Corrugated board humidity on-line automatic monitoring method

Abstract

The invention provides a method for automatically monitoring the humidity of a corrugated board on line, and belongs to the technical field of corrugated board production. Uniform thermal pulses are applied to the surface of the corrugated board through a thermal infrared imaging device in combination with a thermal pulse excitation system, and high-resolution thermal response images are collected in real time. The system calculates thermal diffusivity and thermal attenuation characteristics based on an unsteady state heat conduction equation, generates a temperature distribution matrix, extracts a humidity characteristic spectrum by using a heat and humidity two-parameter neural network model, and establishes a corresponding relation between temperature and water content. Carrying out region segmentation through the humidity characteristic spectrum, identifying a humidity abnormal region, calculating a deviation value, and triggering an alarm when the deviation value exceeds a preset threshold value. The technical problem that in the prior art, corrugated board humidity detection mainly depends on manual sampling and special instruments, and real-time online monitoring on a production line cannot be achieved is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of corrugated board production, and more specifically, relates to a method for on-line automatic monitoring of the humidity of corrugated board. Background Art

[0002] As a basic material in the packaging industry, the physical properties and service life of corrugated board are largely affected by the moisture content distribution. Traditional corrugated board humidity detection technologies mainly rely on manual sampling detection methods, that is, after sampling under the production line, special humidity measuring instruments (such as drying and weighing method, resistive hygrometer, capacitive moisture detector, etc.) are used to measure and analyze the samples. This method has been applied in the packaging industry for decades, and the process is relatively mature, and the detection accuracy can reach a relatively high level in a controlled experimental environment.

[0003] However, the traditional manual sampling detection method has significant defects: First, sampling detection has obvious hysteresis. It usually takes several minutes or even longer from sampling to obtaining the detection result, and it cannot respond in time to the humidity changes during the production process; Second, the randomness and finiteness of sampling are difficult to comprehensively reflect the humidity distribution of the entire batch of products, and it is easy to miss local abnormal areas; Third, there are subjective factors in the manual operation process, and the consistency of the detection results is poor; Finally, this off-line detection method cannot form an automatic feedback adjustment mechanism with the production equipment, and it is difficult to achieve intelligent production control. That is to say, there is a technical problem in the prior art that the humidity detection of corrugated board mainly relies on manual sampling and special instruments, and real-time on-line monitoring on the production line cannot be achieved. Summary of the Invention

[0004] In view of this, the present invention provides a method for on-line automatic monitoring of the humidity of corrugated board, which can solve the technical problem in the prior art that the humidity detection of corrugated board mainly relies on manual sampling and special instruments, and real-time on-line monitoring on the production line cannot be achieved.

[0005] The present invention is implemented as follows: The present invention provides a method for on-line automatic monitoring of the humidity of corrugated cardboard, which includes: using a thermal infrared imaging device at a fixed position on the corrugated cardboard production line, adjusting the focal length to cover the entire width of the corrugated cardboard; using a thermal pulse excitation system to apply a uniform thermal pulse to the surface of the corrugated cardboard; collecting a thermal response image of the surface of the corrugated cardboard through the thermal infrared imaging device; according to the thermal response image, calculating the thermal diffusivity and thermal attenuation characteristics of each pixel point of the corrugated cardboard based on the unsteady heat conduction equation, and generating a temperature distribution matrix; analyzing the temperature distribution matrix based on a thermal humidity dual-parameter neural network model, extracting a humidity characteristic map, and establishing a corresponding relationship between the temperature distribution matrix and the water content of the corrugated cardboard; performing region segmentation through the humidity characteristic map, identifying the humidity abnormal region, and calculating the area ratio and humidity deviation value of the humidity abnormal region; comparing the humidity deviation value with a preset threshold, and triggering an alarm if it exceeds the preset threshold.

[0006] Among them, the thermal pulse excitation system specifically forms a uniform heat source by instantaneously releasing heat energy through a high-power infrared lamp array, and is used to stimulate the surface of the corrugated cardboard to be measured to generate a thermal response image.

[0007] Among them, the thermal pulse intensity is maintained at 5000 watts per square meter, and the thermal pulse duration is 0.5 seconds.

[0008] Among them, the resolution of the thermal response image collected by the thermal infrared imaging device is set to not less than 640×480 pixels, and the sampling frequency of the thermal infrared imaging device is 25 Hz.

[0009] Among them, the thermal diffusivity refers to the parameter of the heat conduction ability of the corrugated cardboard. Since the increase in moisture content will reduce the thermal diffusivity, the area with higher humidity shows the characteristics of more heat accumulation and slower heat dissipation.

[0010] Among them, the thermal attenuation characteristic refers to the curve of the temperature change of the corrugated cardboard with time after absorbing heat energy. By analyzing the differences in the thermal attenuation characteristics of different regions, the internal water content distribution of the corrugated cardboard is reflected.

[0011] Among them, the temperature distribution matrix refers to a two-dimensional array composed of the temperature values of each pixel point of the thermal response image, and is used to quantitatively describe the temperature distribution state of the entire surface of the corrugated cardboard.

[0012] Among them, the humidity deviation value refers to the difference value between the actual measured humidity and the standard humidity, expressed as a percentage. Usually, it is required that the humidity deviation value be controlled within 3%.

[0013] Among them, the unsteady heat conduction equation is used to describe the dynamic process of heat transfer in the corrugated cardboard and calculate the thermal diffusivity and thermal attenuation characteristics. The inputs include the surface temperature change rate of the corrugated cardboard, the initial temperature distribution of the corrugated cardboard, the specific heat capacity of the corrugated cardboard material, the density of the corrugated cardboard material, and the thermal conductivity of the corrugated cardboard material. The outputs are the thermal diffusivity and thermal attenuation characteristics of each point of the corrugated cardboard.

[0014] Among them, the thermal humidity dual-parameter neural network model structure is a five-layer convolutional neural network structure, including three convolutional layers and two fully connected layers. The first convolutional layer is used to extract the basic features of the temperature distribution matrix, the second convolutional layer is used to extract the spatial distribution features of the thermal diffusivity, the third convolutional layer is used to extract the time series features of the thermal attenuation characteristics, the first fully connected layer fuses the three features and combines them with the production process parameters, and the second fully connected layer outputs the humidity feature map.

[0015] Among them, the humidity feature map refers to a visualized image reflecting the water content distribution of the corrugated board obtained by processing the temperature distribution matrix, and different colors represent different humidity levels.

[0016] It also includes performing time series analysis on the humidity deviation value and optimizing the production parameters; associating the humidity deviation value with the production process parameters, and automatically adjusting the drying temperature and drying speed through a feedback control system to achieve closed-loop control.

[0017] The present invention realizes full-width real-time humidity monitoring on the corrugated board production line through a non-contact detection technology combining thermal infrared imaging and thermal pulse excitation. The system uses a high-frequency sampling rate of 25 Hz and a high-resolution imaging of not less than 640×480 pixels, ensuring continuous capture of the humidity state of the cardboard on the fast-running production line, and fundamentally solving the lag problem of traditional sampling detection.

[0018] This method establishes an accurate mapping relationship between the temperature response characteristics and the humidity distribution by combining the heat conduction theory and artificial intelligence algorithms. The system can generate the humidity feature map of the entire cardboard width in real time, automatically identify the abnormal area and calculate the deviation value, expanding the humidity monitoring from traditional point measurement to surface measurement, and comprehensively improving the detection coverage and accuracy. At the same time, the integrated time series analysis and feedback control functions of the system can automatically adjust the production process parameters according to the real-time monitoring data, realizing the transformation from passive detection to active control.

[0019] In summary, the present invention solves the technical problem that the humidity detection of corrugated board mainly relies on manual sampling and special instruments and cannot realize real-time online monitoring on the production line, and provides a new way of fully automatic, intelligent and high-efficiency quality control for corrugated board production. Description of the Drawings

[0020] Figure 1 It is a flowchart of the method of the present invention.

