A method for online automatic monitoring of corrugated cardboard humidity

Abstract

The present invention provides a method for automatically monitoring the humidity of corrugated cardboard online, belonging to the technical field of corrugated cardboard production. The present invention uses a thermal infrared imaging device combined with a thermal pulse excitation system to apply uniform heat pulses to the cardboard surface, capturing high-resolution thermal response images in real time. The system calculates thermal diffusivity and thermal attenuation characteristics based on a non-steady-state heat conduction equation, generates a temperature distribution matrix, and uses a thermal-humidity dual-parameter neural network model to extract a humidity characteristic map, establishing a corresponding relationship between temperature and moisture content. Regional segmentation is performed using the humidity characteristic map, identifying areas with abnormal humidity and calculating deviation values. When the deviation exceeds a preset threshold, an alarm is triggered. This method solves the technical problem in the prior art that corrugated cardboard humidity detection primarily relies on manual sampling and specialized instruments, making it impossible to achieve real-time online monitoring on the production line.

Description

Technical Field

[0001] The invention belongs to the technical field of corrugated cardboard production, and in particular relates to a method for online automatic monitoring of the humidity of corrugated cardboard. Background Art

[0002] Corrugated cardboard, a fundamental material in the packaging industry, has its physical properties and service life significantly impacted by its moisture content. Traditional corrugated cardboard moisture testing relies primarily on manual sampling. This involves sampling off the production line and analyzing the sample using specialized moisture measurement instruments (such as drying and weighing methods, resistance hygrometers, and capacitance moisture detectors). This method has been used in the packaging industry for decades, resulting in a relatively mature process and high accuracy within a controlled laboratory environment.

[0003] However, traditional manual sampling testing methods have significant drawbacks: First, sampling testing has a significant lag, often requiring several minutes or even longer from sampling to obtaining test results, making it impossible to respond promptly to humidity changes during the production process. Second, the randomness and limited nature of sampling make it difficult to fully reflect the moisture distribution of an entire batch of products, and can easily miss localized abnormal areas. Third, subjective factors influence the manual operation process, resulting in poor consistency in test results. Finally, this offline testing method cannot form an automatic feedback adjustment mechanism with production equipment, making it difficult to achieve intelligent production control. In other words, the existing technology for corrugated cardboard moisture testing relies primarily on manual sampling and specialized instruments, making it impossible to achieve real-time online monitoring on the production line. Summary of the Invention

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

[0005] The present invention is implemented as follows: The present invention provides a method for online automatic monitoring of the humidity of corrugated cardboard, comprising: using a thermal infrared imaging device at a fixed position on a 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 corrugated cardboard surface by the thermal infrared imaging device; calculating the thermal diffusivity and thermal attenuation characteristics of each pixel point of the corrugated cardboard based on the thermal response image and a non-steady-state heat conduction equation, and generating a temperature distribution matrix; analyzing the temperature distribution matrix based on a thermal and humidity dual-parameter neural network model, extracting a humidity characteristic map, and establishing a corresponding relationship between the temperature distribution matrix and the moisture content of the corrugated cardboard; performing regional segmentation through the humidity characteristic map, identifying humidity abnormality areas, and calculating the area ratio of the humidity abnormality areas and the humidity deviation value; comparing the humidity deviation value with a preset threshold value, and triggering an alarm if the preset threshold value is exceeded.

[0006] Among them, the thermal pulse excitation system specifically forms a uniform heat source by instantly releasing heat energy through a high-power infrared lamp array, which is used to stimulate the surface of the corrugated cardboard being tested to produce a thermal response image.

[0007] Among them, the heat pulse intensity is maintained at 5000 watts per square meter and the heat 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 be no less than 640×480 pixels, and the sampling frequency of the thermal infrared imaging device is 25 Hz.

[0009] Among them, thermal diffusivity refers to the parameter of corrugated cardboard's ability to conduct heat. Since an increase in moisture content will reduce the thermal diffusivity, areas with higher humidity will show more heat accumulation and slower heat dissipation.

[0010] Among them, the thermal attenuation characteristics refer to the curve of the temperature drop over time after the corrugated cardboard absorbs heat energy. By analyzing the differences in thermal attenuation characteristics in different areas, the distribution of moisture content inside the corrugated cardboard can be reflected.

[0011] The temperature distribution matrix refers to a two-dimensional array of the temperature values ​​of each pixel in the thermal response image, which is used to quantitatively describe the temperature distribution state of the entire corrugated cardboard surface.

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

[0013] Among them, the non-steady-state heat conduction equation is used to describe the dynamic process of heat transfer in corrugated cardboard and calculate the thermal diffusivity and 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 coefficient of the corrugated cardboard material. The output is the thermal diffusivity and thermal attenuation characteristics of each point on the corrugated cardboard.

[0014] Among them, the thermal and humidity dual-parameter neural network model structure is a five-layer convolutional neural network structure, which includes 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 thermal diffusivity, and the third convolutional layer is used to extract the time series features of 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.

[0015] The humidity characteristic map 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.

[0016] This also includes time series analysis of humidity deviation values ​​and optimization of production parameters; associating humidity deviation values ​​with production process parameters, and automatically adjusting drying temperature and drying speed through a feedback control system to achieve closed-loop control.

[0017] This invention utilizes non-contact detection technology combining thermal infrared imaging with heat pulse excitation to achieve real-time, full-width moisture monitoring on corrugated cardboard production lines. The system utilizes a high-frequency sampling rate of 25 Hz and high-resolution imaging of at least 640 x 480 pixels, ensuring continuous capture of the moisture status of cardboard on the fast-moving production line, fundamentally addressing the lag inherent in traditional sampling detection.

