A multimodal temperature control method and system for a three-in-one laminator

Abstract

This application provides a multimodal temperature control method and system for a three-in-one laminator, comprising: using timestamp alignment technology to synchronously acquire data from temperature, pressure, and vision sensors to ensure the temporal consistency of multimodal data; processing the raw data from temperature and pressure sensors using a Gaussian filtering algorithm to filter out high-frequency noise and obtain smoothed sensor data; employing an adaptive illumination compensation algorithm for the image data acquired by the vision system to eliminate the influence of ambient light changes on film detection; using polarization filtering technology to preprocess the image in the reflective area of ​​the film to reduce reflective interference and improve the accuracy of visual detection; inputting the filtered temperature and pressure data and the processed visual data into a preset data fusion model to generate a multimodal feature vector; and optimizing the fuzzy control rules using a genetic algorithm, using historical temperature and pressure data as training samples to generate the optimal control strategy.

Description

Technical Field

[0001] This invention relates to the field of information technology, specifically to control systems for office equipment, and more particularly to a multimodal temperature control method and system for a three-in-one laminator. Background Technology

[0002] The 3-in-1 laminating and sealing machine, as an advanced office device integrating multiple functions, combines lamination, coating, and hot stamping capabilities, and innovatively supports composite processes such as embossing, hot stamping, and subsequent lamination. The machine operates based on the synergistic effect of a precision thermistor and a high-efficiency heating element system. Through an intelligent temperature control mechanism and system, it can precisely adjust to the required temperature range for different operations, thus flexibly achieving the aforementioned functions. This technology not only greatly improves office efficiency and flexibility but also provides users with richer and more diverse document enhancement and protection methods, representing a significant technological innovation in the field of modern office automation. Specifically, the system integrates data from multiple sensors, such as temperature and pressure, along with visual and auditory information, to achieve precise control of the lamination process.

[0003] However, in practical applications, the acquisition and fusion of multimodal data faces a series of technical challenges. First, the sampling frequencies and data formats of different sensors vary, making synchronous acquisition and real-time processing a primary concern. Second, when detecting the uniformity of adhesive film coverage, vision systems may be affected by factors such as changes in ambient light and the reflectivity of the adhesive film material, leading to inaccurate results. Furthermore, data from temperature and pressure sensors may contain noise during dynamic changes; effectively filtering out noise and extracting useful information is another technical challenge. At the control algorithm level, while combining multiple algorithms can improve the system's adaptability, the coordination of parameters and weight allocation between different algorithms is difficult to determine. For example, PID control performs well in steady-state control but may not respond promptly in rapidly changing dynamic processes; while fuzzy control algorithms have advantages in handling uncertainty, their rule formulation and optimization are complex. These intertwined technical contradictions make it challenging for multimodal temperature control systems to achieve automatic adhesive application control. Summary of the Invention

[0004] This invention provides a multimodal temperature control method for a three-in-one laminator, mainly including the following steps:

[0005] S101. Timestamp alignment technology is used to synchronously collect data from temperature, pressure and vision sensors to ensure the temporal consistency of multimodal data;

[0006] S102. The raw data of the temperature and pressure sensors are processed by Gaussian filtering algorithm to filter out high-frequency noise and obtain smoothed sensor data.

[0007] S103. For the image data acquired by the vision system, an adaptive illumination compensation algorithm is adopted to eliminate the influence of changes in ambient light on the film detection.

[0008] S104. In the reflective area of ​​the film, polarization filtering technology is used to preprocess the image to reduce reflective interference and improve the accuracy of visual inspection.

[0009] S105. Input the filtered temperature and pressure data and the processed visual data into the preset data fusion model to generate a multimodal feature vector;

[0010] S106. A fuzzy PID control algorithm is adopted to dynamically adjust the control parameters based on the multimodal feature vector. If the temperature change exceeds the preset threshold, the heating power is adjusted first.

[0011] S107. In the fuzzy control rules, the weights of the PID control algorithm are dynamically allocated by combining the changing trends of temperature and pressure to ensure that the system can respond quickly in both steady-state and dynamic processes.

[0012] S108. Optimize the fuzzy control rules using a genetic algorithm, and generate the optimal control strategy using historical temperature and pressure data as training samples.

[0013] S109. Deploy the optimized control strategy to the temperature control system, monitor the changing trend of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

[0014] This invention provides a multimodal temperature control system for a three-in-one laminator, mainly comprising:

[0015] The timestamp alignment module is used to synchronously acquire data from temperature, pressure, and vision sensors using timestamp alignment technology, ensuring the temporal consistency of multimodal data.

[0016] The Gaussian filter module is used to process the raw data from temperature and pressure sensors using a Gaussian filtering algorithm, filtering out high-frequency noise and obtaining smoothed sensor data.

[0017] The adaptive illumination compensation module is used to eliminate the influence of changes in ambient light on film detection by employing an adaptive illumination compensation algorithm on the image data acquired by the vision system.

[0018] The polarization filtering module is used to preprocess the image in the reflective area of ​​the film using polarization filtering technology, thereby reducing reflective interference and improving the accuracy of visual inspection.

[0019] The data fusion module is used to input the filtered temperature and pressure data and the processed visual data into a preset data fusion model to generate multimodal feature vectors.

[0020] The fuzzy PID control module is used to dynamically adjust control parameters based on multimodal feature vectors using a fuzzy PID control algorithm. If the temperature change exceeds a preset threshold, the heating power is adjusted first.

[0021] The weight allocation module is used to dynamically allocate the weights of the PID control algorithm in the fuzzy control rules by combining the changing trends of temperature and pressure, so as to ensure that the system can respond quickly in both steady-state and dynamic processes.

[0022] The genetic optimization module is used to optimize fuzzy control rules using a genetic algorithm, generating the optimal control strategy using historical temperature and pressure data as training samples.