[0021] Figure 2 It is a thermal response curve graph under different humidity conditions in Embodiment 2.

[0022] Figure 3It is a process diagram for training the thermal and humidity dual-parameter neural network model in Embodiment 2.

[0023] Figure 4 It is an effect diagram of parameter adjustment and quality improvement of the closed-loop control system in Embodiment 2. Specific implementation manners

[0024] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] As Figure 1 shown, it is a flowchart of a method for on-line automatic monitoring of the humidity of corrugated board provided by the present invention. The method includes the following steps: S01. Use a thermal infrared imaging device at a fixed position on the corrugated board production line, and adjust the focal length of the thermal infrared imaging device so that the imaging area covers the entire width of the corrugated board; S02. Apply a uniform thermal pulse to the surface of the corrugated board by using a thermal pulse excitation system. The intensity of the thermal pulse is maintained at 5000 watts per square meter, and the duration of the thermal pulse is 0.5 seconds; S03. Collect the thermal response image of the surface of the corrugated board through the thermal infrared imaging device. The resolution of the thermal response image is set to be not less than 640×480 pixels, and the sampling frequency of the thermal infrared imaging device is 25 Hz; S04. According to the thermal response image, calculate the thermal diffusivity and thermal attenuation characteristics of each pixel point of the corrugated board based on the unsteady heat conduction equation, and generate a temperature distribution matrix; S05. Analyze the temperature distribution matrix based on the thermal and humidity dual-parameter neural network model, extract the humidity characteristic map, and establish the corresponding relationship between the temperature distribution matrix and the water content of the corrugated board; S06. Perform region segmentation through the humidity characteristic map, identify the humidity abnormal region, and calculate the area ratio and humidity deviation value of the humidity abnormal region; S07. Compare the humidity deviation value with a preset threshold. When the humidity deviation value exceeds the preset threshold, trigger the production line abnormal alarm system; S08. Optionally, it further includes performing time series analysis on the humidity deviation value, establishing a humidity historical trend graph, predicting future humidity changes and optimizing production parameters; S09. Optionally, it further includes associating the humidity deviation value with production process parameters, and automatically adjusting the drying temperature and drying speed through a feedback control system to achieve closed-loop control.

[0026] Among them, the thermal pulse excitation system is specifically a uniform heat source formed by the instantaneous release of heat energy by a high-power infrared lamp array, which is used to stimulate the generation of the thermal response image on the surface of the corrugated cardboard to be measured.

[0027] Among them, the thermal diffusivity specifically refers to the parameter of the ability of the corrugated cardboard to conduct heat. Since the increase in moisture content will reduce the thermal diffusivity, the areas with higher humidity are characterized by more heat accumulation and slower heat dissipation.

[0028] Among them, the thermal decay characteristic specifically refers to the change curve of the temperature of the corrugated cardboard decreasing with time after absorbing heat energy. By analyzing the differences in the thermal decay characteristics of different regions, the internal moisture content distribution of the corrugated cardboard is reflected.

[0029] Among them, the temperature distribution matrix specifically refers to a two-dimensional array composed of the temperature values of each pixel point of the thermal response image, which is used to quantitatively describe the temperature distribution state of the entire surface of the corrugated cardboard.

[0030] Among them, the humidity characteristic map specifically refers to a visual image reflecting the moisture content distribution of the corrugated cardboard obtained by processing the temperature distribution matrix, and different colors represent different humidity levels.

[0031] Among them, the humidity deviation value specifically refers to the difference value between the actually measured humidity and the standard humidity, expressed as a percentage. Usually, it is required that the humidity deviation value be controlled within 3%.

[0032] Among them, the closed-loop control specifically means that the system automatically adjusts the drying temperature and the drying speed according to the real-time monitored humidity deviation value, forming a cycle process of detection, analysis, adjustment, and re-detection, and maintaining the product quality stable without manual intervention.

[0033] Among them, the preset threshold specifically refers to the upper limit of the humidity deviation value preset according to the product quality requirements of the corrugated cardboard. When the humidity deviation value exceeds the preset threshold, the system determines it as an abnormal state.

[0034] Among them, the production process parameters specifically refer to the key parameters affecting the humidity control in the production process of the corrugated cardboard, including the drying temperature, the drying speed, the ambient humidity, and the production line speed.

[0035] Among them, the unsteady heat conduction equation is used to describe the dynamic process of heat transfer in the corrugated cardboard and calculate the thermal diffusivity and the thermal attenuation characteristics. The input includes the surface temperature change rate of the corrugated cardboard, the initial temperature distribution of the corrugated cardboard, the specific heat capacity of the corrugated cardboard material, the density of the corrugated cardboard material, and the thermal conductivity of the corrugated cardboard material. The output is the thermal diffusivity and the thermal attenuation characteristics of each point of the corrugated cardboard. The surface temperature change rate of the corrugated cardboard is derived from the temperature difference calculation between consecutive frames of the thermal response image. The initial temperature distribution of the corrugated cardboard is derived from the initial thermal image collected by the thermal infrared imaging device before the thermal pulse excitation. The specific heat capacity of the corrugated cardboard material, the density of the corrugated cardboard material, and the thermal conductivity of the corrugated cardboard material are derived from the corrugated cardboard material database. The thermal diffusivity is used to generate the temperature distribution matrix. The thermal attenuation characteristics are used to analyze the internal moisture content distribution of the corrugated cardboard.

[0036] The specific structure of the thermal humidity two-parameter neural network model is a five-layer convolutional neural network structure, including three convolutional layers and two fully connected layers. The first convolutional layer is used to extract the basic features of the temperature distribution matrix. The second convolutional layer is used to extract the spatial distribution features of the thermal diffusivity. The third convolutional layer is used to extract the time series features of the thermal attenuation characteristics. The first fully connected layer fuses the three features and combines them with the production process parameters. The second fully connected layer outputs the humidity feature map. The size of the convolutional kernel in the thermal humidity two-parameter neural network model structure matches the corrugation structure size of the corrugated cardboard using a preset convolutional kernel correlation function, and is used to capture the differences in humidity distribution in different corrugation structures.

[0037] The steps for establishing the training data set of the thermal humidity two-parameter neural network model specifically include collecting the thermal response image samples under different humidity conditions from the actual production line, synchronously collecting the actual humidity measurement values for each sample as label data, introducing different production conditions including the thickness of the corrugated cardboard, the corrugation type of the corrugated cardboard, and the raw material composition of the corrugated cardboard to construct a multi-dimensional feature vector, and using data augmentation techniques to increase sample diversity including rotation, scaling, and adding noise processing. Finally, a training set containing 10,000 groups of samples and a validation set containing 2,000 groups of samples are formed. The actual humidity measurement value is derived from a contact humidity sensor. The multi-dimensional feature vector is used to enhance the adaptability of the thermal humidity two-parameter neural network model to different types of corrugated cardboard.

[0038] The steps for training the thermal humidity dual-parameter neural network model specifically include optimizing the weights of the thermal humidity dual-parameter neural network model using the batch gradient descent algorithm. The initial learning rate is set to 0.001 and an adaptive adjustment strategy is adopted. The mean square error is used as the loss function to evaluate the prediction accuracy of the thermal humidity dual-parameter neural network model. An early stopping strategy is introduced during training to prevent overfitting. Training stops when the loss function value on the validation set does not decrease for 5 consecutive training cycles. Finally, the parameters of the thermal humidity dual-parameter neural network model with the best performance on the validation set are selected as the final model. The batch gradient descent algorithm is used to efficiently update the weights of the thermal humidity dual-parameter neural network model. The early stopping strategy is used to prevent the thermal humidity dual-parameter neural network model from overfitting to the training data and resulting in a decline in generalization ability.

[0039] The following describes the specific implementation manners of the above steps in detail.