[0018] This method combines heat conduction theory with artificial intelligence algorithms to establish a precise mapping between temperature response characteristics and moisture distribution. The system generates a real-time moisture profile across the entire width of the cardboard, automatically identifying abnormal areas and calculating deviations. This expands moisture monitoring from traditional point measurements to surface measurements, comprehensively improving detection coverage and accuracy. Furthermore, the system's integrated time series analysis and feedback control capabilities automatically adjust production process parameters based on real-time monitoring data, achieving a transition from passive detection to active control.

[0019] In summary, the present invention solves the technical problem that the humidity detection of corrugated cardboard mainly relies on manual sampling and special instruments, and cannot realize real-time online monitoring on the production line. It provides a new fully automatic, intelligent and efficient quality control method for corrugated cardboard production. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the method of the present invention.

[0021] Figure 2 This is a thermal response curve diagram under different humidity conditions in Example 2.

[0022] Figure 3This is a diagram of the training process of the thermal and humidity dual-parameter neural network model in Example 2.

[0023] Figure 4 This is a diagram showing the effect of parameter adjustment and quality improvement of the closed-loop control system in Example 2. DETAILED DESCRIPTION

[0024] In order to make the purpose, 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] like Figure 1 FIG. 1 is a flow chart of a method for online automatic monitoring of corrugated cardboard humidity provided by the present invention, and the method comprises the following steps:

[0026] S01, using a thermal infrared imaging device at a fixed position on a corrugated cardboard production line, and adjusting the focal length of the thermal infrared imaging device so that the imaging area covers the entire width of the corrugated cardboard;

[0027] S02, applying a uniform heat pulse to the surface of the corrugated cardboard using a heat pulse excitation system, wherein the heat pulse intensity is maintained at 5000 watts per square meter and the heat pulse duration is 0.5 seconds;

[0028] S03, collecting a thermal response image of the corrugated cardboard surface by the thermal infrared imaging device, wherein the resolution of the thermal response image is set to be no less than 640×480 pixels, and the sampling frequency of the thermal infrared imaging device is 25 Hz;

[0029] S04. Calculating the thermal diffusivity and thermal attenuation characteristics of each pixel of the corrugated cardboard based on the thermal response image and a non-steady-state heat conduction equation to generate a temperature distribution matrix;

[0030] S05. Analyze the temperature distribution matrix based on a thermal-humidity dual-parameter neural network model, extract the humidity characteristic map, and establish a corresponding relationship between the temperature distribution matrix and the moisture content of the corrugated cardboard;

[0031] S06. Segment the area using the humidity characteristic map to identify the humidity abnormality area, and calculate the area ratio and humidity deviation value of the humidity abnormality area;

[0032] S07, comparing the humidity deviation value with a preset threshold value, and triggering a production line abnormality alarm system when the humidity deviation value exceeds the preset threshold value;

[0033] S08. Optionally, the method further includes performing a time series analysis on the humidity deviation value, establishing a humidity historical trend graph, predicting future humidity changes, and optimizing production parameters;

[0034] S09. Optionally, the method 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.

[0035] The heat 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 excite the surface of the corrugated cardboard to produce the thermal response image.

[0036] The thermal diffusivity specifically refers to the parameter of the corrugated board's ability to conduct heat. Since an increase in moisture content will reduce the thermal diffusivity, areas with higher humidity exhibit the characteristics of greater heat accumulation and slower heat dissipation.

[0037] The thermal attenuation characteristic specifically refers to a curve showing a temperature drop over time after the corrugated cardboard absorbs heat energy. By analyzing the differences in the thermal attenuation characteristics of different regions, the distribution of moisture content inside the corrugated cardboard can be reflected.

[0038] The temperature distribution matrix specifically refers to a two-dimensional array composed of the temperature value 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.

[0039] 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.

[0040] The humidity deviation value specifically refers to the difference between the actual measured humidity and the standard humidity, expressed as a percentage. It is usually required that the humidity deviation value be controlled within 3%.

[0041] Among them, the closed-loop control specifically refers to the system automatically adjusting the drying temperature and the drying speed according to the humidity deviation value monitored in real time, forming a cycle of detection, analysis, adjustment, and re-detection, thereby maintaining stable product quality without human intervention.

[0042] The preset threshold specifically refers to the upper limit of the humidity deviation value preset according to the quality requirements of the corrugated cardboard product. When the humidity deviation value exceeds the preset threshold, the system determines that it is in an abnormal state.

[0043] The production process parameters specifically refer to key parameters that affect humidity control during the production of the corrugated cardboard, including the drying temperature, the drying speed, ambient humidity, and production line speed.

[0044] Among them, the unsteady-state 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 coefficient of the corrugated cardboard material. The output is the thermal diffusivity and the thermal attenuation characteristics of each point on 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 coefficient of the corrugated cardboard material are derived from the corrugated cardboard material database; the thermal diffusivity is used to generate the temperature distribution matrix; and the thermal attenuation characteristics are used to analyze the moisture content distribution inside the corrugated cardboard.

[0045] The specific structure of the thermal and humidity dual-parameter neural network model is a five-layer convolutional neural network structure, which includes 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, and 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 convolution kernel size in the thermal and humidity dual-parameter neural network model structure uses a preset convolution kernel correlation function to match the corrugated structure size of the corrugated cardboard, so as to capture the differences in humidity distribution in different corrugated structures.

[0046] The steps of establishing a training data set for the thermal and humidity dual-parameter neural network model specifically include collecting the thermal response image samples under different humidity conditions from an actual production line, synchronously collecting 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 multidimensional feature vector, and using data enhancement technology to increase sample diversity including rotation, scaling, and noise processing, ultimately forming a training set containing 10,000 groups of samples and a validation set containing 2,000 groups of samples; the actual humidity measurement values ​​are derived from a contact humidity sensor; and the multidimensional feature vector is used to enhance the adaptability of the thermal and humidity dual-parameter neural network model to different types of corrugated cardboard.