[0023] The real-time control module is used to deploy the optimized control strategy to the temperature control system, monitor the changing trends of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

[0024] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0025] This invention discloses a multimodal temperature control method and system for a three-in-one laminator. The method achieves synchronous acquisition of temperature, pressure, and visual sensor data through timestamp alignment technology, and preprocesses the raw data using algorithms such as Gaussian filtering and adaptive illumination compensation. To address the issue of film reflection, polarization filtering technology is employed to improve the accuracy of visual inspection. The processed multimodal data is input into a preset fusion model to generate feature vectors, and control parameters are dynamically adjusted using a fuzzy PID control algorithm. The fuzzy control rules are optimized through a genetic algorithm to achieve real-time dynamic adjustment of temperature and pressure parameters. This method effectively solves problems such as multimodal data synchronization, environmental interference, and control accuracy in traditional film inspection, significantly improving the accuracy of film inspection and the control effect of the lamination process, providing an innovative solution for the field of intelligent manufacturing. Attached Figure Description

[0026] Figure 1 This is a flowchart of a multimodal temperature control method for a three-in-one laminator according to the present invention.

[0027] Figure 2 This is a schematic diagram of a multimodal temperature control method and system for a three-in-one gluing and sealing machine according to the present invention.

[0028] Figure 3 This is another schematic diagram of a multimodal temperature control method and system for a three-in-one glue sealing machine according to the present invention.

[0029] Figure 4 This is a schematic diagram of the multimodal temperature control method and system for a three-in-one glue sealing machine according to the present invention. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0031] like Figure 1-4 The multimodal temperature control method for a three-in-one laminator in this embodiment may specifically include:

[0032] S101. Timestamp alignment technology is used to synchronously collect data from temperature, pressure and vision sensors to ensure the temporal consistency of multimodal data.

[0033] Raw data from temperature, pressure, and vision sensors are acquired, and timestamp information for each data point is recorded. Based on the timestamp information, the temperature, pressure, and vision data are initially aligned to generate a preliminary aligned dataset. For the preliminary aligned dataset, an interpolation algorithm is used to supplement missing timestamp data, resulting in a complete aligned dataset. The complete aligned dataset is then acquired, and it is determined whether timestamp discrepancies exist. If timestamp discrepancies exist, a time compensation algorithm is used to correct them, resulting in a corrected complete aligned dataset. Based on the timestamp information contained in the corrected complete aligned dataset, the temperature, pressure, and vision data are reordered to generate a time-consistent multimodal dataset. A denoising algorithm is used to denoise the time-consistent multimodal dataset, resulting in an optimized multimodal dataset. Finally, a feature extraction algorithm is used to extract temperature, pressure, and vision features from the optimized multimodal dataset, generating a multimodal feature dataset.

[0034] For example, multimodal sensor data acquisition targets monitoring data provided by temperature sensors, pressure gauges, and cameras in a thermal piping system. During the acquisition process, each sensor records data at a different frequency. The temperature sensor acquires three readings per second, the pressure gauge acquires two readings per second, and the camera acquires an image once per second. To achieve data alignment, all data points are first timestamped to milliseconds. During initial alignment, the temperature and pressure data are matched according to the most recent timestamp, based on the acquisition frequency of the visual data. For instance, in pipe leak detection, the data within a single second includes three temperature readings: -13°C, -12.8°C, and -12.5°C (Fahrenheit conversion), two pressure readings: 5.2 and 5.3 atmospheres, and an image of the pipe surface. Through initial alignment, the temperature value of -12.5°C and the pressure value of 5.3 atmospheres closest to the image acquisition time can be associated with the image. For data loss due to sensor malfunctions, linear interpolation methods can be used to supplement the data. Suppose a pressure sensor fails to collect data at a certain moment. Given that the pressure at the previous moment was 5.3 atmospheres and the pressure at the next moment was 5.5 atmospheres, the pressure value at the missing moment can be interpolated to 5.4 atmospheres. In actual operation, timestamp deviations may occur; for example, the temperature sensor's clock may be 0.03 seconds ahead of the standard time. By analyzing data variation patterns, such systematic deviations can be identified, and timestamps can be compensated accordingly to ensure the temporal consistency of multimodal data. Time-consistent data is more conducive to discovering the correlations between various physical quantities. For random noise in the data, median filtering can be used for noise reduction. For example, if temperature data shows abrupt changes, jumping from -12.5 degrees Celsius to -20 degrees Celsius and then back to -13 degrees Celsius, comparing adjacent data points can identify noise and correct it. In the feature extraction stage, statistical features such as mean and variance can be extracted from temperature data, trend features from pressure data, and features such as texture and color of the pipe surface can be extracted from visual data. These features collectively constitute a multimodal feature set reflecting the state of the pipeline system, providing data support for subsequent pipeline health assessment and fault early warning. Through multimodal data analysis, potential problems such as abnormal temperature, pressure fluctuations, or changes in the pipeline surface can be detected in a timely manner, enabling comprehensive monitoring of the pipeline system.

[0035] S102. The raw data of the temperature and pressure sensors are processed by Gaussian filtering algorithm to filter out high-frequency noise and obtain smoothed sensor data.

[0036] The process involves acquiring raw values ​​from temperature and pressure sensors, preprocessing these values ​​to remove outliers and missing values, and generating standardized data. Based on preset Gaussian filtering algorithm parameters, the size and standard deviation of the filter kernel are determined, generating a Gaussian filter kernel matrix. The standardized data is then convolved with the Gaussian filter kernel matrix to obtain preliminary filtered data. This preliminary filtered data is analyzed; if edge effects exist, boundary padding is used to obtain edge-effect-free filtered data. The noise removal effect is calculated by comparing the filtered data with the raw data. If the noise removal effect does not reach a preset threshold, the Gaussian filter kernel parameters are adjusted, and the filtered data is regenerated. The filtered data is converted into smoothed values, and the smoothed values ​​undergo data format conversion to generate final smoothed data. The final smoothed data is then stored, generating a smoothed sensor data file.

