenance as the backup; the health matrix and remaining life The continuous write-back step four weight matrix completes the horizontal needle threading of data between step four-step five. At this point, the intelligent thickness control system completes the closed loop closure from multi-modal acquisition to health self-healing, ensuring that the high-brightness edge-sealed sheet thickness control can still be stable, economical, and long-period running under complex working conditions and equipment aging conditions.

[0224] Model predictive control provides a deterministic optimal solution for the production line, but when faced with sudden disturbances, dramatic changes in raw material batches, or equipment aging beyond the mechanism model description, it often shows parameter adjustment lag and precision decline.

[0225] Step six, the reinforcement learning (RL) policy is fused with the MPC controller through a tunable weight gate to reduce energy consumption and improve thickness control accuracy while meeting hard constraints. The RL agent uses historical experience replay and online reward reweighting to achieve safe fine-tuning. When the confidence interval or health degree touches the warning line, it automatically rolls back to the MPC baseline policy, ensuring that the performance gain brought by exploration and industrial safety are balanced in real time and are interpretable.

[0226] Step 601, reinforcement learning agent architecture and offline pre-training

[0227] The state-action-reward space is designed to perfectly match the edge banding sheet calendering scene, realizing deep reinforcement learning pre-training based on digital twin offline simulation to provide a high-performance initial policy for online use.

[0228] Observation vector of the agent With the digital twin state , the main eigenvalues of the health matrix main features , instantaneous power , and thickness prediction covariance are spliced to obtain the observation vector :

[0229]

[0230] Where: the main eigenvalue vector : extract the most concentrated few-dimensional signals of the health matrix energy through singular value decomposition;

[0231] Instantaneous power : the sum of roller temperature electric heating, servo drive, and cooling fan power.

[0232] Posterior state vector (corrected state): thickness field state, material / equipment parameter drift term, etc.

[0233] Sliding standard deviation or uncertainty estimate of the observation / feature vector .

[0234] Action vector and control vector increment : the same dimension: roller gap displacement increment, roller temperature adjustment, and traction speed increment. To ensure that the agent output is within the hard constraint range, the action is projected through a differentiable projection layer after network output:

[0235]

[0236] Original action vector given by the policy network at time ; to reinforce the control input vector output by the RL module;

[0237] projection operator : project original action into the physical-safety feasible region ; the feasible region step four constraints remain consistent.

[0238] reward function hybrid design, build reinforcement learning instant reward :

[0239]

[0240] wherein: thickness deviation weight , energy consumption weight , health penalty weight , production rhythm reward are positive real numbers;

[0241] health vector one norm : health degradation comprehensive penalty; is the fused real-time thickness estimate; is the target thickness, is the energy consumption index; is the standard deviation vector of the normalized feature vector ; is the production rhythm reward term;

[0242] When used, the composite reward guides the agent to find the Pareto optimal solution among thickness, energy consumption, equipment health and production capacity, rather than simply pursuing dimensional accuracy.

[0243] Build a fast reasoning environment using a digital twin model: encapsulate the step three three-layer mechanism model and the step five health-failure model into a Gym-like interface, and the simulation time of a second of physical process is no more than . In this environment, the Soft-Actor-Critic (continuous action) algorithm is used for pre-training; the policy network and the Q network share two layers of 256-node ReLU backbone, and output Gaussian policy parameters. During the training process, dynamic random injection of raw material viscosity fluctuations, roll gap friction increases and power grid voltage drop disturbances are used to make the strategy have generalization ability.

[0244] Offline simulation enables the agent to master the response strategy to common disturbances before going online, reducing the risk of online detection. Through multi-dimensional state design, projection action safety mapping, composite reward and digital twin fault simulation environment, offline pre-training obtains an initial intelligent agent, laying a high-performance, safe and controllable starting point for online fine-tuning.

[0245] ​By adding shift and batch number encoding to the state vector, the agent has advanced perception of both human and material variables; shell projection and acceleration constraints make the continuous output of deep strategy and mechanical safety boundary have a differentiable track, so that large-scale deep network can first walk with sub-millimeter gap requirements; compound reward steady-state window makes the agent switch to energy-saving mode actively, reflecting the dynamic balance of multi-objective optimization; multi-distribution Monte Carlo disturbance training makes the strategy maintain resilience to black swan extreme conditions.

[0246] Step 602, online safety fine-tuning and decision fusion

[0247] In real production, the reinforcement learning strategy is iteratively reinforced in a controlled manner and fused with the model predictive controller output to form a dual decision-making loop. The model predictive control is the first step and the agent action After weighted fusion:

[0248]

[0249] In the formula: The final issued mixed control input vector (roll gap increment, roll temperature adjustment amount, traction speed increment, etc.);

[0250] The action output by the reinforcement learning strategy; The reference control action, which can be the MPC optimization result, traditional PID / rule control, or the safety action of the previous period;

[0251] Weight coefficient Increases with online sampling volume , mainly relying on MPC at the starting stage, gradually tending to agent decision with experience accumulation, Steady-state maximum weight Rise rate constant.

