A heat transfer printing temperature field uniformity intelligent control method and system

Abstract

The present application relates to the technical field of textile printing and dyeing equipment control, and discloses a heat transfer printing temperature field uniformity intelligent control method and system. The present application embeds the disturbance prior value deduced by the thermodynamic physical model into the waveform matching process as prior knowledge through dynamic time warping of physical constraint, so as to realize accurate retrieval of historical experience under the guidance of physical law, and adaptively selects matching samples through dynamic similarity criterion, combines with the distribution fusion strategy, ensures that the optimal trust distribution between historical experience and physical model can be adaptively selected under different working conditions, and the disturbance estimation obtained by the present application is injected into the active disturbance rejection controller in the form of feedforward, which actively compensates the disturbance before the feedback loop responds, greatly improves the response speed and anti-disturbance ability of temperature control, and through the double-layer mechanism of online model correction and differential bias compensation, it is beneficial to inhibit the temperature field non-uniformity caused by the physical structure difference of the temperature control zone from the root.

Description

Technical Field

[0001] This invention relates to the field of textile printing and dyeing equipment control technology, and in particular to an intelligent control method and system for the uniformity of temperature field in heat transfer printing. Background Technology

[0002] Heat transfer printing is a process that transfers a pattern pre-printed on transfer paper onto a fabric through heat pressing. The core of this process is to apply high temperature and pressure to the transfer paper and fabric using a heated roller, causing the dye to sublimate and penetrate into the fabric fibers. The uniformity of the temperature field on the roller surface directly determines the printing quality. Uneven temperature distribution can lead to quality problems such as color difference and uneven penetration, which seriously affects the product qualification rate.

[0003] The heating plate of a heat transfer printing machine is typically divided into multiple independent temperature control zones, each equipped with an independent temperature sensor and heating element. A controller independently adjusts the temperature of each zone to maintain temperature uniformity across the entire heating plate. The heat transfer printing process is typically periodic and repetitive. Each printing cycle includes stages such as mold closing, pressure holding, and mold opening. Each time the mold closes, the cold substrate comes into contact with the hot plate, causing a significant temperature drop. Furthermore, different batches of substrates vary in material, thickness, and initial temperature, resulting in different magnitudes and recovery characteristics of the temperature drop. The core drawback of conventional zoned PID control is that once the PID parameters are tuned, they remain fixed and cannot adaptively adjust to changes in operating conditions. When the substrate material changes, the temperature control quality (such as overshoot, settling time, and steady-state accuracy) under fixed PID parameters will significantly decrease with changing operating conditions. A single parameter set cannot maintain optimal control performance under all operating conditions. Furthermore, existing temperature field control methods typically rely on temperature deviation feedback for adjustment during the current cycle. However, due to inherent differences in thermal response characteristics between independent temperature control zones caused by differences in physical structures such as heating element layout and heat dissipation boundary conditions, as well as inherent differences in thermal response characteristics between temperature control zones, existing technologies cannot provide differentiated compensation for the personalized thermal response characteristics of each zone when faced with varying operating conditions, model mismatch, and inherent regional differences, resulting in systematic temperature deviations between zones. Summary of the Invention

[0004] The main objective of this invention is to provide an intelligent control method and system for the uniformity of the temperature field in heat transfer printing, aiming to solve the technical problems in the prior art.

[0005] This invention proposes an intelligent control method for the uniformity of the temperature field in heat transfer printing, comprising: The historical waveform library is retrieved, which includes multiple historical temperature waveforms and their corresponding historical disturbance data, and the original temperature signals of each independent temperature control zone of the heating plate are collected by a temperature sensor. The original temperature signal is calibrated and converted to obtain a temperature response waveform. The waveform window length of each zone is determined based on the temperature response waveform in order to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. Based on the pre-calibrated thermodynamic physical model, the initial temperature waveform segments are extrapolated and calculated to obtain the perturbation prior prediction values ​​corresponding to each region. The maximum warping offset is set based on the perturbation prior prediction value, and dynamic time warping of the initial temperature waveform segment and the historical temperature waveform of each region is performed with physical constraints based on the maximum warping offset to obtain the waveform similarity. A dynamic similarity criterion is constructed by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment. The waveform similarity is then used to perform the optimal selection of historical temperature waveforms and the coupling configuration of fusion weights. Based on the historical temperature waveform, disturbance prior prediction value and fusion weight of each zone, fusion and compensation mapping are performed to obtain the feedforward compensation amount of each zone. The feedforward compensation amount is superimposed on the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. After the printing cycle ends, the steady-state disturbance increment and temperature drop integral of each zone are obtained. The thermodynamic physical model is iteratively corrected online, and the temperature field non-uniformity deviation of each zone is quantified to implement differentiated bias compensation for the hysteresis temperature control area.

[0006] This application also provides an intelligent control system for the uniformity of the temperature field in heat transfer printing, including: The data acquisition module is used to retrieve the historical waveform library, which includes multiple historical temperature waveforms and their corresponding historical disturbance data, and to collect the original temperature signals of each independent temperature control zone of the heating plate through a temperature sensor. The waveform extraction module is used to calibrate and convert the original temperature signal to obtain a temperature response waveform, and to determine the waveform window length of each zone based on the temperature response waveform, so as to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. The prior prediction module is used to perform deduction and calculation on each of the initial temperature waveform segments based on a pre-calibrated thermodynamic physical model to obtain the perturbation prior prediction value corresponding to each region. The physical warping module is used to set the maximum warping offset based on the perturbation prior prediction value, and to perform dynamic time warping of the initial temperature waveform segments and historical temperature waveforms of each region based on the maximum warping offset to obtain waveform similarity. The dynamic filtering module is used to construct a dynamic similarity criterion by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment, and to perform optimal filtering of historical temperature waveforms and coupling configuration of fusion weights by combining the waveform similarity. The feedforward compensation module is used to perform fusion and compensation mapping based on the historical temperature waveform, disturbance prior prediction value and fusion weight corresponding to each zone to obtain the feedforward compensation amount of each zone. The feedforward compensation amount is superimposed on the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. The online correction module is used to obtain the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correct the thermodynamic physical model online, and quantify the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control area.

[0007] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent control method for the uniformity of the temperature field in heat transfer printing.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described intelligent control method for the uniformity of the temperature field in heat transfer printing.

[0009] The beneficial effects of this invention are as follows: By dynamically warping the physical constraints, this invention embeds the prior values ​​of disturbances derived from the thermodynamic physical model as prior knowledge into the waveform matching process, overcoming the defect of pure data matching that easily leads to physically infeasible alignment. This achieves accurate retrieval of historical experience under the guidance of physical laws. On this basis, it adaptively filters matching samples through dynamic similarity criteria and combines an allocation fusion strategy to dynamically configure the fusion weights of data-driven and physical priors based on the difference between the current operating conditions and historical experience. This ensures that the optimal trust allocation can be adaptively selected between historical experience and physical models under different operating conditions, which is beneficial to significantly improve the accuracy of disturbance estimation in complex scenarios. Furthermore, this invention injects the fused disturbance estimate into the active disturbance rejection controller in a feedforward form, actively compensating for disturbances before the feedback loop responds. This greatly improves the response speed and disturbance rejection capability of temperature control. Through a two-layer mechanism of online model correction and differentiated bias compensation, it solves the problems of model parameter drift caused by equipment aging and the inherent differences in thermal response characteristics of each region, which is beneficial to suppress the temperature field non-uniformity caused by differences in the physical structure of the temperature control zone from the root. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0012] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0014] like Figure 1 As shown, this application provides an intelligent control method for the uniformity of the temperature field in heat transfer printing, comprising: S1. Retrieve the historical waveform library, wherein the historical waveform library includes multiple historical temperature waveforms and their corresponding historical disturbance data, and the original temperature signals of each independent temperature control zone of the heating plate are collected by a temperature sensor. Exemplarily, this invention is applicable to the temperature control of the heating plate of a heat transfer printing machine, and is particularly suitable for production scenarios where the working parameters such as the material, thickness, and initial temperature of the substrate vary within a certain range. As an example, the types of substrates applicable to this invention include, but are not limited to: cotton, polyester, and blended fabrics, with a thickness range of 0.1~5mm and an initial temperature range of 15~40℃. The historical waveform library in this invention is a pre-built and continuously updated database that stores temperature response waveform sequences of multiple historical printing cycles and historical disturbance data corresponding to each waveform. The historical disturbance data characterizes the external disturbances or internal parameter perturbations that need to be compensated to maintain the uniformity of the temperature field during the historical printing cycle. In practical applications, after each printing cycle ends, if the printing quality of the cycle is qualified, the system will store the temperature response waveforms of each area collected in this cycle and the ideal disturbance compensation amount obtained by reverse calculation as a historical record in the waveform library. The system also collects the original temperature signals of each independent temperature control zone in real time through a temperature sensor at a fixed sampling frequency (e.g., 100Hz). The original temperature signal refers to the original electrical signal directly output by the temperature sensor without calibration conversion.