[0037] For example, sensor data processing first requires acquiring the raw values ​​from temperature and pressure sensors. Sensors typically collect data at a fixed frequency, such as ten times per second for temperature sensors and twenty times per second for pressure sensors. Common outliers in the raw data include values ​​exceeding the measurement range. For instance, a temperature sensor's range is 0 to 100 degrees Celsius; negative values ​​or values ​​exceeding 100 degrees Celsius are considered outliers. Missing values ​​are usually caused by temporary sensor malfunctions or communication interruptions and can be filled by averaging adjacent data points. Data standardization unifies data with different dimensions to the same scale. For example, temperature data ranges from 0 to 100 degrees Celsius, and pressure data ranges from 0 to 1 kPa. Standardization maps both types of data to the 0 to 1 range, facilitating subsequent processing. Gaussian filtering is a commonly used data smoothing method, its core being the use of a Gaussian function as weights for weighted averaging. The size of the Gaussian filter kernel determines the smoothing strength; a larger kernel produces a smoother effect but may lose detail, while a smaller kernel retains more detail but has a weaker noise reduction effect. In practical applications, assuming a 5x5 Gaussian filter kernel with a standard deviation of 1.5, this parameter setting is suitable for processing data with moderate noise levels. Edge effects refer to distortion at the edges of a data sequence due to the lack of a complete filtering window. A common approach is boundary mirroring, which extends the boundary data outwards in a mirror manner to ensure complete filtering calculations are performed at the edges. The effectiveness of noise reduction can be measured by the improvement in signal-to-noise ratio (SNR). For example, if the original data has an SNR of 10 dB, filtering can improve it to 15 dB, which is satisfactory if the preset threshold is 12 dB. If the preset threshold is not met, the filter kernel size can be increased or the standard deviation adjusted. After smoothing, the data needs to be formatted, converting floating-point numbers to fixed-point numbers or adjusting the data bit width to suit the storage system requirements. The storage format of smoothed data needs to include timestamps, data values, and quality labels. Timestamps are accurate to milliseconds, data values ​​are selected according to the actual precision requirements, and quality labels are used to identify whether the data is the original value or an estimated value after interpolation. This structured data storage facilitates subsequent data analysis and feature extraction, and also makes it easy to trace the data processing process. After these processing steps, sensor data with low noise levels and continuous time sequence can be obtained, providing a reliable data foundation for subsequent applications such as condition monitoring and fault diagnosis.

[0038] S103. For the image data acquired by the vision system, an adaptive illumination compensation algorithm is adopted to eliminate the influence of changes in ambient light on the film detection.

[0039] A film image is acquired and preprocessed to remove noise interference, resulting in a preprocessed image. The average brightness value of the preprocessed image is calculated to determine the illumination intensity. If the illumination intensity exceeds a preset threshold, an adaptive illumination compensation algorithm is activated for adjustment. Histogram equalization is used to enhance the contrast of the preprocessed image. Gamma correction is applied to adjust the brightness distribution of the preprocessed image, resulting in an optimized image. Based on the optimized image, film defect detection is performed to identify abnormal areas. A film quality report is generated based on the film defect detection results, and the final processing result is output.

[0040] For example, preprocessing of film images primarily targets noise and interference in the image. In a film production workshop, due to changes in ambient light and equipment vibration, the acquired images may contain salt-and-pepper noise, Gaussian noise, etc. Median filtering can remove salt-and-pepper noise, while mean filtering is used for Gaussian noise, resulting in a clearer image. Average brightness calculation is a crucial indicator for evaluating light intensity. For instance, under standard lighting conditions, the average brightness of a normal film image should remain within a reasonable range. If the measured average brightness of a film image is 250, far exceeding the preset threshold of 180, it indicates excessive light intensity, requiring compensation. Adaptive illumination compensation algorithms can make local adjustments based on the brightness characteristics of different areas. For example, the edge areas of the film may be darker due to uneven lighting, while the central area may be brighter. The algorithm will enhance the brightness of dark areas and suppress bright areas, resulting in a more uniform overall brightness distribution. Through block processing, the image is divided into several sub-regions, and compensation coefficients are calculated independently for each region. Histogram equalization technology enhances image contrast by redistributing pixel grayscale values. For example, if the grayscale values ​​of a certain film image are too concentrated in low-brightness areas, making details difficult to distinguish, equalization processing can make the grayscale value distribution more uniform, making defect features easier to identify. Gamma correction can adjust the overall brightness distribution of the image. When the gamma value is less than one, it enhances details in dark areas; when it is greater than one, it enhances details in bright areas. For example, setting a gamma value of 0.8 can make film defects in shadow areas more clearly visible. Film defect detection is based on feature extraction and analysis of the processed image. Common defects such as bubbles and scratches will form specific grayscale value variation patterns in the image. By setting an appropriate detection threshold, such as a grayscale value difference exceeding 50 in a bubble area, it is determined to be a defect. Combining morphological features, such as area and perimeter, can further confirm the defect type. The quality report generation integrates various detection results. The report content includes specific parameters such as defect location coordinates, defect type, and area size. Through a quantitative scoring mechanism, different types of defects are assigned weights according to their severity to calculate the overall quality score. For example, a bubble defect has a weight of 0.8, a scratch has a weight of 0.6, and a final score below 80 is considered a defective product.

[0041] S104. In the reflective areas of the film, polarization filtering technology is used to preprocess the image to reduce reflective interference and improve the accuracy of visual inspection.

[0042] A raw image with a reflective area from the adhesive film is acquired; the raw image is processed using polarization filtering technology to obtain a de-reflective image; based on a preset threshold, it is determined whether there is a residual reflective area in the de-reflective image; if there is a residual reflective area in the de-reflective image, the polarization angle is adjusted, and the raw image is re-polarized filtered based on the adjusted polarization angle; the contrast and clarity of the de-reflective image are improved using an image enhancement algorithm; the feature contours in the de-reflective image are extracted using an edge detection algorithm; and the position and boundary of the adhesive film area in the raw image are determined based on the feature contours.