[0047] The steps of training the thermal and humidity dual-parameter neural network model specifically include using a batch gradient descent algorithm to optimize the weights of the thermal and humidity dual-parameter neural network model, the learning rate is initially 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 and humidity dual-parameter neural network model, and an early stopping strategy is introduced during the training process to prevent overfitting. When the loss function value of the validation set does not decrease for five consecutive training cycles, the training is stopped, and finally the parameters of the thermal and 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 and humidity dual-parameter neural network model; the early stopping strategy is used to prevent the thermal and humidity dual-parameter neural network model from overfitting the training data, resulting in a decrease in generalization ability.

[0048] The specific implementation of the above steps is described in detail below.

[0049] The specific implementation method of step S01 is to select an appropriate location on the corrugated cardboard production line to fix the thermal infrared imaging device. This location must ensure unobstructed observation of the entire width of the corrugated cardboard. The thermal infrared imaging device uses an uncooled microbolometer focal plane array detector with an operating wavelength range of 7.5 to 14 μm, a temperature resolution of not less than 0.05°C, and a field of view set to 60°×45°. After installation is completed, the focus is precisely adjusted. During the adjustment process, a calibration plate with equidistant reference points is temporarily placed on the corrugated cardboard production line. The calibration plate is marked with equidistant reference points. The thermal infrared imaging device's electric focusing mechanism is used to perform remote fine adjustments until the image is clear and covers the entire corrugated cardboard width of 107×78 cm. The purpose of this step is to establish a stable and reliable thermal imaging observation foundation to ensure the integrity and accuracy of subsequent thermal response data acquisition.

[0050] The specific implementation of step S02 involves activating a heat pulse excitation system to apply a uniform heat pulse to the surface of the corrugated cardboard. This heat pulse excitation system consists of a high-power infrared lamp array consisting of 25 individual 250-watt infrared lamps arranged in a 5×5 matrix, with a coverage area that matches the width of the corrugated cardboard. Before the heat pulse is generated, a power controller precisely regulates the input current to the rated value, ensuring that the heat pulse intensity reaches 5000 watts per square meter. After the heat pulse begins, a precision timer controls the heat pulse duration to 0.5 seconds with a timing accuracy of ±0.01 seconds. During the heat pulse, a uniformity monitoring system monitors the spatial distribution of the heat pulse in real time, ensuring that the heat pulse intensity does not vary by more than ±3% across the entire area. This step creates standardized thermal excitation conditions on the corrugated cardboard surface, providing controllable and repeatable energy input for subsequent thermal response-based moisture analysis.

[0051] The specific implementation of step S03 is to activate the thermal infrared imaging device and immediately begin capturing thermal response images of the corrugated cardboard surface after the heat 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 thermal infrared imaging device sampling frequency is set to 25 Hz, meaning that one thermal image frame is captured every 40 milliseconds for 25 seconds, for a total of 625 frames. During the acquisition process, the thermal infrared imaging device's temperature drift compensation system operates continuously, using a built-in blackbody reference source for real-time calibration to ensure temperature measurement accuracy within ±0.5°C. After image acquisition is completed, the thermal response image sequence is transmitted to the image processing unit via a high-speed data transmission interface for storage and preprocessing. The purpose of this step is to obtain the temporal response sequence of the corrugated cardboard to the heat pulse stimulus and capture the temperature change characteristics caused by differences in thermal conductivity characteristics in different humidity areas.

[0052] The specific implementation of step S04 is to process the thermal response image based on the non-steady-state heat conduction equation in the image processing unit. First, the original thermal image sequence is spatially filtered, and a 5×5 Gaussian filter is used to eliminate random noise. The filter standard deviation is set to 1.2 pixels. Then, the temperature change rate of each pixel between consecutive frames is calculated to construct a temperature change rate field. For each pixel, the initial temperature before the heat pulse is extracted. , peak temperature after heat pulse , the relationship curve between temperature and time during the cooling process . Substituting these parameters into the unsteady heat conduction equation: ,in is the thermal diffusivity, is the Laplace operator of temperature. By numerically solving this equation, the thermal diffusivity of each pixel is obtained and thermal attenuation characteristics ,in The exponential coefficient representing the temperature decay over time satisfies 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 attenuation characteristic parameters at that point. The purpose of this step is to extract thermophysical parameters closely related to humidity from the thermal response data using a heat conduction physical model, providing a theoretical basis for subsequent humidity analysis.

[0053] The specific implementation of step S05 is to input the temperature distribution matrix obtained in step S04 into a dual-parameter thermal and humidity neural network model. This neural network adopts a five-layer convolutional neural network structure, consisting of three convolutional layers and two fully connected layers. The first convolutional layer uses 32 5×5 convolution 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 convolution kernels with a stride of 1 and a ReLU activation function, focusing on extracting the spatial distribution characteristics of thermal diffusivity. The third convolutional layer uses 128 3×3 convolution kernels with a stride of 1 and a ReLU activation function to capture the time series characteristics of thermal attenuation characteristics. The first fully connected layer contains 512 neurons with a ReLU activation function. It combines the features extracted by the first three layers with 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 output 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. The network is trained using the Adam optimizer, with an initial learning rate of 0.001 and a cosine annealing strategy for dynamic adjustment. Mean squared error is used as the loss function during training, achieving a humidity prediction accuracy of ±1.5% on the validation set. After the neural network is processed, a humidity feature map is generated, displaying different humidity regions using pseudo-color. Blue represents low humidity (5%-8%), green represents standard humidity (8%-12%), yellow represents slightly elevated 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 techniques, enabling accurate quantitative analysis of corrugated cardboard humidity.