[0015] S2. The original temperature signal is calibrated and converted to obtain a temperature response waveform, and the waveform window length of each zone is determined according to the temperature response waveform in order to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. For example, the present invention performs analog-to-digital conversion on the original temperature signal: the voltage value or code value is converted into a temperature value in degrees Celsius according to the sensor calibration curve, and a temperature response waveform is obtained. The temperature response waveform refers to the discrete sampling point sequence of temperature changes over time in each independent temperature control zone collected by the temperature sensor from the moment of mold closing trigger. Since the speed and degree of temperature drop are different in different printing cycles, a fixed window length will cause the extracted segment to fail to fully contain the key dynamic change information or contain too many useless steady-state segments. Therefore, the present invention adaptively determines the waveform window length applicable to each zone according to the dynamic characteristics of the temperature response waveform of the current cycle. Based on the window length, the initial temperature waveform segment for subsequent processing is extracted from the complete temperature response waveform, starting from the sampling point index corresponding to the moment of mold closing trigger.

[0016] S3. Based on the pre-calibrated thermodynamic physical model, the initial temperature waveform segments are extrapolated and calculated to obtain the perturbation prior prediction values ​​corresponding to each region. For example, the present invention constructs a thermodynamic physical model, which is a mathematical model describing the heat transfer, heat dissipation, and heat exchange with the outside environment between the independent temperature control zones of the heating plate. Its mathematical form is as follows: ;in, This is the temperature control zone number. For the first The temperature of each temperature-controlled zone For the first The equivalent heat capacity of each temperature control zone For the first The temperature control zone and the first The equivalent thermal resistance between the temperature control zones For the first The equivalent thermal resistance between each temperature-controlled zone and the environment. For the first Heating power of each temperature control zone For the first The equivalent disturbance heat flow of each temperature control zone For ambient temperature, model parameters , , Obtained through identification and correction of experimental data; and in the thermodynamic physical model To estimate the state variables, a Kalman filter (or backpropagation algorithm) is used to estimate them in real time from the measured temperature. As the prior predicted value of the equivalent disturbance compensation, the thermodynamic physical model is used as the equation of state, and the measured temperature is used as the equation of state. For the observed values, a discretized Kalman filter recursive form is established to estimate the equivalent perturbation heat flux of each region at each sampling time using the temperature measurement values. This estimate serves as a priori prediction of the unknown disturbance trends in each region of the current cycle, reflecting the possible unknown disturbance trends in each region of the current cycle, such as sudden changes in the substrate material, a sudden drop in ambient temperature, or a decline in the performance of heating elements. S4. Set the maximum warping offset based on the prior prediction value of the disturbance, and perform dynamic time warping of the initial temperature waveform segment and the historical temperature waveform of each region based on the maximum warping offset to obtain the waveform similarity. For example, the absolute value of the perturbation prior prediction reflects the degree to which the current operating condition deviates from the nominal operating condition of the thermodynamic physical model. When there is an unknown strong perturbation, such as a sudden change in the substrate material, a sudden drop in ambient temperature, or a degradation in the performance of the heating element, the actual temperature response waveform of each independent temperature control zone of the heating plate will produce compression or stretching distortion relative to the nominal waveform on the time axis. Therefore, this invention calculates the maximum warping offset using a piecewise linear function based on the absolute value of the perturbation prior prediction. The maximum warping offset refers to the maximum number of steps allowed for the warping path to deviate from the main diagonal in the Dynamic Time Warping (DTW) algorithm, which is used to limit the physical reasonable range of time axis stretching. This invention embeds physical prior knowledge directly into the constraints of the DTW algorithm by associating the perturbation prior prediction with the maximum warping offset. When the perturbation is small, the warping path is restricted to searching only near the main diagonal, avoiding matching to physically impossible extreme stretching paths. When the perturbation is large, the search range is adaptively widened, preserving the flexibility of matching. In a dynamic programming grid with applied band constraints, the cumulative distance matrix is ​​calculated according to the recursive formula of the dynamic time warping algorithm. The cumulative distance at the endpoint is used as the waveform similarity between the current waveform and the historical waveform. This waveform similarity comprehensively considers the similarity between the two waveforms in terms of time axis scaling and amplitude difference: the smaller the distance, the more similar the two waveforms are within a physically reasonable time scaling range. Compared with traditional DTW, this physically constrained dynamic time warping can output waveform similarity with more physical interpretability and practical reference value, providing a more reliable basis for subsequent selection of historical waveforms.

[0017] S5. Combine the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment to construct a dynamic similarity criterion, and combine the waveform similarity to perform the optimal selection of historical temperature waveforms and the coupling configuration of fusion weights. For example, this invention extracts working condition signal features that characterize the current printing working condition from the initial temperature waveform segment, including the standard deviation of high-frequency noise, and dynamically adjusts the similarity judgment threshold according to the current working condition signal features to construct a dynamic similarity criterion that adapts to the noise level. Based on this dynamic similarity criterion and waveform similarity, historical temperature waveforms in the historical waveform library are selected preferentially. Based on the design that data-driven weights are inversely proportional to waveform similarity, it ensures that historical waveforms more similar to the current working condition contribute more to the fusion, reflecting the core idea of ​​similarity compensation for similar working conditions. Simultaneously, feature difference is extracted from the historical waveform library, and the difference between the perturbation prior prediction value and the data-driven perturbation estimate value is combined. The system employs a method that incorporates physical prior fusion weights based on high-frequency noise variance. This allows the confidence level of the physical model's derivation results to adaptively adjust according to the quality of the measured signal: the model is trusted when the signal is good, and its contribution is reduced when the signal is poor, thus preventing the amplification of model errors. Furthermore, the system calculates the fused predicted perturbation by weighted summation of data-driven weights, data-driven perturbation estimates, physical prior fusion weights, and perturbation prior predictions. This achieves complementarity between two information sources: historical data provides effective compensation experience that has actually occurred, while the physical model provides the ability to respond to transient perturbations that have never occurred before. Compared to perturbation estimation from a single source, this coupled configuration has higher accuracy and robustness.