[0040] The specific implementation manner of step S01 is to fixedly install a thermal infrared imaging device at an appropriate position on the corrugated cardboard production line. This position needs to ensure unobstructed observation of the entire width of the corrugated cardboard. The thermal infrared imaging device uses a non-cooled microbolometer focal plane array detector, with a working wavelength range of 7.5 - 14 μm, a temperature resolution of not less than 0.05 °C, and a field of view angle set to 60°×45°. After installation, precise focal length adjustment is carried out. During the adjustment process, a calibration plate is temporarily placed on the corrugated cardboard production line. The calibration plate is marked with equally spaced reference points. Remote fine adjustment is carried out through the electric focus adjustment mechanism of the thermal infrared imaging device until the image is clear and covers the area of 107×78 cm of the entire width of the corrugated cardboard. The purpose of this step is to establish a stable and reliable thermal imaging observation basis to ensure the integrity and accuracy of subsequent thermal response data collection.

[0041] The specific implementation manner of step S02 is to start the thermal pulse excitation system to apply a uniform thermal pulse to the surface of the corrugated cardboard. The thermal pulse excitation system consists of a high-power infrared lamp array. The array contains 25 infrared lamps with a single power of 250 watts, arranged in a 5×5 matrix, and the coverage area matches the width of the corrugated cardboard. Before the thermal pulse is generated, the input current is precisely adjusted to the rated value through a power controller to ensure that the thermal pulse intensity reaches 5000 watts per square meter. After the thermal pulse starts, a precision timer controls the duration of the thermal pulse to be 0.5 seconds, with a timing accuracy of ±0.01 seconds. During the thermal pulse process, a uniformity monitoring system continuously detects the spatial distribution of the thermal pulse to ensure that the deviation of the thermal pulse intensity in the entire area does not exceed ±3%. The function of this step is to generate standardized thermal excitation conditions on the surface of the corrugated cardboard to provide controllable and repeatable energy input for subsequent humidity analysis based on thermal response.

[0042] The specific implementation of step S03 is to start the thermal infrared imaging device to immediately collect the thermal response image of the corrugated cardboard surface after the thermal pulse is applied. The image acquisition resolution is set to 640×480 pixels, ensuring that each pixel corresponds to an area of approximately 1.67×1.63 mm on the corrugated cardboard surface. The sampling frequency of the thermal infrared imaging device is set to 25 Hz, that is, one thermal image is collected every 40 milliseconds, and 625 thermal images are collected continuously for 25 seconds. During the acquisition process, the temperature drift compensation system of the thermal infrared imaging device continuously works and is calibrated in real time through the built-in blackbody reference source to ensure that the temperature measurement accuracy is within ±0.5°C. After the image acquisition is completed, the thermal response image sequence is transmitted to the image processing unit through a high-speed data transmission interface for storage and preprocessing. The purpose of this step is to obtain the time response sequence of the corrugated cardboard to the thermal pulse stimulus and capture the temperature change characteristics caused by the difference in heat conduction characteristics in different humidity regions.

[0043] The specific implementation of step S04 is to process the thermal response image in the image processing unit based on the unsteady heat conduction equation. First, spatial filtering is performed on the original thermal image sequence, and a 5×5 Gaussian filter is used to eliminate random noise, with the filter standard deviation set to 1.2 pixels. Then, the temperature change rate between consecutive frames is calculated for each pixel point to construct a temperature change rate field. For each pixel point, the initial temperature before the thermal pulse , the peak temperature after the thermal pulse , and the relationship curve between temperature and time during the cooling process are extracted. These parameters are substituted into the unsteady heat conduction equation: , where is the thermal diffusivity, and is the Laplacian operator of temperature. By numerically solving this equation, the thermal diffusivity and the thermal decay characteristics of each pixel point are obtained, where represents the exponential coefficient of temperature decay with time, satisfying . Based on the calculation results, a temperature distribution matrix with a resolution of 640×480 is generated, and each element in the matrix contains the temperature value, thermal diffusivity, and thermal decay characteristic parameters of that point. The purpose of this step is to extract the thermophysical parameters closely related to humidity from the thermal response data through the heat conduction physical model, providing a theoretical basis for subsequent humidity analysis.

[0044] The specific implementation of step S05 is to input the temperature distribution matrix obtained in step S04 into a thermal-humidity dual-parameter neural network model. This neural network adopts a five-layer convolutional neural network structure, including three convolutional layers and two fully connected layers. The first convolutional layer uses 32 5×5 convolutional kernels with a stride of 2 and a ReLU activation function to extract the basic texture features of the temperature distribution matrix. The second convolutional layer uses 64 3×3 convolutional kernels with a stride of 1 and a ReLU activation function, focusing on extracting the spatial distribution features of the thermal diffusivity. The third convolutional layer uses 128 3×3 convolutional kernels with a stride of 1 and a ReLU activation function to capture the time series features of the thermal attenuation characteristics. The first fully connected layer contains 512 neurons with a ReLU activation function, combining the features extracted by the previous three layers with the production process parameters (including drying temperature, drying speed, ambient humidity, and production line speed) to form a comprehensive feature vector. The second fully connected layer contains 640×480 neurons corresponding to the resolution of the output humidity feature map, with a Sigmoid activation function, and the output value ranges from 0 to 1, representing the normalized humidity level. The network is trained using the Adam optimizer with an initial learning rate of 0.001, and the cosine annealing strategy is used for dynamic adjustment. The mean squared error is used as the loss function during training, and the humidity prediction accuracy on the validation set reaches ±1.5%. After the neural network processing is completed, a humidity feature map is generated, and pseudo-color is used to display different humidity regions. Blue represents low humidity (5% - 8%), green represents standard humidity (8% - 12%), yellow represents slightly high humidity (12% - 15%), and red represents excessive humidity (>15%). The purpose of this step is to use deep learning technology to establish a mapping relationship between thermal parameters and humidity and achieve precise quantitative analysis of the humidity of corrugated cardboard.

[0045] The specific implementation of step S06 is to perform regional segmentation and analysis on the humidity characteristic map generated in step S05. First, the multi-threshold segmentation algorithm is used, and the humidity threshold points are set at 8%, 12%, and 15% to divide the humidity characteristic map into a low-humidity region, a standard-humidity region, a slightly high-humidity region, and an excessively high-humidity region. Then, the connected component labeling algorithm is used to identify the spatial distribution of each humidity region and calculate the number of pixels in each region. For the humidity abnormal regions (including the low-humidity region and the excessively high-humidity region), calculate the percentage of their total area in the area of the entire corrugated board to obtain the proportion of the humidity abnormal region area. At the same time, calculate the difference between the actual average humidity and the standard humidity (10%) to obtain the humidity deviation value. The calculation formula for the humidity deviation value is: humidity deviation value = |actual average humidity - standard humidity| / standard humidity × 100%. In addition, the spatial autocorrelation analysis is used to evaluate the uniformity of the humidity distribution and calculate the Moran's I index. When the index value is close to 1, it indicates that the humidity distribution shows a highly aggregated state; when it is close to 0, it indicates a random distribution; when it is close to -1, it indicates a uniform distribution. The purpose of this step is to quantify the abnormal degree and spatial characteristics of the humidity distribution of the corrugated board and provide a decision-making basis for the production process control.