[0054] The specific implementation of step S06 involves regional segmentation and analysis of the humidity characteristic map generated in step S05. First, a multi-threshold segmentation algorithm is used, with humidity thresholds set at 8%, 12%, and 15%. The humidity characteristic map is segmented into low humidity, standard humidity, slightly high humidity, and excessively high humidity regions. A connected component labeling algorithm is then used to identify the spatial distribution of each humidity region and calculate the number of pixels in each region. For regions with abnormal humidity (including low humidity and excessively high humidity), the percentage of their total area relative to the total corrugated cardboard area is calculated to obtain the area ratio of the abnormal humidity region. Simultaneously, the difference between the actual average humidity and the standard humidity (10%) is calculated to obtain the humidity deviation value. The humidity deviation value is calculated as: Humidity deviation value = |actual average humidity - standard humidity| / standard humidity × 100%. Furthermore, spatial autocorrelation analysis is used to assess the uniformity of the humidity distribution, calculating the Moran's index (I). Values ​​close to 1 indicate a highly clustered humidity distribution, those close to 0 indicate a random distribution, and those close to -1 indicate a uniform distribution. The purpose of this step is to quantify the abnormal degree and spatial characteristics of the moisture distribution of corrugated cardboard and provide a decision-making basis for production process control.

[0055] The specific implementation of step S07 is to compare the humidity deviation value calculated in step S06 with a preset threshold value. The preset threshold value is set based on the quality requirements of corrugated cardboard products: the preset threshold value for standard quality grade is 3%, the preset threshold value for high quality grade is 2%, and the preset threshold value for ultra-high quality grade is 1.5%. The system automatically selects the corresponding preset threshold value based on the grade of the product currently being produced. The comparison process uses a fuzzy logic control algorithm, which considers not only the absolute value of the humidity deviation value but also its changing trend. If the humidity deviation value exceeds the preset threshold and exceeds the standard for three consecutive tests, the system determines that it is in an abnormal state and triggers the production line abnormality alarm system. The alarm system has three levels: A level 1 alarm is triggered when the humidity deviation exceeds the preset threshold but is less than 1.5 times the threshold, alerting the operator via a yellow indicator light and a gentle alarm on the operation panel. A level 2 alarm is triggered when the humidity deviation exceeds 1.5 times the threshold but is less than 2 times the threshold, alerting the operator via an orange indicator light and a medium-volume alarm on the operation panel, and recording the abnormal event. A level 3 alarm is triggered when the humidity deviation exceeds 2 times the threshold, alerting the operator via a flashing red indicator light and a high-volume alarm on the operation panel, while also automatically reducing the production line speed by 30% to ensure product quality. This step aims to promptly detect humidity anomalies during production, prevent the continued production of substandard products, and reduce scrap rates.

[0056] Step S08 is optional and involves performing time series analysis and forecasting on humidity deviation values. The system stores humidity deviation data for the past 72 hours, sampling at a 10-minute interval, resulting in a time series of 432 data points. The seasonal trend decomposition (STL) algorithm is first applied to decompose the time series into trend, seasonal, and residual terms, identifying potential cyclical changes and long-term trends. The decomposed time series is then modeled using the autoregressive integrated moving average (ARIMA) model, with model parameters automatically optimized using the Bayesian Information Criterion. Based on the established ARIMA model, humidity trends are predicted for the next two hours with an accuracy within ±0.8%. The system also constructs a historical humidity trend chart, including a curve of actual humidity deviation values, a moving average curve (with a 60-minute window), and a curve of predicted humidity deviation values, providing a visual display of humidity fluctuations. Furthermore, the system uses historical data to identify which production process parameter changes are highly correlated with humidity fluctuations. Multiple regression analysis or Granger causality testing identifies key influencing factors, 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 humidity anomalies.

[0057] Step S09 is optional and involves establishing 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 and the drying speed adjustment. This control model utilizes a model predictive control (MPC) algorithm to optimize control behavior while satisfying constraints based on the identified system dynamics. During the control process, the drying temperature is adjustable from 65°C to 95°C in 0.5°C increments, and the drying speed is adjustable from 80 to 120 m / min in 2 m / min increments. When high humidity is detected, the system automatically increases the drying temperature or decreases the drying speed; when low humidity is detected, the system automatically decreases the drying temperature or increases the drying speed. The controller's response time is no more than 30 seconds, and control accuracy reaches ±1% of the set value. Furthermore, the system incorporates an adaptive control strategy that continuously updates control model parameters through online parameter identification to adapt to changing production conditions. The control system's performance indicators include a steady-state error of no more than 0.5%, an overshoot of no more than 1.5%, and a settling time of no more than 10 minutes. This step aims to achieve automatic closed-loop control of humidity during corrugated cardboard production, reducing manual intervention and improving product quality stability.

[0058] The detailed structure of the thermal and humidity dual-parameter neural network model is as follows: The input layer receives data of 640×480×3 dimensions, including three channels: temperature, thermal diffusivity, and thermal attenuation. The first convolution layer uses 32 5×5 convolution kernels with a stride of 2 and padding of 2, resulting in an output size of 320×240×32. The activation function is Reinforced Lu (ReLU). This layer is followed by a batch normalization layer and a max pooling layer (with a pooling kernel size of 2×2 and a stride of 2), resulting in a pooled size of 160×120×32. The second convolution layer uses 64 3×3 convolution kernels with a stride of 1 and padding of 1, resulting in an output size of 160×120×64. The activation function is Reinforced Lu (ReLU). This layer is followed by a batch normalization layer and a max pooling layer (with a pooling kernel size of 2×2 and a stride of 2), resulting in a pooled size of 80×60×64. The third convolutional layer uses 128 3×3 convolution kernels with a stride of 1 and padding of 1, resulting in an output size of 80×60×128. The activation function is ReLU. This is followed by a batch normalization layer and a max pooling layer (with a pooling kernel size of 2×2 and a stride of 2), resulting in a pooled size of 40×30×128. The feature map is flattened and concatenated with the production process parameters (a 4-dimensional vector). This input is then fed into the first fully connected layer, which contains 512 neurons and uses the ReLU activation function. Dropout (at a ratio of 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.