[0018] S6. Based on the historical temperature waveform, disturbance prior prediction value and fusion weight corresponding to each zone, perform fusion and compensation mapping to obtain the feedforward compensation amount of each zone, and superimpose the feedforward compensation amount onto the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. For example, the present invention converts the fused predicted disturbance into a feedforward compensation amount through a compensation mapping function. This feedforward compensation amount is then superimposed on the total disturbance estimate inside the active disturbance rejection controller to form a corrected control output. Since the feedforward compensation outputs the corresponding compensation power in advance before the feedback loop responds, it can effectively suppress temperature drop and temperature non-uniformity. The corrected control output drives the heating element to perform temperature regulation. In this embodiment, the active disturbance rejection controller adopts a linear active disturbance rejection controller (LADRC), and its extended state observer (ESO) is: , ;in, This is a temperature estimate. This is the estimated total disturbance. , For observer gain, To control the gain, For actual measured temperature, To control the output, the feedforward compensation is superimposed as follows: At the control output of the ADRC, the feedforward compensation is directly added to the control law output of the LADRC, where the control law output is the basic control quantity calculated by the ADRC based on the temperature deviation, and the ADRC parameters... , , Determined according to the standard LADRC parameter tuning rules: , , ,in, The desired closed-loop bandwidth is determined based on the thermal response time constant of the heating plate. The thermal response time constant is obtained through step response experiments. After adding feedforward compensation, the ADRC parameters do not need to be readjusted and can be directly calculated using the above rules. This invention uses historical waveform retrieval and feedforward compensation to output compensation in advance when disturbance signs appear in the current operating condition. This allows temperature regulation to start before the disturbance is fully manifested, eliminating the need to wait for temperature deviation to occur before PID feedback regulation. This solves the problem that fixed PID parameters cannot adapt to changes in operating conditions.

[0019] S7. Obtain the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correct the thermodynamic physical model online, and quantify the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control area.

[0020] For example, after the printing cycle ends, the system obtains the actual disturbance data and temperature drop integral recorded during the cycle, and uses a recursive least squares updater to perform online iterative correction on the thermal gain coefficient in the thermodynamic physical model. The updated thermal gain coefficient is obtained and stored in non-volatile memory, so that the control model can adaptively adjust with the changes in equipment state and maintain long-term control accuracy, thereby solving the problem of model parameter drift caused by equipment aging. Simultaneously, the system further quantifies the temperature field non-uniformity deviation between each independent temperature control zone. By extracting key temperature characteristic parameters of each temperature control zone, normalizing them, and then weighted summing them, the temperature response deviation index of each temperature control zone is obtained. Based on this deviation index, lagging temperature control areas are identified, and bias compensation is calculated for each identified lagging temperature control area. This compensation is then superimposed on the basic control output of the corresponding area at the start of the next printing cycle. The timing relationship between differentiated bias compensation and feedforward compensation is as follows: feedforward compensation is executed in real time within the current printing cycle, targeting transient temperature fluctuations caused by external disturbances within the cycle; differentiated bias compensation is pre-superimposed on the basic control output at the start of the next printing cycle, targeting... The inherent long-term thermal response characteristics of each temperature control zone differ, and the execution sequences of the two are different, so they do not interfere with each other. The calculation of the bias compensation amount is based on the statistical characteristics of historical cycles and does not affect the real-time calculation of the feedforward compensation amount in the current cycle. Moreover, the differentiated bias compensation and the feedforward compensation complement each other: the feedforward compensation is for transient temperature fluctuations caused by external disturbances within the cycle and has cycle specificity; the differentiated bias compensation is for the inherent thermal response characteristics differences between each temperature control zone and has long-term stability and accumulation. The two work together to apply personalized bias amounts according to the historical thermal response characteristics of each temperature control zone through differentiated bias compensation, thereby suppressing the inherent temperature differences between regions from the root and jointly achieving long-term uniformity of the temperature field.

[0021] In one embodiment, prior to step S1 of retrieving the historical waveform library, the following steps are included: S11. After the end of each historical printing cycle, obtain the image detection defect rate and image detection color difference of the printed finished product to obtain the final printing quality assessment value of that cycle. For example, after each printing cycle, the printed product undergoes an online or offline image inspection system. This system analyzes the transfer effect on the surface of the finished product using computer vision algorithms, outputting the image detection defect rate and image detection color difference. When the transfer uniformity is high, the hot melt adhesive or dye can be transferred completely and uniformly from the transfer paper to the substrate surface, and the defect rate decreases exponentially. When the transfer uniformity is low, incomplete transfer leads to defects such as local missing parts and smudging, and the defect rate rises rapidly. Moreover, the occurrence of defects exhibits a threshold effect, that is, defects only appear significantly when the uniformity is below a certain critical value. Therefore, a physical mapping relationship between the defect rate and the transfer uniformity can be established: ,in, Indicates the image detection defect rate. This represents the defect rate saturation coefficient. This indicates the uniformity of the printing transfer; the value is defined as the reciprocal of the standard deviation of the temperature in each temperature-controlled zone. This represents the uniformity characteristic scale; while color difference mainly stems from the deviation of heat exchange efficiency from the optimal value: when the heat exchange efficiency is too low, insufficient heat leads to incomplete dye sublimation, resulting in a lighter and less saturated transferred color; when the heat exchange efficiency is too high, overheating leads to excessive dye diffusion or thermal degradation, producing color deviation or spots. Within a small deviation range, the color difference change is relatively gradual; as the deviation increases, the color difference deteriorates rapidly. To accurately characterize this perceptual characteristic of being initially gradual and then abruptly worsening, the physical mapping relationship between color difference and heat exchange efficiency established in this invention is as follows: ,in, Indicates the detection of color difference in an image. Indicates the color difference gain coefficient. This indicates the heat exchange efficiency, which is defined as the ratio of the actual heat transferred to the substrate to the theoretical heat output of the heating plate. This represents the optimal heat exchange efficiency. Indicates the color difference perception index; This invention measures the saturation defect rate under extremely non-uniform conditions, averages multiple experiments to obtain a defect rate saturation coefficient, and collects multiple sets of sample data with known transfer uniformity and image detection defect rate. A nonlinear least squares method is used to fit and obtain the uniformity feature scale. While keeping other conditions constant, the heat exchange efficiency is changed and the corresponding color difference is measured. The heat exchange efficiency corresponding to the minimum color difference is taken to obtain the optimal heat exchange efficiency. Similarly, color difference is measured under standard experimental conditions where the heat exchange efficiency deviates from the optimal heat exchange efficiency, and the color difference gain coefficient is calculated by reverse calculation. Multiple sets of sample data with known heat exchange efficiency and image detection color difference are collected, and a power function is used to fit and obtain a nonlinear exponent. Based on these pre-calibrated data, combined with the currently measured image detection defect rate and image detection color difference, the two mapping relationship functions mentioned above are substituted to reverse calculate the absolute values ​​of transfer uniformity deviation and heat exchange efficiency deviation. "Calculate the final printing quality assessment value, where, This indicates the final print quality assessment value. Indicates the deviation in transfer uniformity. This indicates the maximum permissible deviation in transfer uniformity. This represents the absolute value of the deviation in heat exchange efficiency. This indicates the maximum permissible deviation in heat exchange efficiency. This represents the uniformity deviation weighting coefficient. The weighting coefficient for heat exchange efficiency deviation can be flexibly configured according to the sensitivity of different printing processes to defect rate and color difference, reflecting the adaptability of the process. This invention makes quality evaluation no longer just a statistical fusion at the data level, but a deductive evaluation based on the physical essence of the process: the quality is interpreted as the degree to which physical parameters deviate from the optimal state. S12. The period is determined to be a valid learning period only when the final printing quality assessment value is higher than the preset quality qualification threshold. For example, this invention uses a preset standard quality qualification threshold based on production quality requirements. Only when the final printing quality assessment value is greater than the quality qualification threshold, that is, the final printed product quality meets the quality standard of the printing cycle, does its data have learning value. If the quality does not meet the standard, the data of that cycle is discarded or marked as invalid, thus avoiding storing inferior data in the historical waveform library. The quality qualification threshold is determined according to industry standards or user-defined product quality requirements. For example, when the final printing quality assessment value is greater than 0.85, it is judged as qualified. S13. For the effective learning period, collect and save the temperature response waveform sequence from the moment of mold closing triggering to the moment of temperature recovery and stabilization within the period. For example, for each effective learning cycle, the present invention takes the mold closing trigger moment as the key time point for the start of the printing action, and starts to collect temperature response waveforms from this moment. The time point when the heating plate temperature recovers to the set temperature after mold opening and the fluctuation is less than a certain threshold is taken as the temperature recovery stabilization moment. The complete waveform sequence from mold closing trigger to temperature recovery stabilization can be collected, which can completely record the entire process of temperature drop, heat preservation and recovery. S14. Based on the temperature field steady-state uniformity result finally achieved in this cycle, the ideal disturbance compensation amount required to achieve the steady-state result is solved in reverse based on the thermodynamic physical model. For example, the present invention obtains the temperature variance between each independent temperature control zone based on the temperature response waveform sequence within each effective learning cycle, and obtains the steady-state uniformity of the temperature field. Under the premise that the temperature field uniformity meets the preset quality requirements, the actual observed temperature waveform and the steady-state uniformity index are input into the thermodynamic physical model together, and the least squares method with regularization constraints is used to estimate an equivalent perturbation compensation amount that can achieve the current uniformity result under the thermodynamic environment. S15. Associate the temperature response waveform sequence with the ideal disturbance compensation amount, store it as a historical record in the historical waveform library, and add a timestamp tag to each record; For example, the present invention treats the collected complete temperature response waveform sequence and the ideal disturbance compensation amount obtained by inverse solving as a pair of correlated data to form a complete historical record stored in the historical waveform library. At the same time, a timestamp tag is added for subsequent first-in-first-out management. In order to avoid the historical waveform library from growing indefinitely and causing excessive storage and computing overhead, the present invention presets a capacity limit. The capacity limit is determined according to the controller's storage resources and real-time matching computing efficiency, for example, storing the most recent 1000 valid records. When the limit is reached, the historical record with the earliest timestamp is automatically deleted according to the first-in-first-out principle to make room for new valid records.