[0046] The specific implementation of step S07 is to compare and judge the humidity deviation value calculated in step S06 with a preset threshold. The preset threshold is set based on the product quality requirements of the corrugated board. The preset threshold for the standard quality grade is 3%, the preset threshold for the high-quality grade is 2%, and the preset threshold for the extra-high-quality grade is 1.5%. The system automatically selects the corresponding preset threshold according to the product grade currently being produced. The comparison process uses the fuzzy logic control algorithm, which not only considers the absolute value of the humidity deviation value but also its change trend. When the humidity deviation value exceeds the preset threshold and all three consecutive detections are out of the standard, the system determines it as an abnormal state and triggers the production line abnormal alarm system. The alarm system is divided into three levels: when the humidity deviation value exceeds the preset threshold but is lower than 1.5 times the preset threshold, the first-level alarm is triggered, and the operator is reminded through the yellow indicator light on the operation panel and a soft alarm; when the humidity deviation value exceeds 1.5 times the preset threshold but is lower than 2 times the preset threshold, the second-level alarm is triggered, and the operator is reminded through the orange indicator light on the operation panel and a medium-volume alarm, and the abnormal event is recorded; when the humidity deviation value exceeds 2 times the preset threshold, the third-level alarm is triggered, and the operator is reminded through the red flashing indicator light on the operation panel and a high-volume alarm. At the same time, the production line speed is automatically reduced by 30% to ensure product quality. The purpose of this step is to timely detect the humidity abnormality in the production process, prevent the continuous production of unqualified products, and reduce the scrap rate.

[0047] Step S08 is an optional step, and its specific implementation is to perform time series analysis and prediction on the humidity deviation value. The system saves the humidity deviation value data within the most recent 72 hours, with a sampling interval of 10 minutes, forming a time series of 432 data points. First, the seasonal trend decomposition (STL) algorithm is applied to decompose the time series into a trend term, a seasonal term, and a residual term to identify potential periodic changes and long-term trends. Then, an autoregressive integrated moving average (ARIMA) model is used to model the decomposed time series, and the model parameters are automatically optimized and selected through the Bayesian information criterion. Based on the established ARIMA model, the humidity change trend within the next 2 hours is predicted, and the prediction accuracy is within ±0.8%. At the same time, the system constructs a humidity historical trend chart, which includes the actual humidity deviation value curve, the moving average curve (window size of 60 minutes), and the predicted humidity deviation value curve, visually showing the humidity change situation. In addition, the system uses historical data to identify which production process parameter changes are highly correlated with humidity fluctuations, and identifies key influencing factors through multiple regression analysis or Granger causality test to provide data support for optimizing production parameters. The purpose of this step is to achieve predictable control of humidity during the corrugated cardboard production process through historical data mining and predictive analysis, adjust production parameters in advance, and prevent abnormal humidity from occurring.

[0048] Step S09 is an optional step, and its specific implementation is to establish a closed-loop control system between the humidity deviation value and production process parameters. First, a multi-input multi-output (MIMO) control model is constructed. The input variables include the current humidity deviation value, the humidity change rate, and the predicted humidity deviation value, and the output variables include the drying temperature adjustment amount and the drying speed adjustment amount. The control model uses the model predictive control (MPC) algorithm. Based on the identified system dynamic characteristics, the control behavior is optimized under the premise of meeting the constraint conditions. During the control process, the adjustment range of the drying temperature is 65°C to 95°C, and the adjustment step is 0.5°C; the adjustment range of the drying speed is 80 to 120 m / min, and the adjustment step is 2 m / min. When it is detected that the humidity is too high, the system automatically increases the drying temperature or decreases the drying speed; when it is detected that the humidity is too low, the system automatically decreases the drying temperature or increases the drying speed. The response time of the controller does not exceed 30 seconds, and the control accuracy reaches ±1% of the set value. In addition, the system also combines an adaptive control strategy to continuously update the control model parameters through online parameter identification to adapt to changes in production conditions. The performance indicators of the control system include a steady-state error not exceeding 0.5%, an overshoot not exceeding 1.5%, and a settling time not exceeding 10 minutes. The purpose of this step is to achieve automatic closed-loop control of humidity during the corrugated cardboard production process, reduce manual intervention, and improve the stability of product quality.

[0049] The detailed structure of the thermal-humidity dual-parameter neural network model is as follows: The input layer receives data with a dimension of 640×480×3, including three channels of temperature value, thermal diffusivity, and thermal attenuation characteristics. The first layer of convolution uses 32 5×5 convolutional kernels, with a stride of 2 and a padding of 2. The output size is 320×240×32. The activation function is ReLU, followed by a batch normalization layer and a max pooling layer (pooling kernel size 2×2, stride 2). After pooling, the size is 160×120×32. The second layer of convolution uses 64 3×3 convolutional kernels, with a stride of 1 and a padding of 1. The output size is 160×120×64. The activation function is ReLU, followed by a batch normalization layer and a max pooling layer (pooling kernel size 2×2, stride 2). After pooling, the size is 80×60×64. The third layer of convolution uses 128 3×3 convolutional kernels, with a stride of 1 and a padding of 1. The output size is 80×60×128. The activation function is ReLU, followed by a batch normalization layer and a max pooling layer (pooling kernel size 2×2, stride 2). After pooling, the size is 40×30×128. After flattening the feature map, it is concatenated with the production process parameters (4D vector) and input into the first fully connected layer, which contains 512 neurons. The activation function is ReLU, and dropout (ratio 0.3) is used to prevent overfitting. The output size of the second fully connected layer is 640×480, corresponding to the resolution of the humidity feature map. The activation function is Sigmoid, and the output value ranges from 0 to 1, representing the normalized humidity level.

[0050] The process of establishing the training dataset for the thermal-humidity dual-parameter neural network model is as follows: First, samples are collected from the actual production line. Thermal response images are collected under different humidity conditions (humidity range 5% - 20%, interval 0.5%). 50 groups of samples are collected for each humidity condition, totaling 1500 groups of original samples. For each sample, a high-precision contact humidity sensor (accuracy ±0.2%) is used to synchronously measure the actual humidity value as the label data. At the same time, the production condition information of the samples is recorded, including the corrugated board thickness (range 1.5 - 7 mm, 5 grades in total), the corrugated board corrugation type (5 types: A, B, C, E, F), and the corrugated board raw material composition (8 different ratios in total), forming a 23D feature vector. Data augmentation techniques are applied to the original samples, including: random rotation (angle range ±15°), random scaling (scaling factor range 0.9 - 1.1), random addition of Gaussian noise (mean 0, standard deviation 0.01), random adjustment of brightness (adjustment range ±10%), random adjustment of contrast (adjustment range ±10%), random horizontal and vertical flipping. 5 enhanced samples are generated for each original sample, finally forming 10000 groups of training samples and 2000 groups of validation samples. The training set and the validation set are divided using the stratified sampling method to ensure that the proportions of different humidity conditions and different corrugated board types in the training set and the validation set are consistent.

[0051] Specifically, the principle of the present invention is as follows: The core technical principle of the present invention is based on the inherent relationship between the thermophysical properties of materials and humidity. Through non-contact thermal imaging technology combined with artificial intelligence analysis, real-time online monitoring of the humidity of corrugated cardboard is achieved. Its working principle can be elaborated from the following aspects:

[0052] First, the principle of thermal pulse excitation and thermal imaging. The present invention uses a high-power infrared lamp array to apply precisely controlled thermal pulses (5000 watts per square meter, lasting for 0.5 seconds) to the surface of the corrugated cardboard, causing the surface temperature of the cardboard to rise uniformly for a short time. Since moisture will significantly affect the thermophysical properties of materials, regions with different moisture contents will exhibit obvious differences in temperature response and heat diffusion behavior after receiving the same thermal pulse. The high-precision thermal infrared imaging device generates time-series image data reflecting the thermal response characteristics by continuously collecting these temperature change processes, providing basic information for subsequent analysis.

[0053] Second, the principle of heat-humidity relationship analysis. The present invention is based on the unsteady heat conduction equation to analyze the heat conduction behavior in the corrugated cardboard after the action of thermal pulses. According to the principles of thermophysics, the moisture content of materials will directly affect their thermal diffusivity and heat capacity. Regions with high moisture content are characterized by more heat accumulation and slower heat dissipation. By accurately calculating the thermal diffusivity and thermal attenuation characteristics of each pixel point, the system can establish a physical relationship between the temperature distribution and the moisture content, providing a theoretical basis for humidity measurement.