[0043] For example, in the process of film inspection, polarization filtering technology can effectively eliminate interference caused by light reflection. The principle of polarization filtering technology is based on the polarization characteristics of light; it filters out reflected light from specific directions by adjusting the angle of the polarizer. When light shines on the film surface, two phenomena occur: specular reflection and diffuse reflection. Specular reflection light usually exhibits strong polarization characteristics. By installing a polarizing filter in front of the camera lens, these reflected lights can be selectively blocked. Reflective areas on the film surface are usually high-brightness areas; for example, in an environment with an illuminance of 1000 lumens, the grayscale value of the reflective area may reach over 240. After applying polarization filtering, the grayscale value of these areas can be reduced to around 120, significantly reducing the brightness difference between them and the surrounding areas. When residual reflective areas are detected, the optimal polarization angle can be found by rotating the polarizer by 15 degrees each time. During image enhancement, histogram equalization is used to improve image contrast. For areas with concentrated grayscale distribution in the original image, the pixel grayscale range is stretched to make dark details clearer. For example, when the main grayscale values ​​of an image are concentrated between 50 and 150, equalization processing can expand them to the full range of 0 to 255, thereby highlighting the subtle features of the film surface. Edge detection employs a multi-scale method, combining the advantages of the Sobel and Cannibal operators. First, the Sobel operator is used for preliminary edge extraction to obtain the main features of the film contour. Then, the Cannibal operator is applied for refinement, setting dual thresholds of 30 (low threshold) and 90 (high threshold) to effectively extract complete edge information. When determining the film position and boundary, a region growing algorithm combined with morphological processing is used. Starting from pre-labeled seed points, the region is expanded using a similarity criterion. A grayscale similarity threshold of ±10 grayscale levels is set to ensure the continuity of region growing. Morphological processing uses circular structuring elements with a radius of five pixels. First, opening operations are performed to remove noise, and then closing operations are used to fill small holes, finally obtaining the complete film region. Through the combined application of these techniques, the position and boundary of the film can be accurately located. For example, in actual inspection, for a detection area of ​​1000 by 1000 pixels, the system can control the boundary positioning accuracy within plus or minus two pixels. This high-precision boundary extraction provides a reliable foundation for subsequent defect detection, effectively improving the overall inspection accuracy. Furthermore, by analyzing the geometric features of the boundary, such as roundness and smoothness, the quality condition of the adhesive film edge can also be determined.

[0044] S105. Input the filtered temperature and pressure data and the processed visual data into the preset data fusion model to generate a multimodal feature vector.

[0045] The process involves acquiring raw temperature data, raw pressure data, and raw visual data; denoising the raw temperature and pressure data using a filtering algorithm to obtain filtered temperature and pressure data; denoising and feature extraction of the raw visual data using an image processing algorithm to obtain processed visual data; inputting the filtered temperature, pressure, and visual data into a preset data fusion model; weighting and fusing the filtered temperature, pressure, and visual data according to a preset weight allocation rule; adjusting the weight ratio of the processed visual data if the data fusion model detects an anomaly in the filtered temperature or pressure data; generating multimodal feature vectors using a multi-layer neural network structure of the data fusion model; and classifying the multimodal feature vectors using a clustering algorithm to obtain the final feature vector categories.

[0046] For example, filtering algorithms are commonly used noise reduction methods in data processing. In a rubber production line, raw data collected by temperature sensors may be affected by environmental interference, resulting in numerical fluctuations. Median filtering can effectively remove sudden noise. For instance, if the raw temperature data is a set of discrete values ​​(185, 190, 192, 188, 187 degrees Celsius), median filtering yields a smoother data sequence. Pressure data can also be optimized using Kalman filtering. For example, in the process of pressure sensors detecting rubber compression, unstable pressure values ​​can be corrected to reliable data through prediction and correction stages. In visual data processing, Gaussian filtering can reduce image noise and improve edge feature extraction. Taking rubber surface defect detection as an example, edge detection algorithms can accurately identify defects such as cracks and bubbles on the product surface, providing reliable visual features for subsequent data fusion. Data fusion models integrate multi-source data through weight allocation. For example, in the rubber vulcanization process, the weight of temperature data can be set to 0.4, pressure data to 0.3, and visual data to 0.3. When an abnormal temperature is detected, such as exceeding 200 degrees Celsius, the system automatically reduces the weight of temperature data while increasing the weight of visual data to ensure the accuracy of the fusion result. Multilayer neural networks play a crucial role in feature extraction. Taking a three-layer network structure as an example, the input layer receives temperature, pressure, and visual data, the hidden layer performs feature transformation, and the output layer generates a fused feature vector. This structure can learn the potential correlations between data, such as material deformation characteristics caused by temperature increases. When clustering algorithms classify feature vectors, density-based clustering methods can be used. For example, in rubber product quality inspection, feature vectors are classified into three categories: normal, minor defects, and severe defects. By calculating the Euclidean distance between feature vectors, the category of a sample is determined, thus achieving accurate product quality grading. The advantage of multimodal data fusion lies in its ability to comprehensively utilize different types of sensor data. For instance, when a vision system struggles to detect minute surface cracks due to lighting conditions, temperature data may reveal localized temperature anomalies, and pressure data may show uneven stress distribution. Through data fusion, these features collectively point to quality problems in the product, improving the accuracy and reliability of the detection. In practical applications, the synergistic effect of this multi-source data can significantly reduce the false negative rate and improve the level of product quality control.

[0047] S106. A fuzzy PID control algorithm is adopted to dynamically adjust the control parameters based on the multimodal feature vector. If the temperature change exceeds the preset threshold, the heating power is adjusted first.

[0048] The system acquires the current temperature value and a preset temperature value, and calculates the temperature change range between the two values. It then compares this temperature change range with a preset threshold to determine if the range exceeds the threshold. If the range exceeds the threshold, it acquires a multimodal feature vector related to temperature control. Based on this feature vector, it employs a fuzzy PID control algorithm to obtain dynamically adjusted control parameters. During the adjustment of these parameters, the heating power parameter is adjusted first. Based on the adjusted control parameters, the heating power value is determined. Finally, based on the heating power value, a new temperature control result is obtained.