[0252] Define safety monitoring index If the safety monitoring index Exceeds twice the tolerance or energy consumption Exceeds baseline , a rollback window is triggered: the weight coefficient Forced to zero , and the weight slowly rises again during this period to prevent the agent from continuously outputting undesirable actions under unknown extreme disturbances.

[0253] Real running data is stored in the form of priority experience pool, with priority Scoring according to TD error and thickness deviation events; every The background resampling updates the network parameters, and the learning rate is automatically halved with the mean of the health matrix, reducing excessive exploration during device aging and focusing on difficult samples and abnormal samples, improving sample efficiency.

[0254] Every cumulative Online training freeze strategy generation version , and the current running version alternates for 10 minutes AB comparison; if the thickness mean square error and energy consumption are better than the old version, upgrade; otherwise, roll back and write the failed samples to the low-priority pool to avoid shock. When using, versioning and small window AB testing ensure that each online is quantifiable improvement.

[0255] New direction consistency factor :

[0256]

[0257] When and the included angle exceeds ninety degrees, the weight is immediately reduced to (typical value 0.2), at the same time, record a direction conflict event and put the conflict samples into high priority into the experience pool to prevent the two sets of control quantities from canceling each other out and causing thickness jitter.

[0258] Online fine-tuning realizes a safe adaptive closed loop of stable takeover-progressive learning-index guardianship-continuous iteration through dynamic weight fusion, safety monitoring, priority experience playback, and strategy versioning verification.

[0259] The weight health gate uses device signs as a strategy empowerment valve, achieving parameter-level safety to strategy-level safety, an academic challenge; extreme deviation mode allows monitoring logic to go beyond threshold cutoff and can pinch soft landing according to deviation direction and speed; multi-scale experience pool focuses on abnormal samples, making limited online computing contribute more to the value interval; gray upgrade path increases along the order tail-segment-full day, avoiding direct impact of new strategies on high-value orders.

[0260] Step 601 provides an available initial strategy, step 602 ensures safe and controllable online learning, and merges with the MPC output through adjustable weight fusion, maintaining original stability and gradually injecting creative scheduling capabilities. The intelligent thickness control system completes the ultimate closed loop from multi-modal perception to adaptive intelligent decision-making, laying the foundation for the evolution of strategies for continuous optimization of pressure-laminated high-gloss light edge sheet in wider working conditions, longer life, and lower energy consumption.

[0261] The prior art high light edge sealing strip is coated with a UV coating on the surface to make the surface of the decorative edge sealing strip smooth, and the glossiness reaches 85GU or above, but it cannot be applied to the pull handle-free edge sealing process, because the shape of the pull handle-free edge sealing strip process has different shapes, and the edge sealing strip needs to be bent into different shapes. The high light edge sealing strip produced by surface coating UV coating is easy to crack and turn white on the surface when bent, which affects the appearance.

[0262] Further, the present application provides a soft-formable high-gloss edge sealing decorative strip. From the selection of raw materials, SG-5 or SG-8 type PVC resin powder is selected, and the vinyl chloride monomer is controlled within <1.0 mg / kg. The lubricant is selected from high-melting-point and high-molecular-weight PE wax, and the filler is selected from nano calcium powder. An environmentally friendly high-performance calcium-zinc stabilizer is used.

[0263] The process uses mirror roller smoothing forming technology, and the surface does not need to be printed again, so that the glossiness of the decorative strip reaches 85GU (glossiness unit) or above.

[0264] The formula of the paint-free high light edge sealing decorative strip of the present application is as follows:

[0265] PVC resin (SG-8) 100 parts; nano calcium carbonate 5-10 parts; processing aid: 1-1.5 parts; toughening agent: 7-10 parts; high-molecular-weight lubricant: 0.5-1.5 parts; antioxidant: 0.5-1.5 parts; high-performance calcium-zinc stabilizer: 3.5-5.5 parts; brightener: several; plasticizer: 2-4 parts; color powder: several; The main purpose of the present application is to fill the gap of the demand for high-gloss products in the soft-formable edge sealing decorative strip, and to solve the problems of cracking or surface paint film blistering during the soft-forming process of printed high-gloss products.

[0266] The high light edge sealing decorative strip of the present application has a glossiness of 85GU or above;