[0054] Third, the principle of intelligent recognition and mapping. The thermal and humidity dual-parameter neural network model designed in the present invention adopts a five-layer convolutional neural network structure, which matches the corrugated structure characteristics of the corrugated cardboard and can effectively capture the humidity distribution characteristics of different regions. Through large-scale training with 10,000 groups of samples, the model establishes a complex non-linear mapping relationship between the thermal response characteristics and the actual humidity, enabling the system to accurately extract humidity information from the temperature distribution matrix and generate an intuitive humidity characteristic map.

[0055] Finally, the principle of closed-loop control. The system correlates the identified humidity deviation value with the production process parameters to form a complete feedback control mechanism. When abnormal humidity is detected, the system can automatically adjust key process parameters such as drying temperature and speed according to preset rules, realizing a closed-loop control process of detection - analysis - adjustment - re-detection. At the same time, the system performs time-series analysis on the humidity data to predict future change trends, providing a decision-making basis for process parameter optimization.

[0056] Based on the above principles, the present invention organically combines advanced thermal imaging technology, thermophysical analysis methods, and artificial intelligence algorithms to construct a humidity monitoring and control system that can operate in real time on the production line, fundamentally solving the problem that traditional sampling detection methods cannot achieve real-time online monitoring.

[0057] A specific Embodiment 1 of the present invention is provided below. The specific implementation of each step in this Embodiment 1 is described in detail as follows.

[0058] The specific implementation of step S01 is to fixedly install a thermal infrared imaging device at an appropriate position on the corrugated cardboard production line. This position needs to ensure unobstructed observation of the entire width of the corrugated cardboard. The thermal infrared imaging device uses a non-cooled microbolometer focal plane array detector, with a working wavelength range of 7.5 - 14 μm, a temperature resolution of not less than 0.05 °C, and a field of view angle set to 60°×45°. After installation, precise focus adjustment is carried out. During the adjustment process, a calibration plate is temporarily placed on the corrugated cardboard production line. The calibration plate is marked with equally spaced reference points, and remote fine adjustment is carried out through the electric focus adjustment mechanism of the thermal infrared imaging device until the image is clear and covers the area of the entire corrugated cardboard width of 107×78 cm. The purpose of this step is to establish a stable and reliable thermal imaging observation basis to ensure the integrity and accuracy of subsequent thermal response data collection.

[0059] The specific implementation of step S02 is to start the thermal pulse excitation system to apply a uniform thermal pulse to the surface of the corrugated cardboard. The thermal pulse excitation system consists of a high-power infrared lamp array. The array contains 25 infrared lamps with a single power of 250 watts, arranged in a 5×5 matrix, and the coverage area matches the width of the corrugated cardboard. Before the thermal pulse is generated, the input current is precisely adjusted to the rated value through a power controller to ensure that the thermal pulse intensity reaches 5000 watts per square meter. After the thermal pulse starts, a precision timer controls the thermal pulse duration to be 0.5 seconds, with a timing accuracy of ±0.01 seconds. During the thermal pulse process, a uniformity monitoring system continuously detects the spatial distribution of the thermal pulse to ensure that the deviation of the thermal pulse intensity in the entire area does not exceed ±3%. The function of this step is to generate standardized thermal excitation conditions on the surface of the corrugated cardboard to provide controllable and repeatable energy input for subsequent humidity analysis based on thermal response.

[0060] The specific implementation of step S03 is to start the thermal infrared imaging device to immediately collect the thermal response image of the corrugated cardboard surface after the thermal pulse is applied. The image acquisition resolution is set to 640×480 pixels, ensuring that each pixel corresponds to an area of approximately 1.67×1.63 mm on the corrugated cardboard surface. The sampling frequency of the thermal infrared imaging device is set to 25 Hz, that is, one thermal image is collected every 40 milliseconds, and it is continuously collected for 25 seconds, with a total of 625 thermal images collected. During the collection process, the temperature drift compensation system of the thermal infrared imaging device continuously works and is calibrated in real time through the built-in blackbody reference source to ensure that the temperature measurement accuracy is within ±0.5°C. After the image collection is completed, the thermal response image sequence is transmitted to the image processing unit through the high-speed data transmission interface for storage and preprocessing. The purpose of this step is to obtain the time response sequence of the corrugated cardboard to the thermal pulse stimulation and capture the temperature change characteristics generated by the difference in heat conduction characteristics in different humidity regions.

[0061] The specific implementation of step S04 is to process the thermal response image in the image processing unit based on the unsteady heat conduction equation. First, spatial filtering is performed on the original thermal image sequence, and a 5×5 Gaussian filter is used to eliminate random noise. The filter kernel function can be expressed as , where is the Gaussian standard deviation, set to 1.2 pixels; and are the coordinates of the pixel in the filter kernel. Then, calculate the temperature change rate between consecutive frames for each pixel point and construct the temperature change rate field. For each pixel point , extract the initial temperature before the thermal pulse, the peak temperature after the thermal pulse, and the relationship curve between temperature and time during the cooling process. Substitute these parameters into the unsteady heat conduction equation: , where is the thermal diffusivity, is the Laplacian operator of temperature, which can be expressed as . In the discrete pixel environment, the finite difference method is used for approximate calculation, , , where and are the pixel spacings. By numerically solving this equation, the thermal diffusivity and the thermal decay characteristics of each pixel point are obtained, where represents the exponential coefficient of temperature decay with time, satisfying . Based on the calculation results, a temperature distribution matrix with a resolution of 640×480 is generated. Each element in the matrix contains the temperature value , thermal diffusivity and thermal decay characteristic parameters . The purpose of this step is to extract the thermophysical parameters closely related to humidity from the thermal response data through a heat conduction physical model, providing a theoretical basis for subsequent humidity analysis.

[0062] The specific implementation of step S05 is to input the temperature distribution matrix obtained in step S04 into a thermal humidity two-parameter neural network model. This neural network adopts a five-layer convolutional neural network structure, including three convolutional layers and two fully connected layers. The first convolutional layer uses 32 5×5 convolutional kernels, with a stride of 2 and a ReLU activation function, and its mathematical expression is , extracting the basic texture features of the temperature distribution matrix. The second convolutional layer uses 64 3×3 convolutional kernels, with a stride of 1 and a ReLU activation function, focusing on extracting the spatial distribution features of the thermal diffusivity. The third convolutional layer uses 128 3×3 convolutional kernels, with a stride of 1 and a ReLU activation function, used to capture the time series features of the thermal decay characteristics. The convolution operation can be expressed as , where is the output feature map, is the convolutional kernel, is the input feature map, is the convolutional kernel size. The first fully connected layer contains 512 neurons, with a ReLU activation function, combining the features extracted by the first three layers with the production process parameters (including drying temperature, drying speed, ambient humidity, and production line speed) to form a comprehensive feature vector . The second fully connected layer contains 640×480 neurons, corresponding to the resolution of the output humidity feature map, with a Sigmoid activation function, and the expression is , and the output value ranges from 0 to 1, representing the normalized humidity level. The network is trained using the Adam optimizer, with the initial learning rate set to 0.001, and dynamically adjusted using the cosine annealing strategy. The expression for the change in the learning rate is , where is the learning rate at the th step, and are the minimum and maximum learning rates respectively, is the total number of steps. The mean squared error is used as the loss function during the training process , where is the true humidity value, is the predicted humidity value, is the number of samples. After the neural network processing is completed, a humidity feature map is generated, and pseudo-color is used to display different humidity regions. Blue represents low humidity (5% - 8%), green represents standard humidity (8% - 12%), yellow represents slightly high humidity (12% - 15%), and red represents excessive humidity (>15%). The purpose of this step is to establish a mapping relationship between thermal parameters and humidity using deep learning technology to achieve precise quantitative analysis of the humidity of corrugated cardboard.