[0049] For example, in temperature control, obtaining the difference between the current temperature and the preset temperature is a fundamental step. In an industrial furnace temperature control system, assuming the preset temperature is 800 degrees Celsius and the current temperature is 795 degrees Celsius, the system calculates a temperature change of 5 degrees Celsius. The calculation of this temperature change can use a sliding window method, continuously sampling multiple data points to reduce the impact of instantaneous fluctuations. The setting of the preset threshold needs to consider the specific application scenario. Taking ceramic firing as an example, the threshold can be set to 10 degrees Celsius during the heating stage, while the holding stage requires stricter control, and the threshold can be set to 3 degrees Celsius. When the temperature change exceeds the threshold, the system triggers the acquisition of multimodal feature vectors. These multimodal feature vectors contain multi-dimensional data such as temperature, pressure, and visual data. In glass annealing, the feature vectors not only contain temperature data but also information such as the stress distribution image features on the glass surface and the furnace pressure. These data are normalized to form a unified feature vector, which is then used for subsequent fuzzy PID control. The fuzzy PID control algorithm dynamically adjusts the proportional, integral, and derivative parameters through fuzzy rules. In the heat treatment of steel, when a rapid temperature rise is detected, the system reduces the proportional gain to avoid overshoot. Conversely, when the temperature deviates slowly from the target value, the weight of the integral term is increased to eliminate steady-state error. Prioritizing the adjustment of heating power parameters is reflected in the control strategy. Taking an industrial resistance furnace as an example, when the temperature is below the target value and the rate of change is small, the system prioritizes increasing the heating power rather than adjusting other parameters. Typical power adjustment ranges from 30% to 80%, avoiding drastic temperature fluctuations due to excessive power. New temperature control results form a complete control cycle through closed-loop feedback. In semiconductor wafer annealing processes, the system samples the temperature every 0.5 seconds and continuously optimizes the control parameters based on the deviation between the actual and target temperatures. This dynamic adjustment mechanism adapts to different process stages, ensuring accurate temperature control. The overall control effect can be evaluated through the stability of the temperature curve. For example, in precision heat treatment equipment, a 2% overshoot is allowed during the heating phase, while temperature fluctuations are required to be controlled within ±1 degree Celsius during the isothermal phase. The system continuously monitors temperature change trends, predicts potential temperature fluctuations, and adjusts control parameters in advance to achieve precise temperature control.

[0050] S107. In the fuzzy control rules, the weights of the PID control algorithm are dynamically allocated by combining the changing trends of temperature and pressure to ensure that the system can respond quickly in both steady-state and dynamic processes.

[0051] The system acquires temperature values ​​from a temperature sensor and pressure values ​​from a pressure sensor, and calculates the changes in temperature and pressure values. Using time series analysis, it obtains temperature and pressure trend values ​​based on the temperature and pressure changes, respectively, and establishes a trend prediction model. Based on the trend prediction model, it determines whether the system is in a steady-state or dynamic state, and defines a control value range. The control value range is mapped to a preset fuzzy value space, and fuzzy inference is performed on the fuzzy values ​​in the fuzzy value space using preset fuzzy rules to obtain fuzzy inference results. Based on the fuzzy inference results, the weights of the proportional, integral, and derivative terms in the PID control algorithm are dynamically adjusted. The adjusted weights are applied to the PID control algorithm to generate system execution instructions. The system execution instructions are used to adjust the actuators so that the system output reaches the expected response value.

[0052] For example, in a temperature and pressure sensor acquisition system, temperature measurement can use thermocouple sensors to collect temperature values ​​in real time during the control process. For instance, in a boiler system, when the steam temperature rises from room temperature to operating temperature, temperature data is collected every 0.5 seconds. Pressure measurement can use diaphragm pressure sensors to continuously monitor system pressure. For example, in a chemical reactor, as the pressure rises from 0.3 MPa to 1.5 MPa, pressure data is collected every second. Time series analysis of the collected temperature and pressure data can be performed using a moving average method to obtain trend values. If the temperature shows a stable upward trend over the past five minutes, rising from 85 degrees to 90 degrees, it can be predicted that the temperature will continue to rise to around 95 degrees within the next five minutes. Similarly, pressure trend analysis is performed; if the pressure fluctuation is detected to be less than 0.1 MPa recently, the system is considered to be in a steady-state. In system state judgment, steady state means that both the temperature and pressure change rates are within preset ranges, such as a temperature change rate of less than one degree per minute and a pressure change rate of less than 0.05 MPa per minute. Dynamic state indicates that the system parameters are changing rapidly, requiring a faster control response. The control value range can be set to allowable temperature deviation of ±1 degree and allowable pressure deviation of ±0.02 MPa in steady state. In the fuzzy value space mapping, temperature deviation can be divided into five levels: negative large, negative small, zero, positive small, and positive large, each corresponding to a different membership function. For example, when the temperature deviation is negative 2 degrees, the membership degree for the negative large level is 0.8, and the membership degree for the negative small level is 0.2. The fuzzy rule can be set as follows: when the temperature deviation is negative large and the pressure deviation is positive small, increase the proportional coefficient. Regarding the dynamic adjustment of PID control parameters, the proportional coefficient can be appropriately increased in the initial heating stage to accelerate the response speed, such as adjusting the proportional coefficient from the original value of 1.5 to 2.0. When approaching the target value, the proportional coefficient is decreased to avoid overshoot, while the integral coefficient is appropriately increased to eliminate steady-state error. For sudden disturbances, the derivative coefficient can be temporarily increased to improve the system's ability to suppress disturbances. In actuator adjustment, system control can be achieved by adjusting the heating power and the opening of the flow valve. For example, when the temperature is low, the controller outputs 70% of the heating power; when it detects that the pressure is about to exceed the limit, it automatically adjusts the flow valve opening to 40% to prevent the pressure from continuing to rise. This coordinated control ensures the safe and stable operation of the system.

[0053] S108. Optimize the fuzzy control rules using a genetic algorithm, and generate the optimal control strategy using historical temperature and pressure data as training samples.