[0059] The training dataset for the thermal-humidity dual-parameter neural network model was established as follows: First, samples were collected from actual production lines. Thermal response images were captured under different humidity conditions (ranging from 5% to 20% in 0.5% increments). Fifty sets of samples were collected for each humidity condition, totaling 1,500 sets of raw samples. For each sample, the actual humidity value was simultaneously measured using a high-precision contact humidity sensor (accuracy ±0.2%) as label data. Production condition information was also recorded, including corrugated cardboard thickness (ranging from 1.5 to 7 mm, with five levels), corrugated cardboard corrugation type (A, B, C, E, and F), and the composition of the corrugated cardboard raw materials (eight different ratios), to form a 23-dimensional feature vector. Data augmentation techniques were 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 brightness adjustment (adjustment range ±10%), random contrast adjustment (adjustment range ±10%), and random horizontal and vertical flipping. Five augmented samples were generated for each original sample, ultimately resulting in 10,000 training samples and 2,000 validation samples. Stratified sampling was used to divide the training and validation sets to ensure consistent proportions of samples from different humidity conditions and corrugated cardboard types in the training and validation sets.

[0060] Specifically, the core technology of this invention is based on the inherent relationship between the thermophysical properties of materials and humidity. It uses non-contact thermal imaging technology combined with artificial intelligence analysis to achieve real-time online monitoring of corrugated cardboard humidity. Its working principle can be explained from the following aspects:

[0061] First, let's discuss the principles of thermal pulse excitation and thermal imaging. This method uses a high-power infrared lamp array to apply precisely controlled heat pulses (5000 watts / m², lasting 0.5 seconds) to the surface of the corrugated cardboard, causing a brief and uniform increase in surface temperature. Because moisture significantly affects the thermophysical properties of a material, areas with different moisture contents will exhibit distinct temperature responses and heat diffusion behaviors after receiving the same heat pulse. A high-precision thermal infrared imaging device continuously captures these temperature changes, generating time-series image data reflecting the thermal response characteristics, providing essential information for subsequent analysis.

[0062] Secondly, the principle of heat-humidity relationship analysis. Based on the non-steady-state heat conduction equation, this invention analyzes the heat conduction behavior in corrugated cardboard after a heat pulse. According to the principles of thermophysics, the moisture content of a material directly affects its thermal diffusivity and heat capacity. Areas with high moisture content exhibit high heat accumulation and slow heat dissipation. By accurately calculating the thermal diffusivity and thermal attenuation characteristics of each pixel, the system can establish a physical correlation between temperature distribution and moisture content, providing a theoretical basis for humidity measurement.

[0063] Finally, the principle of intelligent recognition and mapping. The thermal and humidity dual-parameter neural network model designed in this paper utilizes a five-layer convolutional neural network structure that matches the corrugated structure of corrugated cardboard and effectively captures the humidity distribution characteristics of different regions. Through large-scale training with 10,000 sets of samples, the model establishes a complex nonlinear mapping relationship between thermal response characteristics and actual humidity. This enables the system to accurately extract humidity information from the temperature distribution matrix and generate intuitive humidity characteristic maps.

[0064] Finally, the closed-loop control principle applies. The system correlates identified humidity deviations with production process parameters, forming a comprehensive feedback control mechanism. When humidity anomalies are detected, the system automatically adjusts key process parameters such as drying temperature and speed according to pre-set rules, implementing a closed-loop control process of detection-analysis-adjustment-redetection. The system also performs time series analysis on humidity data to predict future trends and provide a basis for decision-making in process parameter optimization.

[0065] 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 run in real time on the production line, which in principle solves the fundamental problem that traditional sampling detection methods cannot achieve real-time online monitoring.

[0066] 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.

[0067] The specific implementation method of step S01 is to select an appropriate location on the corrugated cardboard production line to fix the thermal infrared imaging device. This location must ensure unobstructed observation of the entire width of the corrugated cardboard. The thermal infrared imaging device uses an uncooled microbolometer focal plane array detector with an operating wavelength range of 7.5 to 14 μm, a temperature resolution of not less than 0.05°C, and a field of view set to 60°×45°. After installation is completed, the focus is precisely adjusted. During the adjustment process, a calibration plate with equidistant reference points is temporarily placed on the corrugated cardboard production line. The calibration plate is marked with equidistant reference points. The thermal infrared imaging device's electric focusing mechanism is used to perform remote fine adjustments until the image is clear and covers the entire corrugated cardboard width of 107×78 cm. The purpose of this step is to establish a stable and reliable thermal imaging observation foundation to ensure the integrity and accuracy of subsequent thermal response data acquisition.

[0068] The specific implementation of step S02 involves activating a heat pulse excitation system to apply a uniform heat pulse to the surface of the corrugated cardboard. This heat pulse excitation system consists of a high-power infrared lamp array consisting of 25 individual 250-watt infrared lamps arranged in a 5×5 matrix, with a coverage area that matches the width of the corrugated cardboard. Before the heat pulse is generated, a power controller precisely regulates the input current to the rated value, ensuring that the heat pulse intensity reaches 5000 watts per square meter. After the heat pulse begins, a precision timer controls the heat pulse duration to 0.5 seconds with a timing accuracy of ±0.01 seconds. During the heat pulse, a uniformity monitoring system monitors the spatial distribution of the heat pulse in real time, ensuring that the heat pulse intensity does not vary by more than ±3% across the entire area. This step creates standardized thermal excitation conditions on the corrugated cardboard surface, providing controllable and repeatable energy input for subsequent thermal response-based moisture analysis.