[0063] The specific implementation of step S06 is to perform region segmentation and analysis on the humidity feature map generated in step S05. First, a multi-threshold segmentation algorithm is used, and humidity threshold points are set at 8%, 12%, and 15% to segment the humidity feature map into low humidity region, standard humidity region, slightly high humidity region, and excessive humidity region. Then, the connected component labeling algorithm is used to identify the spatial distribution of each humidity region and calculate the number of pixels in each region , where represents different humidity region categories. For humidity abnormal regions (including low humidity region and excessive humidity region), calculate the percentage of its total area in the area of the entire corrugated cardboard, that is, the proportion of the area of the humidity abnormal region , where is the number of pixels in the low humidity region, is the number of pixels in the excessive humidity region, is the total number of pixels of the corrugated cardboard. At the same time, calculate the actual average humidity and the difference between the standard humidity to obtain the humidity deviation value . The calculation formula of the humidity deviation value is: . In addition, the uniformity of the humidity distribution is evaluated through spatial autocorrelation analysis, and the Moran's I index is calculated , where is the number of pixels, is the element of the spatial weight matrix, is the pixel at the humidity value, is the average humidity value. The value of this index close to 1 indicates that the humidity distribution shows a highly aggregated state, close to 0 indicates a random distribution, and close to -1 indicates a uniform distribution. The purpose of this step is to quantify the abnormal degree and spatial characteristics of the humidity distribution of corrugated cardboard to provide a decision-making basis for production process control.

[0064] The specific implementation of step S07 is to compare the humidity deviation value calculated in step S06 with the preset threshold Perform comparison and judgment. The preset threshold is set based on the quality requirements of corrugated cardboard products. The preset threshold for the standard quality grade is 3%, the preset threshold for the high-quality grade is 2%, and the preset threshold for the extra-high-quality grade is 1.5%. The system automatically selects the corresponding preset threshold according to the product grade of the current production. The comparison process uses a fuzzy logic control algorithm, which not only considers the absolute value of the humidity deviation value but also its change trend , that is, the difference between the humidity deviation value at the current moment and the previous moment. The fuzzy logic rules define the abnormal state determination conditions, such as "if is large and is large, then the degree of abnormality is high", "if is large but is small, then the degree of abnormality is medium", etc. The fuzzy membership function uses a trapezoidal function , where are the parameters of the membership function. When the humidity deviation value exceeds the preset threshold and all three consecutive detections exceed the standard, the system determines it as an abnormal state and triggers the production line abnormal alarm system. The alarm system is divided into three levels: when the humidity deviation value exceeds the preset threshold but is less than 1.5 times the preset threshold, a first-level alarm is triggered; when the humidity deviation value exceeds 1.5 times the preset threshold but is less than 2 times the preset threshold, a second-level alarm is triggered; when the humidity deviation value exceeds 2 times the preset threshold, a third-level alarm is triggered, and at the same time, the production line speed is automatically reduced by 30% to ensure product quality. The purpose of this step is to timely detect humidity abnormalities in the production process, prevent the continued production of unqualified products, and reduce the scrap rate

[0065] The specific implementation of step S08 is to perform time series analysis and prediction on the humidity deviation value. The system saves the humidity deviation value data within the most recent 72 hours, with a sampling interval of 10 minutes, forming a time series of 432 data points . First, apply the seasonal trend decomposition (STL) algorithm to decompose the time series into a trend term , a seasonal term and a residual term , satisfying . The trend term reflects the long-term change trend, the seasonal term reflects the periodic change, and the residual term is the random fluctuation. Then use the autoregressive integrated moving average (ARIMA) model to model the decomposed time series. The ARIMA model can be expressed as , where is the lag operator, is the autoregressive parameter, is the moving average parameter, is the differencing order, is the constant term, is the white noise process. The model parameters are automatically optimized and selected through the Bayesian information criterion (BIC). BIC is defined as , where is the sum of squared residuals, is the number of samples, is the number of model parameters. Based on the established ARIMA model, the humidity change trend within the next 2 hours is predicted , and the prediction accuracy is within ±0.8%. Meanwhile, the system constructs a humidity historical trend chart, including the actual humidity deviation value curve, the moving average curve (window size is 60 minutes), and the predicted humidity deviation value curve, visually showing the humidity change situation. In addition, the system uses historical data to identify which production process parameter changes are highly correlated with humidity fluctuations, and identifies key influencing factors through multiple regression analysis and Granger causality test, providing data support for optimizing production parameters. The purpose of this step is to achieve predictable control of humidity during the corrugated cardboard production process through historical data mining and predictive analysis, adjust production parameters in advance, and prevent abnormal humidity from occurring.

[0066] The specific implementation of step S09 is to establish a closed-loop control system between the humidity deviation value and production process parameters. First, a multi-input multi-output (MIMO) control model is constructed. The input variables include the current humidity deviation value , the humidity change rate , and the predicted humidity deviation value . The output variables include the drying temperature adjustment amount and the drying speed adjustment amount . The control model adopts the model predictive control (MPC) algorithm. The system state equation can be expressed as , and the output equation is , where is the state vector, including the humidity deviation value and its historical values, is the control vector, including the adjustment amounts of the drying temperature and drying speed, is the output vector, that is, the predicted future humidity deviation value, is the system matrix, obtained through system identification. The objective function of model predictive control is , where is the prediction horizon, is the control horizon, is the reference trajectory, and is the weight matrix. During the control process, the adjustment range of the drying temperature is 65°C to 95°C, and the adjustment step is 0.5°C; the adjustment range of the drying speed is 80 to 120 m / min, and the adjustment step is 2 m / min. When the detected humidity is too high, the system automatically increases the drying temperature or decreases the drying speed; when the detected humidity is too low, the system automatically decreases the drying temperature or increases the drying speed. The response time of the controller does not exceed 30 seconds, and the control accuracy reaches ±1% of the set value. In addition, the system also incorporates an adaptive control strategy, continuously updating the control model parameters through online parameter identification to adapt to changes in production conditions. The performance indicators of the control system include a steady-state error not exceeding 0.5%, an overshoot not exceeding 1.5%, and a settling time not exceeding 10 minutes. The purpose of this step is to achieve automatic closed-loop control of humidity during the corrugated cardboard production process, reduce manual intervention, and improve the stability of product quality.

[0067] The detailed structure of the thermal-humidity dual-parameter neural network model is as follows: The input layer receives a data tensor with a dimension of , which contains three channels: temperature value, thermal diffusivity, and thermal attenuation characteristics. The first layer of convolution uses 32 5×5 convolutional kernels, with a stride of 2 and a padding of 2. The convolution operation can be expressed as , where is the convolutional kernel weight tensor, is the bias vector, is the ReLU activation function, and represents the convolution operation. The output size of the first layer of convolution is , followed by a batch normalization layer, with the expression , where and are the mean and variance of the channel respectively, and are learnable parameters, and is a small constant to prevent division by zero. After batch normalization, it is followed by a max pooling layer, with a pooling kernel size of 2×2 and a stride of 2, and the expression is , and the size after pooling is . The second layer of convolution uses 64 3×3 convolutional kernels, with a stride of 1 and a padding of 1, and the expression is , and the output size is , followed by a batch normalization layer and a max pooling layer, and the size after pooling is . The third layer of convolution uses 128 3×3 convolutional kernels, with a stride of 1 and a padding of 1, and the expression is , and the output size is , followed by a batch normalization layer and a max pooling layer, and the size after pooling is . After the feature map is flattened, a vector is obtained, with a dimension of , concatenated with the production process parameters (4D vector ) to obtain the vector , which is input into the first fully connected layer, and the expression is , where is the weight matrix, is the bias vector. This layer contains 512 neurons, and dropout is used to prevent overfitting. The dropout probability is 0.3, and the expression is , where is a binary random variable with a Bernoulli distribution probability of , and represents element-wise multiplication. The output size of the second fully connected layer is , and the expression is , where is the Sigmoid activation function, is the weight matrix, and is the bias vector. The output value ranges from 0 to 1, representing the normalized humidity level, forming a humidity feature map.