[0022] In one embodiment, step S2, which involves determining the waveform window length of each zone based on the temperature response waveform to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform, includes: S21. Extract temperature response waveforms from the historical waveform library for multiple completed printing cycles, and obtain the average drop slope peak value based on the multiple temperature response waveforms. For example, this invention extracts the temperature response waveforms of the most recent (e.g., 10) completed printing cycles from a historical waveform library, and obtains the peak value of the temperature drop slope for each cycle based on the temperature response waveforms. The peak value of the temperature drop slope refers to the maximum value of the temperature derivative with respect to time during the rapid temperature drop caused by the cold substrate contacting the hot plate at the moment of mold closing. This peak value reflects the most severe degree of heat being rapidly removed in the current printing cycle. By statistically analyzing the waveforms of the most recent completed cycles, the recent trend of temperature drop can be obtained. This invention performs a moving average processing on the extracted multiple peak values ​​of temperature drop slope to smooth out random fluctuations and obtain a relatively stable average peak value of temperature drop slope. This average value characterizes the average thermal shock intensity under recent printing conditions, realizing adaptive perception of the thermal shock intensity of the current printing conditions. S22. Based on the historical waveform library, identify the temperature drop start time and temperature recovery stabilization time of multiple historical temperature response waveforms to obtain the characteristic recovery time, and perform statistical analysis on multiple characteristic recovery times to obtain the minimum window length and the maximum window length; For example, whenever a certain number of valid learning cycle records are added and accumulated in the historical waveform library, the present invention extracts all stored valid historical temperature response waveforms from the historical waveform library. For each waveform, it first identifies the start time of its temperature drop, which corresponds to the mold closing trigger time of the printing cycle. Then, it identifies the time when the temperature recovers and stabilizes, that is, the time when the heating plate temperature rises back to the set temperature after mold opening and the fluctuation does not exceed a preset threshold (e.g., ±0.5℃) for more than 1 second. The difference between the two is calculated to obtain the feature recovery time. The feature recovery time reflects the heating plate temperature under the specific thermal shock conditions of the historical printing cycle. The actual time required for the system to fully recover to steady state after being subjected to thermal shock; after traversing all valid historical temperature waveforms, the system obtains a statistical sample set consisting of a large number of characteristic recovery times. Statistical analysis is performed on this set to calculate its 5th percentile and 95th percentile. The 5th percentile represents the critical value corresponding to the 5% of historical temperature waveforms with shorter characteristic recovery times, reflecting the typical recovery speed of the equipment under weak thermal shock conditions. The 95th percentile represents the critical value corresponding to the 5% of historical temperature waveforms with longer characteristic recovery times, reflecting the limit of the equipment's recovery capability under extreme conditions with stronger thermal shock. To prevent boundary values ​​from falling precisely on outliers and causing subsequent new operating conditions to exceed the boundaries, the system adds a preset margin based on the quantiles. This margin is pre-calibrated according to the thermal inertia characteristics of the equipment and is used to cover the additional recovery time caused by extreme operating conditions that have not yet occurred or the slow degradation of equipment performance. Finally, combined with the sampling frequency of the temperature sensor, the updated minimum and maximum window lengths are calculated and stored in non-volatile memory for use in subsequent printing cycles. Through the above adaptive update mechanism, when long-term operation of the equipment leads to aging of the heating element, degradation of the thermal conductivity, or changes in the thermal capacity characteristics, the statistical distribution of the feature recovery time is observed. The window length will gradually shift to the right, and the minimum and maximum window lengths will automatically increase to adapt to the slower thermal response speed. Conversely, when the performance recovers after equipment maintenance or the thermal shock is reduced after production process optimization, the window length boundary will also shrink accordingly. This mechanism ensures that the boundary value of the waveform window length always matches the current actual physical state of the equipment, avoiding the problem of fixed boundary values ​​gradually failing within the service life. At the same time, it enables subsequent dynamic time warping matching to be performed within the optimal window length range, so that key dynamic information is not missed due to the window being too short, and redundant steady-state segment data is not introduced due to the window being too long, thus reducing computational efficiency. S23. Obtain the waveform window length of the current printing cycle based on the average drop slope peak value, the minimum window length, and the maximum window length; For example, the stronger the thermal shock, the faster and deeper the temperature drop, requiring a longer time window to fully record the entire process from temperature drop to recovery and stabilization. Conversely, with weaker thermal shocks, the temperature change is gradual, and a shorter window can contain complete dynamic information. Therefore, this invention determines the waveform window length for the current printing cycle based on the magnitude of the average drop slope peak and a preset segmented mapping rule: if the average drop slope peak is less than the low thermal shock threshold, the minimum window length is used as the waveform window length for the current printing cycle; if the average drop slope peak is greater than the high thermal shock threshold, the maximum window length is used as the waveform window length for the current printing cycle; if the average drop slope peak is between the low and high thermal shock thresholds, the waveform window length for the current printing cycle is calculated through linear interpolation to ensure that the window length changes smoothly with the thermal shock intensity, avoiding discontinuous matching results caused by jumps in window length. S24. Starting from the sampling point index corresponding to the mold closing trigger moment, extract the initial temperature waveform segment from the temperature response waveform of the current printing cycle according to the waveform window length. For example, after determining the waveform window length of the current printing cycle, the present invention, when actually performing dynamic time warping matching of physical constraints, only extracts sampling points within the waveform window length range of the current temperature response waveform to participate in the matching calculation. The specific extraction operation is as follows: Assume the current temperature response waveform is a complete time sequence: in, Indicates the first Temperature values ​​at each sampling point Indicates the sampling point index. This represents the total number of sampling points collected throughout the entire printing cycle. The index of the sampling point corresponding to the mold closing trigger moment is denoted as "". "This moment marks the beginning of the temperature drop; the starting point of the interception operation is set at the mold closing trigger moment." The endpoint is set as " ",in, This represents the waveform window length for the current printing cycle, specifically the portion extracted from the moment the die closes to the trigger point, with a waveform window length of "". "A continuous subsequence of sampling points is used as the initial temperature waveform segment, and the mold closing trigger time is determined by the mold closing signal of the printing equipment (such as the pressure switch signal or the timing of the mold closing command output by the controller). The steady-state segment data outside the window, i.e., the index..." "less than" "Prefix and index" "greater than" The segment following the "" does not participate in subsequent DTW matching calculations. The core of the above truncation method is: taking the moment of mold closing triggering as a fixed anchor point, not truncating before it, because the steady-state temperature information before mold closing does not contribute to the description of the impact process, and truncating after it to a dynamically variable length, so that the truncated waveform segment always contains the core dynamic information from the beginning of the impact to the early stage of recovery, while excluding the long steady-state segment after recovery, avoiding the problem that the fixed window truncates insufficient information when the thermal shock is strong and truncates too much redundant data when the thermal shock is weak, thereby minimizing the amount of data participating in DTW calculations without losing key information and improving matching efficiency.