[0069] The specific implementation of step S03 is to activate the thermal infrared imaging device and immediately begin capturing thermal response images of the corrugated cardboard surface after the heat 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 thermal infrared imaging device sampling frequency is set to 25 Hz, meaning that one thermal image frame is captured every 40 milliseconds for 25 seconds, for a total of 625 frames. During the acquisition process, the thermal infrared imaging device's temperature drift compensation system operates continuously, using a built-in blackbody reference source for real-time calibration to ensure temperature measurement accuracy within ±0.5°C. After image acquisition is completed, the thermal response image sequence is transmitted to the image processing unit via a high-speed data transmission interface for storage and preprocessing. The purpose of this step is to obtain the temporal response sequence of the corrugated cardboard to the heat pulse stimulus and capture the temperature change characteristics caused by differences in thermal conductivity characteristics in different humidity areas.

[0070] The specific implementation of step S04 is to process the thermal response image based on the non-steady-state heat conduction equation in the image processing unit. First, the original thermal image sequence is spatially filtered, and a 5×5 Gaussian filter is used to eliminate random noise. The filter kernel function can be expressed as ,in is the Gaussian standard deviation, set to 1.2 pixels; and is the coordinate of the pixel in the filter kernel. Then calculate the temperature change rate of each pixel between consecutive frames , construct the temperature change rate field. For each pixel , extract the initial temperature before the heat pulse , peak temperature after heat pulse , the relationship curve between temperature and time during the cooling process . Substituting these parameters into the unsteady heat conduction equation: ,in is the thermal diffusivity, is the Laplace operator of temperature, which can be expressed as In a discrete pixel environment, the finite difference method is used for approximate calculation. , ,in and is the pixel spacing. By numerically solving this equation, the thermal diffusivity of each pixel is obtained and thermal attenuation characteristics ,in The exponential coefficient representing the temperature decay over time satisfies 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 of the point , thermal diffusivity and thermal attenuation characteristic parameters 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.

[0071] The specific implementation of step S05 is to input the temperature distribution matrix obtained in step S04 into the thermal and humidity dual-parameter neural network model. The 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 convolution kernels with a step size of 2 and an activation function of ReLU, which is mathematically expressed as follows: , extracting the basic texture features of the temperature distribution matrix. The second convolution layer uses 64 3×3 convolution kernels with a step size of 1 and an activation function of ReLU, focusing on extracting the spatial distribution features of thermal diffusivity. The third convolution layer uses 128 3×3 convolution kernels with a step size of 1 and an activation function of ReLU, which is used to capture the time series features of thermal attenuation characteristics. The convolution operation can be expressed as ,in is the output feature map, is the convolution kernel, is the input feature map, The first fully connected layer contains 512 neurons, and the activation function is ReLU. The features extracted by the first three layers are combined 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 output humidity feature map resolution, and the activation function is Sigmoid, expressed as The output value ranges from 0 to 1, representing the normalized humidity level. The network is trained using the Adam optimizer, with the learning rate initially set to 0.001 and dynamically adjusted using the cosine annealing strategy. The learning rate change expression is: ,in For the The learning rate of the step, and are the minimum and maximum learning rates, respectively, is the total number of steps. The mean square error is used as the loss function during training. ,in is the true humidity value, To predict humidity values, is the number of samples. After the neural network processing is complete, a humidity characteristic map is generated, using pseudo-color to display different humidity areas. 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, enabling accurate quantitative analysis of corrugated cardboard humidity.

[0072] The specific implementation of step S06 is to segment and analyze the humidity characteristic map generated in step S05. First, a multi-threshold segmentation algorithm is used to set the humidity threshold points to 8%, 12% and 15%, and the humidity characteristic map is segmented into low humidity area, standard humidity area, slightly high humidity area and excessively high humidity area. Then, a connected region labeling algorithm is used to identify the spatial distribution of each humidity area and calculate the number of pixels in each area. ,in Indicates different humidity area categories. For abnormal humidity areas (including low humidity areas and high humidity areas), calculate the percentage of their total area to the entire corrugated cardboard area, that is, the percentage of the abnormal humidity area area. ,in is the number of pixels in the low humidity area, is the number of pixels in the area with excessively high humidity, is the total number of pixels in the corrugated cardboard. At the same time, calculate the actual average humidity With standard humidity The difference between the two is the humidity deviation value. The calculation formula of humidity deviation value is: In addition, the uniformity of humidity distribution was evaluated by spatial autocorrelation analysis and Moran's I was calculated. ,in is the number of pixels, is the spatial weight matrix element, Pixels The humidity value at is the average humidity value. A value close to 1 indicates a highly clustered humidity distribution, close to 0 indicates a random distribution, and close to -1 indicates a uniform distribution. The purpose of this step is to quantify the degree of anomaly and spatial characteristics of the moisture distribution of corrugated cardboard, providing a basis for decision-making in production process control.

[0073] The specific implementation of step S07 is to calculate the humidity deviation value obtained in step S06 With preset threshold The preset threshold is set based on the quality requirements of corrugated cardboard products. The preset threshold for standard quality grade is 3%, the preset threshold for high quality grade is 2%, and the preset threshold for ultra-high quality grade is 1.5%. The system automatically selects the corresponding preset threshold based on the current product grade. The comparison process uses a fuzzy logic control algorithm, which not only considers the absolute value of the humidity deviation value, but also considers its changing trend. , that is, the difference between the humidity deviation value at the current moment and the previous moment. Fuzzy logic rules define the abnormal state judgment conditions, such as "If Big and If the value is large, the abnormality is high. Big but If the value is small, the abnormality is medium. The fuzzy membership function adopts the trapezoidal function ,in is the parameter of the membership function. When the humidity deviation value exceeds the preset threshold and exceeds the standard for three consecutive tests, the system determines that it is in an abnormal state and triggers the production line abnormality alarm system. The alarm system has three levels: when the humidity deviation value exceeds the preset threshold but is less than 1.5 times the preset threshold, a level one 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 level two alarm is triggered; when the humidity deviation value exceeds 2 times the preset threshold, a level three alarm is triggered, and the production line speed is automatically reduced by 30% to ensure product quality. The purpose of this step is to promptly detect humidity anomalies in the production process, prevent the continued production of substandard products, and reduce the scrap rate.