[0068] The convolution kernel size of the thermal-humidity dual-parameter neural network model is matched with the corrugated structure size of the corrugated board by using a preset convolution kernel correlation function. The correlation function's calculation formula for the convolution kernel size is , where is the convolution kernel size, is the characteristic size of the corrugated structure of the corrugated board, in millimeters, is the number of pixels per millimeter, and represents rounding down. For the standard A-type corrugated structure, is approximately 8.0 - 9.0 millimeters, and the corresponding convolution kernel size is 5×5; for the standard B-type corrugated structure, is approximately 6.0 - 6.5 millimeters, and the corresponding convolution kernel size is 3×3; for the standard C-type corrugated structure, is approximately 7.0 - 7.5 millimeters, and the corresponding convolution kernel size is 3×3 or 5×5; for the E-type and F-type corrugated structures, is smaller, and the corresponding convolution kernel size is 3×3.

[0069] The process of establishing the training data set for the thermal-humidity dual-parameter neural network model is as follows: First, collect samples from the actual production line. Collect thermal response images under different humidity conditions (humidity range 5% - 20%, interval 0.5%). 50 groups of samples are collected for each humidity condition, totaling 1500 groups of original samples. For each sample, use a high-precision contact humidity sensor (accuracy ±0.2%) to synchronously measure the actual humidity value as the label data. At the same time, record the production condition information of the samples, including the corrugated board thickness ​​​​​​​​​​(Range: 1.5 to 7 mm, with 5 grades), corrugation types of corrugated cardboard (There are 5 types in total: A type, B type, C type, E type, and F type), raw material components of corrugated cardboard (There are 8 different ratios in total), forming a feature vector with a total dimension of 23. Apply data augmentation techniques to the original samples, including: random rotation (angle range ±15°), random scaling (scaling factor range 0.9 to 1.1), random addition of Gaussian noise (mean is 0, standard deviation is 0.01), random adjustment of brightness (adjustment range ±10%), random adjustment of contrast (adjustment range ±10%), random horizontal and vertical flipping. Each original sample generates 5 augmented samples, and finally 10,000 groups of training samples and 2,000 groups of validation samples are formed. The training set and the validation set are divided using the stratified sampling method to ensure that the proportions of different humidity conditions and different corrugated cardboard types in the training set and the validation set are the same.

[0070] In this embodiment, the design that the convolution kernel size of the thermal humidity dual-parameter neural network model matches the corrugated cardboard corrugation structure size ensures that the neural network can accurately capture the characteristic information of different types of corrugated structures. This matching relationship makes the receptive field of the convolutional layer correspond to the actual corrugation physical size, improving the model's recognition ability for local humidity distribution characteristics. By selecting appropriate sizes of convolution kernels for different specifications of corrugated cardboard (such as A type, B type, C type, etc.), the system can adapt to various product specifications, reduce the model generalization error, and at the same time improve the accuracy and adaptability of humidity detection, providing technical guarantee for realizing high-precision humidity monitoring.

[0071] Types A, B, and C refer to different structural types of corrugated cardboard, and these classifications are industry standards based on the shape dimensions and characteristics of the corrugations.

[0072] Type A corrugation (also known as large corrugation) has a relatively large corrugation height and period, and the characteristic dimensions of its corrugation structure are approximately 8.0 to 9.0 mm. This structure provides relatively high compressive strength and buffering performance, and is usually used for packaging products that require strong protection performance.

[0073] Type B corrugation (also known as medium corrugation) has a smaller corrugation structure than Type A, and the characteristic dimensions are approximately 6.0 to 6.5 mm. It provides good compressive performance and a printable surface, and performs well in balancing strength and printing quality, and is widely used in general commodity packaging.

[0074] Type C corrugation is between Type A and Type B, and the characteristic dimensions of its corrugation structure are approximately 7.0 to 7.5 mm. This structure has both relatively good compressive strength and stacking performance, and is one of the most widely used corrugation types in the packaging industry, suitable for packaging a variety of medium-weight products.

[0075] The structural characteristics of different corrugation types affect the heat conduction behavior. Therefore, when designing a humidity monitoring system, it is necessary to select a convolution kernel of an appropriate size for different types to match their physical characteristics.

[0076] To better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: Workers implemented an on-line humidity monitoring system based on thermal infrared imaging on a certain corrugated cardboard production line. This production line mainly produces type B corrugated cardboard for high-end electronic product packaging, with strict requirements for the humidity uniformity of the cardboard. The standard humidity is 9.5%, and the allowable deviation does not exceed 1.5%. The system is installed at the outlet of the drying section of the production line, and the monitoring area covers the entire width of the cardboard, which is 107×78 cm.

[0077] The thermal infrared imaging device uses a non-cooled microbolometer focal plane array detector of the FLIR A655sc model, with a working wavelength range of 7.5~14μm, a temperature resolution of 0.03℃, and a field of view angle set to 62°×45°. The installation height of the device is 95 cm, the imaging resolution is set to 640×480 pixels, and each pixel corresponds to an area of approximately 1.67×1.63 mm on the cardboard surface. The sampling frequency is set to 25 Hz.

[0078] The thermal pulse excitation system consists of 25 infrared lamps with a single power of 250 watts, arranged in a 5×5 matrix. The pulse intensity is maintained at 5000 watts per square meter, and the duration is precisely controlled to be 0.5 seconds. The system monitors the spatial distribution of thermal pulses in real time, and the deviation within the entire area is controlled within ±2.5% to ensure the consistency of thermal excitation.

[0079] The workers collected multiple sets of thermal response data under different humidity conditions and conducted a detailed analysis. As shown in Table 1, the thermal parameters under four typical humidity states show obvious differences.

[0080] Table 1 Thermal parameter characteristics table under different humidity conditions

[0081] The data in Table 1 show that as the humidity increases, the peak temperature and thermal diffusivity decrease significantly, the thermal attenuation parameter decreases, and the thermal response duration prolongs. This provides an important training basis for the thermal humidity dual-parameter neural network model. Figure 2Shows the thermal response curves of corrugated cardboard to thermal pulse excitation under different humidity conditions. The figure contains four curves, representing the temperature variation with time under low humidity (6.0% - 7.5%), standard humidity (8.0% - 11.0%), slightly high humidity (11.5% - 14.0%), and excessive humidity (14.5% - 17.0%) conditions respectively. It can be seen from the figure that the corrugated cardboard with low humidity has the highest peak temperature (about 46.5°C) and the fastest thermal decay rate (β = 0.30 s^-1), and the shortest thermal response duration; while with the increase of humidity, the peak temperature gradually decreases, the thermal decay rate slows down, and the thermal response duration prolongs. The thermal pulse excitation stage (0.5 seconds) and the thermal response duration threshold line (reference temperature + 1°C) are also marked in the figure.

[0082] The training of the thermal - humidity two - parameter neural network model used 10,000 groups of samples, including diverse data of different humidity conditions and different types of corrugated cardboard. The model verification performance is shown in Table 2.

[0083] Table 2 Verification Performance Table of Thermal - Humidity Two - Parameter Neural Network Model

[0084] The data in Table 2 show that the model has a high prediction accuracy for various types of corrugated cardboard, with an average root - mean - square error of only 0.44%, an average absolute error of 0.35%, and an average detection time of 639 milliseconds, fully meeting the requirements of on - line monitoring. Figure 3 Details the training process and network structure of the thermal - humidity two - parameter neural network model. The main figure shows the change curves of training loss and validation loss with the number of training epochs. It can be seen that as the number of training epochs increases, both loss values show a downward trend and reach an early stopping point at about 78 training epochs, at which time the validation loss no longer decreases.

[0085] The staff conducted region - segmentation analysis on the predicted humidity feature maps, as shown in Table 3.

[0086] Table 3 Humidity Region Analysis Table during a Certain Production Process

[0087] Table 3 shows that the overall humidity of this batch of corrugated cardboard is qualified, but there are 14.64% of abnormal regions, mainly concentrated on the cardboard edges and local areas. The system triggered a first - level alarm accordingly and at the same time initiated closed - loop control to automatically adjust the drying parameters. The closed - loop control system automatically adjusts the production parameters according to the humidity deviation value, as shown in Table 4.