[0023] In one embodiment, step S4, which involves setting a maximum warping offset based on the perturbation prior prediction value and performing dynamic time warping of the initial temperature waveform segments and historical temperature waveforms of each region based on the maximum warping offset to obtain waveform similarity, includes: S41. Obtain the regularization offset boundary interval, and obtain the maximum regularization offset based on the regularization offset boundary interval and the perturbation prior prediction value. For example, the absolute value of the prior prediction of the disturbance reflects the degree to which the current operating condition deviates from the nominal operating condition of the model. The physical meaning of this deviation is that when there is an unknown strong disturbance, such as a sudden change in the substrate material, a sudden drop in ambient temperature, or a degradation in the performance of the heating element, the actual temperature response waveform of each independent temperature control zone of the heating plate will produce compression or stretching distortion on the time axis relative to the nominal waveform. For example, a strong heat dissipation disturbance will prolong the duration of the temperature drop phase, and the waveform will be stretched, while a strong heating disturbance may accelerate the temperature recovery phase, and the waveform will be compressed. This stretching distortion on the time axis is essentially driven by the change in the equivalent thermal response time caused by the disturbance. Therefore, this invention estimates the change in the equivalent thermal response time of the current period relative to the nominal operating condition based on the prior prediction of the disturbance. ,in, This represents the change in equivalent thermal response time. The characteristic time window length represents the historical standard waveform. This represents the prior predicted value of the perturbation. This indicates the equivalent heat capacity of the current temperature control zone. The average temperature change rate of the historical standard waveform is represented. The larger the potential time span of the DTW warping path that is allowed to deviate from the main diagonal, the larger the corresponding maximum number of warping steps. The ratio of the equivalent thermal response time change to the temperature sampling period is calculated to obtain the theoretical offset. This invention obtains the regularized offset boundary interval, which includes a first allowable offset and a second allowable offset. The first allowable offset represents the maximum allowable regularized offset under ideal conditions with no disturbance or minimal disturbance, and is determined as follows: Under the steady-state, undisturbed condition of the heating plate, pairwise DTW matching is performed on multiple temperature response waveforms under the same condition, and the 95th percentile of the number of steps the matching path deviates from the main diagonal is taken as the first allowable offset; the second allowable offset represents the maximum allowable regularized offset under extreme disturbance conditions limited by physical laws, and is taken as 30% to 50% of the number of sampling points required for the heating plate to reach steady state from the initial temperature step response; simultaneously, 0.5 times the average absolute value of historical disturbance data is taken. To reach the low threshold of disturbance, take twice the average absolute value of historical disturbance data to obtain the high threshold of disturbance. The average absolute value of historical disturbances is calculated using a sliding window, and the window length is determined based on the number of equipment operation records. If the prior predicted value of disturbance is less than the low threshold of disturbance, the first allowable offset is used as the maximum normalized offset. If the prior predicted value of disturbance is greater than the high threshold of disturbance, the second allowable offset is used as the maximum normalized offset. If the prior predicted value of disturbance is between the low threshold of disturbance and the high threshold of disturbance, the theoretical offset is used as the basis, and the first and second allowable offsets are combined for amplitude limiting to obtain the final maximum normalized offset. This ensures that the offset reflects the change in thermal response time caused by the current disturbance without exceeding the physically achievable range. S42. The initial temperature waveform segment and the historical temperature waveform are respectively used as two sequences to be matched. A dynamic programming grid is constructed, and the straight line formed by the points with equal indices of the two sequences in the grid is used as the main diagonal. The normalized deviation magnitude is calculated for each candidate alignment point. For example, when performing dynamic time warping of physical constraints, let the initial temperature waveform segment of the current waveform be a sequence. , , length is Each sampling point, and a specific historical temperature waveform in the historical waveform library, is a sequence. , , length is Each sampling point; the traditional DTW algorithm constructs a " A grid of points, where grid points Indicates the current waveform's first... sampling points The first of the historical temperature waveforms sampling points Alignment relationships between them, the main diagonal refers to the alignment relationships in the mesh that satisfy " and The straight line formed by the points marked "" represents the ideal path for two waveforms to be perfectly aligned on the time axis. This assumes that the dynamic processes of the two waveforms are completely synchronized, without any scaling or distortion on the time axis. In the actual dynamic programming recursive process, the system considers each candidate alignment point... Calculate the magnitude of its deviation from the main diagonal. Since the lengths of the two waveforms may be different, a normalized deviation magnitude needs to be used. This value represents the equivalent number of steps the alignment point deviates from the main diagonal at the original sampling point scale. The system compares this deviation magnitude with the maximum normalization offset to determine whether the alignment point is allowed to participate in the state transition. S43. Compare the normalized deviation magnitude with the maximum normalization offset to obtain a set of candidate alignment points that are allowed to participate in the state transition. For example, in the dynamic programming recursion process, the present invention uses all grid positions that may be points traversed by the path as candidate alignment points, and the dynamic programming recursion starts from the starting point. Initially, the cumulative distance of each grid point is calculated sequentially according to time. The cumulative distance of each point depends only on its predecessor point, i.e., its left neighbor. , lower neighbor point and diagonal neighbor The cumulative distance; in unconstrained traditional DTW, all " "Each grid point can be used as a candidate alignment point in the recursion. However, with the introduction of physical constraints in this invention, when a candidate alignment point..." When the normalized deviation exceeds the maximum warping offset, the system marks the point as unreachable, sets its cumulative distance to infinity, and prohibits state transitions from that point to its subsequent neighbors, thus obtaining a set of candidate alignment points that are allowed to participate in state transitions. In this way, the search path of dynamic programming is restricted to a strip area centered on the main diagonal, and the half width of this strip area is the maximum warping offset. All points outside the strip are masked. S44. In the dynamic programming grid with applied band constraints, the cumulative distance matrix is ​​calculated according to the dynamic time warping algorithm, and the cumulative distance at the endpoint is used as the waveform similarity between the initial temperature waveform segment and the historical temperature waveform based on the cumulative distance matrix. For example, the present invention calculates the cumulative distance matrix recursively using the conventional DTW algorithm in a dynamic programming mesh with applied band constraints, ultimately reaching the endpoint. The minimum cumulative distance obtained is the waveform similarity between the current waveform and the historical temperature waveform. This waveform similarity takes into account the similarity between the two waveforms in terms of time axis scaling and amplitude difference: the smaller the distance, the more similar the two waveforms are within a physically reasonable time scaling range. Since alignment paths that deviate too much from the main diagonal are prohibited, these paths physically imply extremely unreasonable time domain scaling, such as aligning the temperature drop stage to the heating stage, the final waveform similarity has stronger physical interpretability than traditional DTW.