[0054] Historical temperature and pressure data are acquired and preprocessed to remove outliers and missing values, resulting in preprocessed historical temperature and pressure data. Based on this preprocessed data, a fuzzy control rule base is established, fuzzy control rule parameters are initialized, and an initial control strategy set is generated. A genetic algorithm is used to encode the fuzzy control rule parameters, construct a fitness function, and calculate the fitness value of each control strategy in the initial control strategy set. A selection operation is performed on the initial control strategy set based on the fitness values, retaining high-fitness control strategies and eliminating low-fitness ones. Crossover and mutation operations are performed on the retained control strategies to obtain a new control strategy set. The selection, crossover, and mutation operations are iteratively executed until the fitness value reaches a preset threshold or the number of iterations reaches a preset upper limit. The control strategy with the highest fitness value is output as the optimal control strategy, and this optimal control strategy is applied to the temperature and pressure control scenario.

[0055] For example, the acquisition and preprocessing of historical data is crucial for temperature and pressure control systems. Taking an industrial boiler as an example, if a temperature sensor collects data once per second, continuous monitoring for one hour will generate 3,600 data points. The moving average method can eliminate random fluctuations in temperature data, and sub-zero temperatures can be considered outliers and discarded. Similarly, outliers in pressure data exceeding the boiler's pressure limit must also be discarded. The establishment of a fuzzy control rule base needs to consider the mutual influence of temperature and pressure. Taking a thermal system as an example, increased temperature leads to increased pressure, which in turn affects the rate of temperature change. The rate of temperature change and the rate of pressure change can be set as input variables, divided into five levels: negative high, negative low, zero, positive low, and positive high, with the actuator's adjustment amount as the output variable. Genetic algorithm encoding can use real-number encoding to transform the membership function parameters in the fuzzy rules into chromosomes. Taking a triangular membership function as an example, each membership function is determined by three parameters. If there are fifteen membership functions in total for the input and output variables, then each chromosome consists of forty-five real numbers. The design of the fitness function needs to comprehensively consider the system response time and overshoot. Taking the temperature control of a chemical reactor as an example, the fitness value can be the weighted sum of the system rise time, settling time, and overshoot. The weighting coefficients can be determined according to the actual process requirements. If the process has strict requirements on overshoot, the weight of overshoot in the fitness calculation should be increased. The selection operation of the genetic algorithm adopts the roulette wheel selection method, where individuals with higher fitness values ​​have a higher probability of being selected. The crossover operation can use arithmetic crossover, which linearly combines two parent chromosomes in a certain proportion to produce offspring. The mutation operation applies small-amplitude random perturbations to certain genes in the chromosomes, which helps to escape local optima. The practical application of the optimal control strategy requires online fine-tuning. Taking the temperature control of a plastic extruder as an example, the control strategy obtained through offline optimization is used in the initial stage. After running for a period of time, fine-tuning is performed based on the actual control effect. When the temperature fluctuation is large, the weight of the negative feedback is increased; when the pressure fluctuation is large, the weight of the pressure compensation term is increased to ensure that the system can maintain good control performance under different operating conditions. In this way, coordinated control of temperature and pressure can be achieved, improving the overall performance of the system.

[0056] S109. Deploy the optimized control strategy to the temperature control system, monitor the changing trend of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

[0057] The system acquires multimodal data from the temperature control system, including temperature and pressure parameters. For this multimodal data, a time series analysis method is used to model the data trend, resulting in a data trend model. It is then determined whether the data trend model deviates from a preset range; if so, a dynamic adjustment mechanism is triggered. Based on the data trend model, target values ​​for adjusting the temperature and pressure parameters are determined, and the parameter configuration is updated using an optimized control strategy. The execution effect of the updated parameters is monitored in real time, and monitoring data is acquired. If the monitoring data does not reach the preset target range, trend analysis and parameter adjustment are repeated. Historical temperature and pressure parameter data are acquired, and a machine learning algorithm is used to train the historical data to obtain a temperature-pressure correlation model. Based on the prediction results of the correlation model, the control strategy is optimized. Through iterative execution of monitoring and adjustment, the temperature and pressure parameters are stabilized within the target range, achieving optimal control of the temperature control system.

[0058] For example, in the processing of multimodal data in a temperature control system, real-time monitoring data of temperature and pressure parameters includes equipment operating temperature, cavity pressure, and ambient temperature. Taking the rubber sheet pressing process as an example, when the production line is running, temperature data is collected by sensors placed at various parts of the press, and pressure data is collected by pressure sensors, allowing for the acquisition of multi-dimensional operating parameters at each time point. When modeling these data using time series analysis methods, a moving average can be used to analyze temperature change trends. Assuming that the temperature should fluctuate around 180 degrees Celsius during normal pressing, if the temperature is detected to rise continuously within ten minutes and exceed 190 degrees Celsius, the system will determine this as an abnormal trend and trigger adjustment. The implementation of the dynamic adjustment mechanism needs to be combined with process requirements. Taking the constant temperature and pressure stage in the pressing process as an example, when the temperature is too high, the control strategy will calculate cooling parameters based on the rate of temperature rise. If a synchronous increase in pressure is detected, the pressure parameters need to be adjusted simultaneously to ensure product quality. The correlation between temperature and pressure parameters is reflected in the process. Taking the vulcanization process as an example, in the initial stage of vulcanization, an increase in temperature will cause the rubber molecules to move more rapidly, requiring an appropriate increase in pressure to ensure the molding effect. By analyzing historical data using machine learning algorithms, a mapping relationship between temperature and pressure can be established to predict the optimal pressure value required at different temperatures. Continuous optimization of control strategies requires accurate performance evaluation. For example, in the molding process, if monitoring data after parameter adjustments shows a product thickness deviation exceeding 0.2 mm or surface hardness failing to meet standards, parameter optimization needs to be repeated. The system continuously improves the temperature-pressure correlation model based on actual production data, enhancing prediction accuracy. By establishing a closed-loop control linking temperature and pressure, the production process can be managed more precisely. In practical applications, when an abnormal temperature rise is detected, the system predicts its impact on pressure and adjusts relevant parameters in advance to avoid quality fluctuations. This proactive adjustment effectively reduces defective products caused by parameter misalignment, improving production efficiency and product quality stability.