[0074] 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 in the last 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 trend terms. , Seasonal items and the residual ,satisfy The trend term reflects the long-term trend, the seasonal term reflects the cyclical change, and the residual term is the random fluctuation. Then the autoregressive integrated moving average (ARIMA) model is used to model the decomposed time series. The model can be expressed as ,in is the lag operator, is the autoregressive parameter, is the sliding average parameter, is the difference order, is a constant term, is a white noise process. The model parameters are automatically optimized and selected using the Bayesian Information Criterion (BIC), which is defined as ,in is the residual sum of squares, is the sample size, is the number of model parameters. Based on the established ARIMA model, the humidity change trend in the next 2 hours is predicted. , with a prediction accuracy within ±0.8%. Simultaneously, the system constructs a humidity historical trend chart, including an actual humidity deviation curve, a moving average curve (with a 60-minute window), and a predicted humidity deviation curve, visually displaying humidity fluctuations. Furthermore, the system uses historical data to identify which production process parameter changes are highly correlated with humidity fluctuations. Multiple regression analysis and Granger causality tests identify key influencing factors, providing data support for optimizing production parameters. The purpose of this step is to achieve predictable control of humidity during corrugated cardboard production through historical data mining and predictive analysis, allowing for preemptive adjustment of production parameters to prevent humidity anomalies.

[0075] The specific implementation of step S09 is to establish a closed-loop control system between the humidity deviation value and the production process parameters. First, a multi-input multi-output (MIMO) control model is constructed, and the input variables include the current humidity deviation value. , humidity change rate , predicted humidity deviation value , the output variables include the drying temperature adjustment and drying speed adjustment The control model adopts the model predictive control (MPC) algorithm, and the system state equation can be expressed as , the output equation is ,in is the state vector, which contains the humidity deviation value and its historical value, is the control vector, which includes the adjustment of drying temperature and drying speed. is the output vector, i.e. the predicted future humidity deviation value, is the system matrix, which is obtained through system identification. The objective function of model predictive control is ,in For the prediction time domain, To control the time domain, is the reference trajectory, and is the weight matrix. During the control process, the drying temperature is adjusted between 65°C and 95°C, with an adjustment step of 0.5°C; the drying speed is adjusted between 80 and 120 m / min, with an adjustment step of 2 m / min. When high humidity is detected, the system automatically increases the drying temperature or decreases the drying speed; when low humidity is detected, the system automatically decreases the drying temperature or increases the drying speed. The controller's response time 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, which continuously updates 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 of no more than 0.5%, an overshoot of no more than 1.5%, and an adjustment time of no more than 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 product quality stability.

[0076] The detailed structure of the thermal and humidity dual-parameter neural network model is as follows: the input layer receives the dimension Data tensor , containing three channels: temperature, thermal diffusivity, and thermal attenuation characteristics. The first convolution layer uses 32 5×5 convolution kernels with a stride of 2 and a padding of 2. The convolution operation can be expressed as ,in is the convolution kernel weight tensor, is the bias vector, is the ReLU activation function, Represents the convolution operation. The output size of the first convolution layer is , followed by a batch normalization layer, the expression is ,in and Respectively The mean and variance of the channel, and is a learnable parameter, is a small constant to prevent division by zero. Batch normalization is followed by a maximum pooling layer with a pooling kernel size of 2×2 and a stride of 2. The expression is , after pooling the size is The second convolution layer uses 64 3×3 convolution kernels with a stride of 1 and a padding of 1, which is expressed as , the output size is , followed by a batch normalization layer and a maximum pooling layer, the size after pooling is The third convolution layer uses 128 3×3 convolution kernels with a stride of 1 and a padding of 1. The expression is , the output size is , followed by a batch normalization layer and a maximum pooling layer, the size after pooling is After flattening the feature map, we get the vector , the dimension is , and the production process parameters (4-dimensional vector ) splicing to get the vector , input the first fully connected layer, the expression is ,in is the weight matrix, is the bias vector. This layer contains 512 neurons. Dropout is used to prevent overfitting. The dropout probability is 0.3. The expression is ,in The probability of Bernoulli distribution is A binary random variable, Represents element-by-element multiplication. The output size of the second fully connected layer is , the expression is ,in is the Sigmoid activation function, is the weight matrix, is the bias vector. The output value ranges from 0 to 1, representing the normalized humidity level, forming a humidity feature map.

[0077] The convolution kernel size of the heat and humidity dual-parameter neural network model uses a preset convolution kernel correlation function to match the corrugated structure size of the corrugated cardboard. The correlation function calculates the convolution kernel size as follows: ,in is the convolution kernel size, is the characteristic dimension of the corrugated structure of the corrugated cardboard, in millimeters. is the number of pixels per millimeter, Indicates rounding down. For standard A-type corrugated structure, It is about 8.0 to 9.0 mm, and the corresponding convolution kernel size is 5×5; for the standard B-type corrugated structure, It is about 6.0 to 6.5 mm, and the corresponding convolution kernel size is 3×3; for the standard C-type corrugated structure, It is about 7.0 to 7.5 mm, and the corresponding convolution kernel size is 3×3 or 5×5; for E-type and F-type corrugated structures, Smaller, the corresponding convolution kernel size is 3×3.