[0088] Table 4 Closed - Loop Control Parameter Adjustment Table

[0089] Table 4 shows that the closed-loop control system automatically fine-tunes the drying temperature and speed according to the humidity deviation, gradually reducing the humidity deviation from the initial 1.8% to 0.8%, meeting the quality requirements of high-end products.

[0090] The staff statistically analyzed the long-term operation data of the system, and Table 5 shows the quality improvement effects before and after the implementation of the system.

[0091] Table 5 Comparison Table of Quality Indicators Before and After System Implementation

[0092] The data in Table 5 show that after the implementation of the system, the humidity control accuracy of corrugated board has been significantly improved, the unqualified product rate has decreased by 71.0%, customer complaints have decreased by 78.1%, and the energy consumption in the production process has decreased by 9.1%. Figure 4 It shows the automatic adjustment process of production parameters by the closed-loop control system and the quality improvement effect. The main curve (red) shows the change of humidity deviation value over time, gradually decreasing from the initial 1.8% to stabilizing at about 0.6%, which is lower than the preset threshold of 1.5%. The figure also includes the change curves of drying temperature (blue) and drying speed (green) before and after adjustment. It can be seen that the system first increases the drying temperature and decreases the drying speed to reduce the humidity, and then makes fine adjustments to maintain during the stable stage. The control process is divided into three stages: initial regulation (0 - 20 minutes), fine regulation (20 - 50 minutes), and stable maintenance (50 - 90 minutes). The quality improvement effects before and after the implementation of the system are given in the lower left corner of the figure, including a 63.7% reduction in average humidity deviation, a 68.1% reduction in humidity standard deviation, a 71.0% reduction in unqualified product rate, and a 9.1% reduction in energy consumption. The controller performance indicators are shown in the lower right corner, including a steady-state error less than 0.5%, an overshoot less than 1.5%, and a regulation time less than 10 minutes, reflecting the excellent performance of this control system.

[0093] The time series analysis module established a humidity historical trend chart and predicted future changes. As shown in Table 6, the prediction accuracy decreases with the extension of the prediction time, but the prediction error within 2 hours is controlled within ±0.8%, meeting the production regulation requirements.

[0094] Table 6 Humidity Prediction Accuracy Evaluation Table

[0095] Traditional corrugated cardboard humidity detection mainly uses contact humidity sensors or off-line sampling detection methods, which have problems such as limited measurement points, inability to cover the entire area, inability to adjust online in real time, and large interference to the production process. The specific manifestations are as follows: Contact sensors are easily worn and need to be replaced frequently; they can only measure the humidity at a few fixed points and cannot obtain the humidity distribution of the entire cardboard; off-line sampling detection has obvious lag and it is difficult to discover and solve problems in a timely manner; the processing performance and finished product quality of corrugated cardboard under different humidity states vary greatly, and it is difficult to establish an effective quality prediction model with traditional methods.

[0096] The infrared thermal imaging humidity on-line monitoring method of the present invention overcomes the limitations of traditional methods and realizes non-contact, full-coverage, high-precision, and real-time monitoring of the humidity on the entire surface of corrugated cardboard. By combining thermal pulse excitation with thermal infrared imaging technology, the thermal response differences in different humidity regions are captured; using the unsteady heat conduction equation and neural network model, an accurate mapping relationship between temperature characteristics and humidity is established; integrating functions such as humidity anomaly recognition, automatic alarm, and closed-loop control, intelligent adjustment of production parameters is realized; introducing time series analysis and prediction modules, the humidity change trend can be predicted in advance to provide support for production decisions. Through these technological innovations, this system significantly improves the quality stability and production efficiency of corrugated cardboard production, reduces energy consumption and the rejection rate, and provides an effective solution for the intelligent upgrade of the corrugated cardboard manufacturing industry.

[0097] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 7 and 8 below.

[0098] Table 7 Variable Explanation Table (Part 1)

[0099] Table 8 Variable Explanation Table (Part 2)

[0100] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A method for on-line automatic monitoring of the humidity of corrugated cardboard, characterized in that, Including: A thermal infrared imaging device is used at a fixed position on the corrugated cardboard production line, and the focal length is adjusted to cover the entire width of the corrugated cardboard; A thermal pulse excitation system is used to apply a uniform thermal pulse to the surface of the corrugated cardboard; the thermal response image of the corrugated cardboard surface is collected by the thermal infrared imaging device; According to the thermal response image, based on the unsteady heat conduction equation, the thermal diffusivity and thermal decay characteristics of each pixel point of the corrugated cardboard are calculated to generate a temperature distribution matrix; based on the thermal humidity dual-parameter neural network model, the temperature distribution matrix is analyzed to extract the humidity characteristic map, and the corresponding relationship between the temperature distribution matrix and the water content of the corrugated cardboard is established; regional segmentation is performed through the humidity characteristic map to identify the humidity abnormal area, and the area ratio and humidity deviation value of the humidity abnormal area are calculated; the humidity deviation value is compared with the preset threshold, and if it exceeds the preset threshold, an alarm is triggered.

2. The method for on-line automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, Specifically, the thermal pulse excitation system forms a uniform heat source by instantaneously releasing heat energy through a high-power infrared lamp array, which is used to stimulate the surface of the corrugated cardboard under test to generate a thermal response image.

3. The method for on-line automatic monitoring of the humidity of corrugated cardboard according to claim 2, characterized in that, The thermal pulse intensity is maintained at 5000 watts per square meter, and the thermal pulse duration is 0.5 seconds.

4. The method for online automatic monitoring of the humidity of corrugated cardboard according to claim 1, wherein, The resolution of the thermal response image collected by the thermal infrared imaging device is set not less than 640×480 pixels, and the sampling frequency of the thermal infrared imaging device is 25 Hz.

5. The method for online automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The thermal diffusivity refers to the parameter of the heat conduction ability of the corrugated cardboard. Since the increase in moisture content will reduce the thermal diffusivity, the area with higher humidity shows the characteristics of more heat accumulation and slower heat dissipation.

6. The method for on-line automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The thermal decay characteristic refers to the curve of the temperature change of the corrugated cardboard over time after absorbing heat energy. By analyzing the differences in the thermal decay characteristics of different regions, the internal water content distribution of the corrugated cardboard is reflected.

7. The method for on-line automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The temperature distribution matrix refers to a two-dimensional array composed of the temperature values of each pixel point of the thermal response image, which is used to quantitatively describe the temperature distribution state of the entire corrugated cardboard surface.

8. The method for online automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The humidity deviation value refers to the difference value between the actual measured humidity and the standard humidity.

9. The method for on-line automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The unsteady heat conduction equation is used to describe the dynamic process of heat transfer in the corrugated cardboard and calculate the thermal diffusivity and thermal decay characteristics. The inputs include the surface temperature change rate of the corrugated cardboard, the initial temperature distribution of the corrugated cardboard, the specific heat capacity of the corrugated cardboard material, the density of the corrugated cardboard material, and the thermal conductivity of the corrugated cardboard material, and the outputs are the thermal diffusivity and thermal decay characteristics of each point of the corrugated cardboard.

10. The method for online automatic monitoring of the humidity of corrugated cardboard according to claim 1, characterized in that, The structure of the thermal humidity dual-parameter neural network model is a five-layer convolutional neural network structure, including three convolutional layers and two fully connected layers. The first convolutional layer is used to extract the basic features of the temperature distribution matrix, the second convolutional layer is used to extract the spatial distribution features of the thermal diffusivity, the third convolutional layer is used to extract the time series features of the thermal decay characteristics, the first fully connected layer fuses the three features and combines them with the production process parameters, and the second fully connected layer outputs the humidity characteristic map.

Images