[0059] This invention provides a multimodal temperature control system for a three-in-one laminator, mainly comprising:

[0060] The timestamp alignment module is used to synchronously acquire data from temperature, pressure, and vision sensors using timestamp alignment technology, ensuring the temporal consistency of multimodal data.

[0061] The Gaussian filter module is used to process the raw data from temperature and pressure sensors using a Gaussian filtering algorithm, filtering out high-frequency noise and obtaining smoothed sensor data.

[0062] The adaptive illumination compensation module is used to eliminate the influence of changes in ambient light on film detection by employing an adaptive illumination compensation algorithm on the image data acquired by the vision system.

[0063] The polarization filtering module is used to preprocess the image in the reflective area of ​​the film using polarization filtering technology, thereby reducing reflective interference and improving the accuracy of visual inspection.

[0064] The data fusion module is used to input the filtered temperature and pressure data and the processed visual data into a preset data fusion model to generate multimodal feature vectors.

[0065] The fuzzy PID control module is used to dynamically adjust control parameters based on multimodal feature vectors using a fuzzy PID control algorithm. If the temperature change exceeds a preset threshold, the heating power is adjusted first.

[0066] The weight allocation module is used to dynamically allocate the weights of the PID control algorithm in the fuzzy control rules by combining the changing trends of temperature and pressure, so as to ensure that the system can respond quickly in both steady state and dynamic processes.

[0067] The genetic optimization module is used to optimize fuzzy control rules using a genetic algorithm, using historical temperature and pressure data as training samples to generate the optimal control strategy.

[0068] The real-time control module is used to deploy the optimized control strategy to the temperature control system, monitor the changing trends of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

[0069] The above description is merely an 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 multimodal temperature control method for a three-in-one laminator, characterized in that, The method includes the following steps: S101. Timestamp alignment technology is used to synchronously collect data from temperature, pressure and vision sensors to ensure the temporal consistency of multimodal data; S102. The raw data of the temperature and pressure sensors are processed by Gaussian filtering algorithm to filter out high-frequency noise and obtain smoothed sensor data. S103. For the image data acquired by the vision system, an adaptive illumination compensation algorithm is adopted to eliminate the influence of changes in ambient light on the film detection. S104. In the reflective area of ​​the film, polarization filtering technology is used to preprocess the image to reduce reflective interference and improve the accuracy of visual inspection. S105. Input the filtered temperature and pressure data and the processed visual data into the preset data fusion model to generate a multimodal feature vector; S106. A fuzzy PID control algorithm is adopted to dynamically adjust the control parameters based on the multimodal feature vector. If the temperature change exceeds the preset threshold, the heating power is adjusted first. S107. In the fuzzy control rules, the weights of the PID control algorithm are dynamically allocated by combining the changing trends of temperature and pressure to ensure that the system can respond quickly in both steady-state and dynamic processes. S108. Optimize the fuzzy control rules using a genetic algorithm, and generate the optimal control strategy using historical temperature and pressure data as training samples. S109. Deploy the optimized control strategy to the temperature control system, monitor the changing trend of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

2. The multimodal temperature control method for a three-in-one laminator according to claim 1, characterized in that, S101 includes: Acquire raw data from temperature, pressure, and vision sensors, and record the timestamp information for each data point; Based on the timestamp information, the temperature data, pressure data, and visual data are initially aligned to generate a preliminary aligned dataset. For the initial aligned dataset, an interpolation algorithm is used to supplement the missing timestamp data in the initial aligned dataset to obtain the complete aligned dataset; Obtain the complete aligned dataset and determine whether there is a timestamp discrepancy in the complete aligned dataset; If there is a timestamp discrepancy, a time compensation algorithm is used to correct it for the complete aligned dataset to obtain a corrected complete aligned dataset. Based on the timestamp information contained in the corrected complete aligned dataset, the temperature data, pressure data, and visual data are reordered to generate a time-consistent multimodal dataset. A denoising algorithm is used to denoise the time-consistent multimodal dataset to obtain an optimized multimodal dataset. Temperature, pressure, and visual features are extracted from the optimized multimodal dataset using a feature extraction algorithm to generate a multimodal feature dataset.

3. The multimodal temperature control method for a three-in-one laminator according to claim 1, characterized in that, S102 includes: Obtain the raw values ​​from the temperature and pressure sensors, preprocess the raw values ​​to remove outliers and missing values, and generate standardized data; Based on the preset Gaussian filter algorithm parameters, determine the size and standard deviation of the filter kernel, and generate the Gaussian filter kernel matrix; The standardized data is convolved with the Gaussian filter kernel matrix to obtain preliminary filtered data. The preliminary filtered data is analyzed. If edge effects exist, a boundary filling method is used to process them and obtain filtered data without edge effects. Based on the comparison between the filtered data and the original data, the noise filtering effect is calculated. If the noise filtering effect does not reach the preset threshold, the Gaussian filter kernel parameters are adjusted and the filtered data is regenerated. The filtered data is converted into smoothed values, and the smoothed values ​​are then converted into data formats to generate the final smoothed data. The final smoothed data is stored to generate a smoothed sensor data file.

4. The multimodal temperature control method for a three-in-one laminator according to claim 1, characterized in that, S103 includes: Acquire an image of the adhesive film, and preprocess the image of the adhesive film to remove noise interference, thereby obtaining a preprocessed image; Based on the preprocessed image, calculate its average brightness value and determine the light intensity; If the light intensity exceeds a preset threshold, an adaptive light compensation algorithm is activated for adjustment. Histogram equalization is used to enhance the contrast of the preprocessed image; The brightness distribution of the preprocessed image is adjusted using gamma correction to obtain an optimized image. Based on the optimized image, perform film defect detection and identify abnormal areas; Based on the defect detection results of the adhesive film, an adhesive film quality report is generated, and the final processing result is output.

5. A multimodal temperature control method for a three-in-one laminator according to any one of claims 1-4, characterized in that, S104 includes: Acquire the original image containing the reflective area of ​​the adhesive film; The original image is processed using polarization filtering technology to obtain a de-reflected image; Based on a preset threshold, determine whether there are residual reflective areas in the de-reflected image; If there are residual reflective areas in the de-reflected image, the polarization angle is adjusted, and the original image is re-polarized filtered according to the adjusted polarization angle. The contrast and clarity of the de-reflective image are improved through image enhancement algorithms; An edge detection algorithm is used to extract the feature contours in the de-reflected image; Based on the feature contour, the position and boundary of the film region in the original image are determined.