[0024] In one embodiment, step S51, which combines the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment to construct a dynamic similarity criterion, includes: S511. Extract high-frequency components from the original temperature signal and obtain the high-frequency noise standard deviation based on the high-frequency components; For example, the present invention extracts high-frequency components from the raw temperature signal collected by the current temperature sensor through high-pass filtering or wavelet transform, and calculates the standard deviation of the high-frequency components to obtain the high-frequency noise standard deviation. This index reflects the degree of environmental noise pollution such as electromagnetic interference and vibration of the current measurement signal. S512. Based on all the historical temperature waveforms stored in the historical waveform library, a dynamic time warping algorithm is used to obtain the dynamic time warping distance between any two historical temperature waveforms, and a waveform intrinsic difference threshold is obtained based on multiple dynamic time warping distances. For example, for all historical temperature waveforms stored in the historical waveform library, the present invention calculates the dynamic time warping distance between every two waveforms using conventional DTW, and then averages all the calculated distances as the waveform intrinsic difference threshold to measure the average difference between waveforms under normal operating conditions. S513. Obtain the preset normal operating condition noise standard deviation, and obtain the threshold amplification factor based on the high-frequency noise standard deviation, the normal operating condition noise standard deviation, and the waveform intrinsic difference threshold. For example, this invention obtains a preset normal operating condition noise standard deviation, which is a typical high-frequency noise level measured in advance during normal equipment operation. When the current high-frequency noise standard deviation increases, it indicates a decline in signal quality. At this time, even if two waveforms are physically very similar, the calculated DTW distance will be too large due to noise interference. Therefore, it is necessary to increase the judgment threshold to avoid misjudging waveforms that should match as mismatched. This invention addresses this by… Calculate the threshold amplification factor, where, Indicates the threshold amplification factor. This represents the preset sensitivity coefficient, which is tuned using a grid search method with the goal of achieving the highest accuracy in matching historical waveforms. It is used to control the amplification factor's sensitivity to noise. This represents the standard deviation of the measured high-frequency noise in the current period. This represents the standard deviation of noise under normal operating conditions. Indicates the intrinsic difference threshold of the waveform. This represents the reference distance threshold, which is the minimum self-matching DTW distance of each waveform in the historical waveform library under noise-free conditions (usually the waveform is divided into two equal segments for DTW alignment calculation), and it must be a positive number; through the formula " "and" "This shows that the degree of noise pollution on DTW distance depends not only on the intensity of the noise itself, but also on the degree of difference in the waveform itself. When the waveforms in the historical waveform library are significantly different, the relative deviation caused by noise is small, and the threshold amplification factor can be appropriately reduced. When the waveforms are highly consistent, the noise has a more significant impact on the matching results, and a larger amplification factor is needed to avoid misjudgment." S514. Obtain the dynamic similarity determination threshold based on the waveform intrinsic difference threshold and the threshold amplification factor; For example, the present invention obtains a similarity determination threshold that is dynamically adjusted according to the current noise level by multiplying the waveform intrinsic difference threshold by a threshold amplification factor.

[0025] In one embodiment, step S52, which combines the waveform similarity to perform the optimal selection of historical temperature waveforms and the coupling configuration of the fusion weights, includes: S521. Compare the minimum waveform similarity with the similarity determination threshold: If the minimum waveform similarity is greater than the similarity judgment threshold, the current load is determined to be an unknown operating condition, and a conservative control strategy is adopted. If the minimum waveform similarity is not greater than the similarity judgment threshold, it is determined that there is a matching historical working condition, and all historical temperature waveforms with waveform similarity not greater than the similarity judgment threshold are selected as reference historical temperature waveforms. For example, after performing dynamic time warping of physical constraints and obtaining the waveform similarity between the current waveform and each historical temperature waveform in the historical waveform library, the system finds the minimum waveform similarity and compares the minimum waveform similarity with a similarity judgment threshold. If the minimum waveform similarity is greater than the similarity judgment threshold, it means that the similarity between the current waveform and all waveforms in the historical library is lower than the dynamic threshold, and the current load is determined to be an unknown working condition, such as a substrate material, thickness, or new environmental conditions that have never appeared before. If the minimum waveform similarity is not greater than the similarity judgment threshold, it is determined that there is a matching historical working condition, and all historical temperature waveforms with waveform similarity not greater than the similarity judgment threshold are selected as reference historical temperature waveforms. When the current load is determined to be an unknown operating condition, the present invention adopts a conservative control strategy: reducing or completely disabling the feedforward compensation fusion weight based on historical temperature waveform matching, for example, setting the data-driven weight coefficient to zero, relying only on the feedback control of the active disturbance rejection controller and the prior prediction value of the disturbance in the current cycle for temperature adjustment, and marking the temperature response waveform of the current cycle as a state to be learned, and deciding whether to include it in the historical waveform library after the printing cycle ends and the quality is qualified. This mechanism avoids blindly using mismatched historical disturbance data under unknown operating conditions, which would lead to inaccurate control output. S522. For each of the selected reference historical temperature waveforms, obtain the data-driven weight based on the waveform similarity, and calculate the data-driven disturbance estimate based on the data-driven weight and the historical disturbance data corresponding to each of the reference historical temperature waveforms. For example, the present invention calculates data-driven weights for each selected historical temperature waveform based on its waveform similarity to the current waveform. Specifically, the calculation can use an exponential decay function or an inverse proportional function. The smaller the distance, the more similar the historical temperature waveform is to the current operating condition, and the greater the reference value of its corresponding historical disturbance data, thus the higher the weight. This invention is achieved through " "Calculate the data-driven disturbance estimate, where, This represents the data-driven disturbance estimate. Indicates the first Weights are driven by data from a reference historical temperature waveform. Indicates the first Historical disturbance data corresponding to a reference historical temperature waveform This indicates the number of reference historical temperature waveforms. Indicates the index number of the reference historical temperature waveform; S523. Obtain the historical disturbance prior value associated with each valid historical record in the historical waveform library, and obtain the feature difference degree based on multiple historical disturbance data and the historical disturbance prior value; For example, the present invention calculates the historical difference between the corresponding historical disturbance data and the historical disturbance prior value derived from the physical model of the historical temperature waveform itself for each valid historical record in the historical waveform library, and obtains a set of historical difference. Then, the median of the set is taken as the feature difference. The feature difference refers to the typical deviation between the physical model derivation value and the actual required disturbance compensation amount in the historical excellent printing cycle. S524. Obtain the physical prior fusion weights based on the perturbation prior prediction value, the data-driven perturbation estimate value, and the feature difference degree. For example, this invention calculates the absolute value of the difference between the prior predicted value of the disturbance and the estimated value of the data-driven disturbance to obtain the degree of difference between the two. This degree of difference physically characterizes the extent of energy trace deviation between the current heat transfer path and the historical experience path: when the degree of difference is small, it indicates that the current operating condition and the historical operating condition are highly consistent in thermodynamic behavior, and the physical model deduction and historical data experience corroborate each other; when the degree of difference is large, it indicates that a new disturbance mode has appeared in the current operating condition that has not been fully covered by the historical waveform library. At this time, the credibility of the physical model deduction is relatively higher than that of the historical data matching result, because the physical model contains the basic thermodynamic structure information of the equipment, while the historical data only relies on past experience. Based on this physical logic, this invention designs an inverse proportional allocation fusion function, so that the fusion weight of the physical prior is positively correlated with the degree of difference, while the fusion weight of the data-driven disturbance is negatively correlated with the degree of difference. Specifically, the physical prior fusion weight is defined as: ,in, Represents the physics prior fusion weights. Indicates the degree of difference. This represents a non-negative power coefficient used to adjust the steepness of the weight change. This value is determined based on the distribution characteristics of historical variability; it takes a small value (close to 1) when the variability distribution is concentrated, a large value (close to 4) when the distribution is dispersed, and a value of 2 when there is no prior knowledge. This represents the feature difference degree, which is defined as the median of the absolute values ​​of the differences between the historical disturbance data (i.e., the historical ideal disturbance compensation amount) and the corresponding historical disturbance prior value in each record of the historical waveform library.