[0267] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0268] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0269] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is only a logical function division, and there can be another division manner for actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0270] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0271] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent thickness control system for calendering high-gloss edge-lit sheeting, comprising: a multi-modal acquisition module for synchronously acquiring the web using multiple physical principles, combining specular reflection suppression and dynamic range stitching, and collecting laser thickness data, terahertz thickness data, and ultrasonic thickness data in the same frequency and phase, and simultaneously recording environmental disturbance signals; a drift compensation fusion module for modeling drift mechanisms based on the environmental disturbance signals, extracting drift vectors and noise distributions for each sensing channel, adaptively adjusting a gain matrix using a Bayesian-Kalman fusion algorithm, and outputting a fused thickness vector with a confidence level; a digital twin synchronization module for constructing a digital twin model including three-layer mechanism submodels of roll system elasticity-geometry, material rheology-stress, and thermal field-curing, assimilating measured data using a Kalman-particle hybrid assimilation algorithm, and outputting a rolling thickness prediction sequence and corresponding uncertainty information; a closed-loop execution module for solving a control input sequence using a rolling quadratic programming model predictive control algorithm with confidence constraints based on the rolling thickness prediction sequence and the uncertainty information, and outputting a first-step control input for the current period to be sent to an actuator for adjusting roll gap displacement, roll temperature, and pulling speed; a health assessment module for establishing a multi-domain health matrix of electrical, mechanical, and environmental domains, performing real-time health assessment of the acquisition-fusion-control chain, and triggering shadow drive hot standby and self-healing strategies using exponential entropy weight scoring and Markov residual life prediction; a strategy fusion module for constructing an observation vector based on a state vector of the digital twin model, the multi-domain health matrix, and energy consumption indicators, calling a reinforcement learning agent pre-trained by digital twin offline simulation to output a candidate control input, weighting and fusing the candidate control input and the first-step control input according to a gating weight to obtain a final control input under the premise of meeting line speed, energy consumption, and safety constraints, and setting the gating weight to zero to roll back to the first-step control input when a safety monitoring indicator exceeds a threshold.

2. The intelligent thickness control system of claim 1, wherein, The multi-modal acquisition module acquires thickness signals in real time in the same spatial profile and temporal granularity using a laser displacement sensor, a terahertz pulse probe, and an air-coupled ultrasonic probe in a double-axis up-and-down shooting mode, and improves thickness signal quality through mirror reflection suppression and double-redundancy calibration mechanisms.

3. The intelligent thickness control system of claim 2, wherein, The multi-modal acquisition module further includes an atomic clock and fiber synchronization technology to ensure millisecond-level phase synchronization of different sensing channels, a master-slave clock structure and least squares bias correction to unify time references, and full-width air running self-detection and mirror reflection target calibration to achieve a priori alignment of thickness data and coordinate systems.

4. The intelligent thickness control system of claim 3, wherein, The drift compensation fusion module includes a drift prediction function constructed using a linear-nonlinear mixed kernel regression based on the intrinsic correlation between environmental disturbances and historical thickness sequences, real-time extraction of slow-varying drift vectors and instantaneous noise distributions for each sensing channel, and adaptive adjustment of a gain matrix of the Bayesian-Kalman fusion algorithm through a confidence matrix.

5. The intelligent thickness control system of claim 4, wherein, The drift compensation fusion module further comprises triggering an abnormal replacement mechanism to replace the abnormal channel with the predicted value of the digital twin model to maintain the continuity and stability of the fused thickness vector through Kalman residual and Mahalanobis distance threshold judgment when single-channel anomaly occurs.

6. The intelligent thickness control system of claim 5, wherein, The digital twin synchronization module comprises state assimilation with real measurement data through Kalman-particle hybrid assimilation algorithm, and real-time ingestion of process parameter flow to update model boundary conditions.

7. The intelligent thickness control system of claim 6, wherein, The digital twin synchronization module further comprises rolling prediction using a fourth-order explicit Runge-Kutta-Cash-Karp variable step strategy, outputting a thickness prediction curve and an uncertainty band, and encapsulating and forwarding the thickness prediction curve and the uncertainty band to the model predictive controller through the message bus.

8. The intelligent thickness control system of claim 7, wherein, The closed-loop execution module comprises using the rolling thickness prediction sequence to solve the control input sequence that minimizes the weighted square sum of thickness deviation using the rolling quadratic programming model predictive control algorithm with confidence constraints, and issuing the first-step control input to the execution mechanism through a high-speed scheduling mechanism.

9. The intelligent thickness control system of claim 8, wherein, The closed-loop execution module further comprises translating the first-step control input into servo pulses, roll temperature valve pulse width, and traction drive reference frequency within a millisecond-level network cycle, and ensuring the accuracy and continuity of execution response through real-time feedback nesting and redundant switching mechanism.

10. The intelligent thickness control system of claim 9, wherein, The health assessment module comprises millisecond-level health assessment of the acquisition-fusion-control whole link, construction of a three-domain health matrix of electrical, mechanical and environmental domains, and real-time discrimination of device performance degradation using exponential entropy weight scoring and Markov residual life prediction.

11. The intelligent thickness control system of claim 10, wherein, The health assessment module further comprises triggering shadow drive hot standby, cloud operation and maintenance, and self-healing strategies when the health degree is below the threshold, and achieving long-period high availability and low downtime risk of the system through software and hardware hybrid redundancy preemption.

12. The intelligent thickness control system of claim 11, wherein, The strategy fusion module comprises designing a reinforcement learning agent architecture compatible with the edge-coated sheet calendering scene, pre-training a deep reinforcement learning through digital twin offline simulation to generate a high-performance initial strategy, and providing a safe and controllable starting point for online fine-tuning.

13. The intelligent thickness control system of claim 12, wherein, The strategy fusion module further comprises iteratively reinforcing the learning strategy in real production in a controlled manner, fusing it with the first-step control input of the rolling quadratic programming model predictive control algorithm with confidence constraints through adjustable weight gating, and achieving robust adaptive intelligent decision-making through safety monitoring and versioned verification.

Images