6. The multimodal temperature control method for a three-in-one laminator according to any one of claims 1-4, characterized in that, S105 includes: Acquire raw temperature data, raw pressure data, and raw visual data; The original temperature data and the original pressure data are denoised using a filtering algorithm to obtain filtered temperature data and filtered pressure data. The original visual data is denoised and feature extracted using image processing algorithms to obtain processed visual data. The filtered temperature data, the filtered pressure data, and the processed visual data are input into a preset data fusion model; The data fusion model performs weighted fusion of the filtered temperature data, the filtered pressure data, and the processed visual data according to a preset weight allocation rule. If the data fusion model detects an anomaly in the filtered temperature data or the filtered pressure data, the weight ratio of the processed visual data is adjusted. Multimodal feature vectors are generated through the multi-layer neural network structure of the data fusion model. Clustering algorithms are used to classify the multimodal feature vectors to obtain the final feature vector categories.

7. The multimodal temperature control method for a three-in-one laminator according to any one of claims 1-4, characterized in that, S107 includes: The temperature value collected by the temperature sensor and the pressure value collected by the pressure sensor are acquired, and the change in the temperature value and the change in the pressure value are calculated. A time series analysis method is used to obtain temperature trend values ​​based on the changes in temperature values ​​and pressure trend values ​​based on the changes in pressure values, thereby establishing a trend prediction model. Based on the trend prediction model, determine whether the system is in a steady state or a dynamic state, and determine the range of control values; The control value range is mapped to a preset fuzzy value space, and the fuzzy values ​​in the fuzzy value space are subjected to fuzzy inference using preset fuzzy rules to obtain the fuzzy inference result. Based on the fuzzy inference results, the weight values ​​of the proportional term, integral term, and derivative term in the PID control algorithm are dynamically adjusted. The adjusted proportional term weight value, integral term weight value, and derivative term weight value are applied to the PID control algorithm to generate system execution instructions; The system executes instructions to adjust the actuator so that the system output reaches the expected response value.

8. The multimodal temperature control method for a three-in-one laminator according to claim 1, characterized in that, S108 includes: Historical temperature and pressure data are acquired, and the historical temperature and pressure data are preprocessed to remove outliers and missing values, resulting in preprocessed historical temperature and pressure data. Based on the preprocessed historical temperature and pressure data, a fuzzy control rule base is established, fuzzy control rule parameters are initialized, and an initial control strategy set is generated. A genetic algorithm is used to encode the parameters of the fuzzy control rules, construct a fitness function, and calculate the fitness value of each control strategy in the initial control strategy set. The initial set of control strategies is selected based on the fitness value, retaining control strategies with high fitness and eliminating control strategies with low fitness. Perform crossover and mutation operations on the retained control strategies to obtain a new set of control strategies; The selection operation, the crossover operation, and the mutation operation are performed iteratively until the fitness value reaches a preset threshold or the number of iterations reaches a preset upper limit. The control strategy with the highest output fitness value is taken as the optimal control strategy, and the optimal control strategy is applied to the temperature and pressure control scenario.

9. The multimodal temperature control method for a three-in-one laminator according to claim 1, characterized in that, S109 includes: Acquire multimodal data from the temperature control system, the multimodal data including temperature parameters and pressure parameters; For the aforementioned multimodal data, time series analysis methods are used to model the data trends, resulting in a data trend model; Determine whether the data trend model deviates from the preset range; if so, trigger a dynamic adjustment mechanism. Based on the data trend model, target values ​​for adjusting temperature and pressure parameters are determined, and parameter configurations are updated using the optimized control strategy. The system monitors the effect of updated parameters in real time and obtains monitoring data. If the monitoring data does not reach the preset target range, the system performs trend analysis and parameter adjustment again. Historical temperature and pressure parameter data are obtained, and machine learning algorithms are used to train the historical data to obtain a correlation model between temperature and pressure. Based on the prediction results of the correlation model, optimize the control strategy; By iteratively performing monitoring and adjustment, the temperature and pressure parameters are stabilized within the target range, achieving optimal control of the temperature control system.

10. A multimodal temperature control system for a three-in-one laminator, characterized in that, This system is used to implement the multimodal temperature control method for a three-in-one laminator as described in any one of claims 1-9, the system comprising: The timestamp alignment module is used to synchronously acquire data from temperature, pressure, and vision sensors using timestamp alignment technology, ensuring the temporal consistency of multimodal data. The Gaussian filter module is used to process the raw data from temperature and pressure sensors using a Gaussian filtering algorithm, filtering out high-frequency noise and obtaining smoothed sensor data. The adaptive illumination compensation module is used to eliminate the influence of changes in ambient light on film detection by employing an adaptive illumination compensation algorithm on the image data acquired by the vision system. The polarization filtering module is used to preprocess the image in the reflective area of ​​the film using polarization filtering technology, thereby reducing reflective interference and improving the accuracy of visual inspection. The data fusion module is used to input the filtered temperature and pressure data and the processed visual data into a preset data fusion model to generate multimodal feature vectors. The fuzzy PID control module is used to dynamically adjust the control parameters based on the multimodal feature vector using the fuzzy PID control algorithm. If the temperature change exceeds the preset threshold, the heating power will be adjusted first. The weight allocation module is used to dynamically allocate the weights of the PID control algorithm in the fuzzy control rules by combining the changing trends of temperature and pressure, so as to ensure that the system can respond quickly in both steady state and dynamic processes. The genetic optimization module is used to optimize fuzzy control rules using a genetic algorithm, using historical temperature and pressure data as training samples to generate the optimal control strategy. The real-time control module is used to deploy the optimized control strategy to the temperature control system, monitor the changing trends of multimodal data in real time, and dynamically adjust the temperature and pressure parameters during the gluing process.

Images