[0026] In one embodiment, step S7, which involves obtaining the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correcting the thermodynamic physical model online, and quantifying the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control region, includes: S71. Obtain the steady-state disturbance increment after the current printing cycle enters a steady state, and obtain the temperature drop integral value based on the temperature response waveform. Recursively obtain the update gain based on the steady-state disturbance increment and the temperature drop integral value. For example, the present invention obtains the temperature drop integral value by integrating the temperature deviation (relative to the set temperature) during the temperature waveform drop phase in the current printing cycle over time, and obtains the stable increment of the total disturbance estimate inside the automatic disturbance rejection controller relative to the nominal value after the current printing cycle enters steady state, thus obtaining the steady-state disturbance increment. This value reflects the equivalent disturbance size actually observed and compensated by the controller in order to maintain temperature stability. The recursive least squares updater is an online parameter identification algorithm that takes the temperature drop integral value as the input observation and the steady-state disturbance increment as the output observation, and inputs it into the RLS updater. The RLS updater calculates an update gain based on the current covariance matrix. S72. Calculate the predicted steady-state disturbance increment based on the thermodynamic physical model and the temperature drop integral value, and update the thermal gain coefficient in the thermodynamic physical model based on the difference between the predicted steady-state disturbance increment and the steady-state disturbance increment, as well as the update gain amount. For example, the initial calibration of the model parameters is obtained as follows: with the heating plate at its rated operating temperature, a step heating power is applied to each temperature control zone, the temperature response curve of each zone is recorded, and the thermodynamic physical model is fitted using the least squares method to obtain... , , The initial estimate is used as the initial value for subsequent online RLS correction; This invention uses the current thermodynamic physical model to calculate a predicted value of the steady-state disturbance increment based on the integral value of the temperature drop. Then, it calculates the deviation between this predicted value and the steady-state disturbance increment, multiplies this deviation by the update gain, and obtains the correction amount for the thermal gain coefficient. This invention adds the correction amount to the current thermal gain coefficient to obtain an updated thermal gain coefficient. This updated coefficient reflects the actual thermodynamic gain characteristics of the equipment and is stored in non-volatile memory, ensuring that the latest model parameters can be loaded after a power outage and restart, without the need for re-offline calibration. When the rate of change of the thermal gain coefficient is less than a preset threshold within several consecutive printing cycles, RLS convergence is determined, and updates are stopped. The preset threshold is set by those skilled in the art based on the actual equipment characteristics using conventional methods. S73. Collect the complete temperature response waveforms of each independent temperature control zone after the current printing cycle ends, and extract features from the complete temperature response waveforms to obtain the temperature response deviation index of each zone. For example, after the current printing cycle ends, the system collects and records the complete temperature response waveform of each independent temperature control zone throughout the entire printing cycle, extracts the key temperature characteristic parameters of each temperature control zone, including but not limited to: temperature drop trough value, temperature recovery time, and temperature drop integral. Based on the above characteristic parameters, the system normalizes them and sets the weight coefficients corresponding to each characteristic parameter according to the sensitivity of the process to each parameter. Then, the temperature response deviation index is calculated by weighted summation algorithm. S74. Identify the hysteresis temperature control area based on the temperature response deviation index, and calculate the bias compensation amount based on the temperature response deviation index and the preset nominal control amount. At the beginning of the next printing cycle, add the bias compensation amount to the basic control output of the corresponding area. For example, the present invention identifies hysteresis temperature control zones based on the temperature response deviation index. Specifically, it calculates the mean and standard deviation of the temperature response deviation index for all temperature control zones. When the temperature response deviation index of a certain temperature control zone meets the following condition... When this occurs, the area is determined to be a delayed temperature control area, among which, Indicates the first Temperature response deviation index of each lagging temperature control zone Indicates the serial number of the lagging temperature control zone. This represents the mean of the temperature response deviation index. This represents the preset threshold coefficient, for example, 1.5 to 2.0. The standard deviation of the temperature response deviation index is represented by the physical meaning of the hysteresis temperature control region: the thermal response speed of this region is slower than that of other regions. Under the same control output, its temperature drops more and the recovery time is longer, resulting in a worse uniformity of the temperature field of the entire heating plate. For the identified hysteresis temperature control regions, this invention implements a differentiated bias compensation strategy. Specifically, an additional bias compensation amount is calculated for each hysteresis temperature control region. ,in, Indicates the first The offset compensation amount for each lagging temperature control zone. This represents the bias compensation coefficient, used to control the compensation intensity and avoid overcompensation leading to new non-uniformities. This value is tuned through routine experiments, with the goal of bringing the temperature response deviation index of each temperature control zone to near the mean. For example, a value of 0.3 is used. Indicates the first Temperature response deviation index of each lagging temperature control zone This represents the mean of the temperature response deviation index. It represents the nominal control quantity, used to convert the dimensionless deviation index into a control increment with physical meaning. It can be defined as the average steady-state heating power of the heating plate in each temperature control zone at the rated operating temperature. At the start of the next printing cycle, the differential bias compensation is superimposed on the basic control output of the corresponding area. It should be noted that differential bias compensation and feedforward compensation are two complementary mechanisms: feedforward compensation targets transient temperature fluctuations caused by external disturbances (such as changes in the substrate material) within each cycle and has cycle specificity; while differential bias compensation targets the inherent and long-term differences in thermal response characteristics between each temperature control zone (such as response lag caused by differences in physical structures such as heating element layout, heat dissipation boundary conditions, and sensor installation positions), and has long-term stability and accumulation. The two work together to suppress transient disturbances within the cycle and eliminate system deviations between areas at the root.

[0027] This application also provides an intelligent control system for the uniformity of the temperature field in heat transfer printing, including: The data acquisition module is used to retrieve the historical waveform library, which includes multiple historical temperature waveforms and their corresponding historical disturbance data, and to collect the original temperature signals of each independent temperature control zone of the heating plate through a temperature sensor. The waveform extraction module is used to calibrate and convert the original temperature signal to obtain a temperature response waveform, and to determine the waveform window length of each zone based on the temperature response waveform, so as to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. The prior prediction module is used to perform deduction and calculation on each of the initial temperature waveform segments based on a pre-calibrated thermodynamic physical model to obtain the perturbation prior prediction value corresponding to each region. The physical warping module is used to set the maximum warping offset based on the perturbation prior prediction value, and to perform dynamic time warping of the initial temperature waveform segments and historical temperature waveforms of each region based on the maximum warping offset to obtain waveform similarity. The dynamic filtering module is used to construct a dynamic similarity criterion by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment, and to perform optimal filtering of historical temperature waveforms and coupling configuration of fusion weights by combining the waveform similarity. The feedforward compensation module is used to perform fusion and compensation mapping based on the historical temperature waveform, disturbance prior prediction value and fusion weight corresponding to each zone to obtain the feedforward compensation amount of each zone. The feedforward compensation amount is superimposed on the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. The online correction module is used to obtain the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correct the thermodynamic physical model online, and quantify the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control area.

[0028] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent control method for the uniformity of the temperature field in heat transfer printing.

[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described intelligent control method for the uniformity of the temperature field in heat transfer printing.

[0030] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0031] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0032] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent control of temperature field uniformity in heat transfer printing, characterized in that, include: The historical waveform library is retrieved, which includes multiple historical temperature waveforms and their corresponding historical disturbance data, and the original temperature signals of each independent temperature control zone of the heating plate are collected by a temperature sensor. The original temperature signal is calibrated and converted to obtain a temperature response waveform. The waveform window length of each zone is determined based on the temperature response waveform in order to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. Based on the pre-calibrated thermodynamic physical model, the initial temperature waveform segments are extrapolated and calculated to obtain the perturbation prior prediction values ​​corresponding to each region. The maximum warping offset is set based on the perturbation prior prediction value, and dynamic time warping of the initial temperature waveform segment and the historical temperature waveform of each region is performed with physical constraints based on the maximum warping offset to obtain the waveform similarity. A dynamic similarity criterion is constructed by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment. The waveform similarity is then used to perform the optimal selection of historical temperature waveforms and the coupling configuration of fusion weights. Based on the historical temperature waveform, disturbance prior prediction value and fusion weight of each zone, fusion and compensation mapping are performed to obtain the feedforward compensation amount of each zone. The feedforward compensation amount is superimposed on the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. After the printing cycle ends, the steady-state disturbance increment and temperature drop integral of each zone are obtained. The thermodynamic physical model is iteratively corrected online, and the temperature field non-uniformity deviation of each zone is quantified to implement differentiated bias compensation for the hysteresis temperature control area.

2. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, Before the step of retrieving the historical waveform library, the following steps are included: After the end of each historical printing cycle, obtain the image detection defect rate and image detection color difference of the printed finished product to obtain the final printing quality assessment value for that cycle; The period is considered a valid learning period only if the final printing quality assessment value is higher than the preset quality qualification threshold. For the effective learning period, the temperature response waveform sequence from the moment of mold closing triggering to the moment of temperature recovery and stabilization within the period is collected and saved. Based on the temperature field steady-state uniformity result finally achieved in this cycle, the ideal perturbation compensation amount required to achieve the steady-state result is solved in reverse based on the thermodynamic physical model. The temperature response waveform sequence is associated with the ideal disturbance compensation amount and stored as a historical record in the historical waveform library, with a timestamp tag added to each record.

3. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, The step of determining the waveform window length of each zone based on the temperature response waveform, in order to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform, includes: Extract temperature response waveforms from multiple completed printing cycles from the historical waveform library, and obtain the average drop slope peak value based on the multiple temperature response waveforms. Based on the historical waveform library, the temperature drop start time and temperature recovery stabilization time of multiple historical temperature response waveforms are identified to obtain the characteristic recovery time. Statistical analysis is then performed on multiple characteristic recovery times to obtain the minimum window length and the maximum window length. The waveform window length for the current printing cycle is obtained based on the average drop slope peak value, the minimum window length, and the maximum window length. Starting from the sampling point index corresponding to the mold closing trigger moment, an initial temperature waveform segment is extracted from the temperature response waveform of the current printing cycle according to the waveform window length.

4. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, The step of setting a maximum warping offset based on the perturbation prior prediction value, and performing dynamic time warping of the initial temperature waveform segments and historical temperature waveforms of each region based on the maximum warping offset to obtain waveform similarity includes: Obtain the regularization offset boundary interval, and obtain the maximum regularization offset based on the regularization offset boundary interval and the perturbation prior prediction value; The initial temperature waveform segment and the historical temperature waveform are respectively used as two sequences to be matched. A dynamic programming grid is constructed, and the straight line formed by the points with equal indices of the two sequences in the grid is used as the main diagonal. The normalized deviation magnitude is calculated for each candidate alignment point. The normalized deviation magnitude is compared with the maximum normalization offset to obtain a set of candidate alignment points that are allowed to participate in the state transition. In the dynamic programming grid with applied band constraints, the cumulative distance matrix is ​​calculated according to the dynamic time warping algorithm, and the cumulative distance at the endpoint is used as the waveform similarity between the initial temperature waveform segment and the historical temperature waveform based on the cumulative distance matrix.

5. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, The step of constructing a dynamic similarity criterion by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment includes: High-frequency components are extracted from the original temperature signal, and the standard deviation of high-frequency noise is obtained based on the high-frequency components; Based on all the historical temperature waveforms stored in the historical waveform library, a dynamic time warping algorithm is used to obtain the dynamic time warping distance between any two historical temperature waveforms, and a waveform intrinsic difference threshold is obtained based on multiple dynamic time warping distances. Obtain the preset normal operating condition noise standard deviation, and obtain the threshold amplification factor based on the high-frequency noise standard deviation, the normal operating condition noise standard deviation, and the waveform intrinsic difference threshold. The dynamic similarity determination threshold is obtained based on the waveform intrinsic difference threshold and the threshold amplification factor.

6. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, The steps of combining the waveform similarity to perform the optimal selection of historical temperature waveforms and the coupled configuration of the fusion weights include: The minimum waveform similarity is compared with the similarity determination threshold: If the minimum waveform similarity is greater than the similarity judgment threshold, the current load is determined to be an unknown operating condition, and a conservative control strategy is adopted. If the minimum waveform similarity is not greater than the similarity judgment threshold, it is determined that there is a matching historical working condition, and all historical temperature waveforms with waveform similarity not greater than the similarity judgment threshold are selected as reference historical temperature waveforms. For each of the selected reference historical temperature waveforms, a data-driven weight is obtained based on the waveform similarity, and a data-driven perturbation estimate is calculated based on the data-driven weight and the historical perturbation data corresponding to each of the reference historical temperature waveforms. Obtain the historical perturbation prior value associated with each valid historical record in the historical waveform library, and obtain the feature difference degree based on multiple historical perturbation data and the historical perturbation prior value; The physical prior fusion weights are obtained based on the perturbation prior prediction, the data-driven perturbation estimate, and the feature difference.

7. The intelligent control method for temperature field uniformity in heat transfer printing according to claim 1, characterized in that, The steps of obtaining the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correcting the thermodynamic physical model online, and quantifying the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control area include: The steady-state disturbance increment after the current printing cycle enters a steady state is obtained, and the temperature drop integral value is obtained based on the temperature response waveform. The update gain is then recursively obtained based on the steady-state disturbance increment and the temperature drop integral value. The steady-state disturbance increment prediction value is calculated based on the thermodynamic physical model and the temperature drop integral value. The thermal gain coefficient in the thermodynamic physical model is updated based on the difference between the steady-state disturbance increment prediction value and the steady-state disturbance increment, as well as the update gain amount. Collect the complete temperature response waveforms of each independent temperature control zone after the current printing cycle ends, and extract features from the complete temperature response waveforms to obtain the temperature response deviation index of each zone; The temperature response deviation index is used to identify the hysteresis temperature control area, and the bias compensation amount is calculated based on the temperature response deviation index and the preset nominal control amount. At the beginning of the next printing cycle, the bias compensation amount is superimposed on the basic control output of the corresponding area.

8. An intelligent control system for the uniformity of temperature field in heat transfer printing, characterized in that, include: The data acquisition module is used to retrieve the historical waveform library, which includes multiple historical temperature waveforms and their corresponding historical disturbance data, and to collect the original temperature signals of each independent temperature control zone of the heating plate through a temperature sensor. The waveform extraction module is used to calibrate and convert the original temperature signal to obtain a temperature response waveform, and to determine the waveform window length of each zone based on the temperature response waveform, so as to extract the initial temperature waveform segments of each zone in the current printing cycle from the temperature response waveform. The prior prediction module is used to perform deduction and calculation on each of the initial temperature waveform segments based on a pre-calibrated thermodynamic physical model to obtain the perturbation prior prediction value corresponding to each region. The physical warping module is used to set the maximum warping offset based on the perturbation prior prediction value, and to perform dynamic time warping of the initial temperature waveform segments and historical temperature waveforms of each region based on the maximum warping offset to obtain waveform similarity. The dynamic filtering module is used to construct a dynamic similarity criterion by combining the historical waveform library and the operating condition signal features extracted based on the initial temperature waveform segment, and to perform optimal filtering of historical temperature waveforms and coupling configuration of fusion weights by combining the waveform similarity. The feedforward compensation module is used to perform fusion and compensation mapping based on the historical temperature waveform, disturbance prior prediction value and fusion weight corresponding to each zone to obtain the feedforward compensation amount of each zone. The feedforward compensation amount is superimposed on the control output of the active disturbance rejection controller to drive the heating element to perform temperature regulation. The online correction module is used to obtain the steady-state disturbance increment and temperature drop integral of each zone after the printing cycle ends, iteratively correct the thermodynamic physical model online, and quantify the temperature field non-uniformity deviation of each zone to implement differentiated bias compensation for the hysteresis temperature control area.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images