[0078] The training dataset for the heat and humidity dual-parameter neural network model was established as follows: First, samples were collected from the actual production line. Thermal response images were acquired under different humidity conditions (humidity range 5% to 20%, interval 0.5%). Fifty sets of samples were collected for each humidity condition, for a total of 1,500 sets of original samples. For each sample, the actual humidity value was simultaneously measured using a high-precision contact humidity sensor (accuracy ±0.2%). As label data. At the same time, record the production conditions of the sample, including the thickness of the corrugated cardboard (Range 1.5 ~ 7 mm, 5 levels), corrugated cardboard corrugation type (5 types in total: A, B, C, E, and F), raw material composition of corrugated cardboard (8 different ratios in total), forming a feature vector , with a total dimension of 23. Data augmentation techniques were 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%), and random horizontal and vertical flipping. Five augmented samples were generated for each original sample, ultimately resulting in 10,000 training samples and 2,000 validation samples. Stratified sampling was used to divide the training and validation sets to ensure consistent proportions of samples from different humidity conditions and corrugated cardboard types in the training and validation sets.

[0079] In this embodiment, the convolution kernel size of the thermal and humidity dual-parameter neural network model is designed to match the corrugated cardboard corrugation structure size, ensuring that the neural network can accurately capture the characteristic information of different types of corrugated structures. This matching relationship ensures that the receptive field of the convolution layer corresponds to the actual physical size of the corrugations, improving the model's ability to identify local moisture distribution characteristics. By selecting the appropriate convolution kernel size for different specifications of corrugated cardboard (Type A, Type B, Type C, etc.), the system can adapt to a variety of product specifications, reducing model generalization error, while improving the accuracy and adaptability of humidity detection, providing technical support for achieving high-precision humidity monitoring.

[0080] Type A, Type B, and Type C refer to different structural types of corrugated cardboard. These classifications are industry standards based on the shape, size, and characteristics of the corrugated waves.

[0081] Type A corrugated paper (also known as large corrugated paper) has a larger corrugation height and period, with a characteristic corrugation structure size of approximately 8.0 to 9.0 mm. This structure provides high compressive strength and cushioning properties and is typically used for packaging products requiring strong protection.

[0082] Type B fluting (also known as medium fluting) has a smaller corrugation structure than type A, with a characteristic dimension of approximately 6.0 to 6.5 mm. It offers excellent compression resistance and a good print surface, balancing strength and print quality, and is widely used in general product packaging.

[0083] Type C fluting is between Types A and B, with a characteristic corrugation dimension of approximately 7.0 to 7.5 mm. This structure combines good compressive strength and stacking performance, making it one of the most widely used types of fluting in the packaging industry, suitable for packaging a variety of medium-weight products.

[0084] The structural characteristics of different corrugated types will affect the heat conduction behavior. Therefore, when designing a humidity monitoring system, it is necessary to select the appropriate size of the convolution kernel for different types to match their physical characteristics.

[0085] To better understand and implement the present invention, Example 2, a specific application scenario, is provided below: A thermal infrared imaging-based online humidity monitoring system was implemented on a corrugated cardboard production line. This line primarily produces B-type corrugated cardboard used in high-end electronic product packaging, with strict requirements for uniform moisture content. The standard humidity is 9.5%, with an allowable deviation of no more than 1.5%. The system was installed at the exit of the drying section of the production line, and the monitoring area covered the entire width of the cardboard, 107 x 78 cm.

[0086] The thermal infrared imaging device uses a FLIR A655sc uncooled microbolometer focal plane array detector with an operating wavelength range of 7.5 to 14 μm, a temperature resolution of 0.03°C, and a field of view of 62° × 45°. The device is installed at a height of 95 cm, with an imaging resolution of 640 × 480 pixels, each pixel corresponding to an area of ​​approximately 1.67 × 1.63 mm on the cardboard surface, and a sampling frequency of 25 Hz.

[0087] The thermal pulse excitation system consists of 25 infrared lamps, each with a power of 250 watts, arranged in a 5×5 matrix. Pulse intensity is maintained at 5000 watts per square meter, and duration is precisely controlled to 0.5 seconds. The system monitors the spatial distribution of thermal pulses in real time, limiting deviations within the entire area to within ±2.5%, ensuring consistent thermal excitation.

[0088] The researchers collected and analyzed multiple sets of thermal response data under different humidity conditions. As shown in Table 1, the thermal parameters under four typical humidity conditions showed significant differences.

[0089] Table 1 Thermal parameter characteristics under different humidity conditions

[0090]

[0091] The data in Table 1 show that as humidity increases, the peak temperature and thermal diffusivity decrease significantly, the thermal attenuation parameter decreases, and the thermal response duration increases. This provides an important training basis for the heat-humidity dual-parameter neural network model. Figure 2The thermal response curves of corrugated cardboard to thermal pulse excitation under different humidity conditions are shown. The figure contains four curves, representing the temperature variation over time under low humidity (6.0%-7.5%), standard humidity (8.0%-11.0%), slightly elevated humidity (11.5%-14.0%), and excessive humidity (14.5%-17.0%) conditions. The figure shows that low-humidity corrugated cardboard has the highest peak temperature (approximately 46.5°C), the fastest thermal decay rate (β=0.30s^-1), and the shortest thermal response duration. As humidity increases, the peak temperature gradually decreases, the thermal decay rate slows, and the thermal response duration increases. The figure also indicates the thermal pulse excitation phase (0.5 seconds) and the thermal response duration threshold (baseline temperature + 1°C).

[0092] The thermal and humidity dual-parameter neural network model was trained using 10,000 sets of samples, including diverse data from different humidity conditions and corrugated cardboard types. The model validation performance is shown in Table 2.

[0093] Table 2 Verification performance of thermal and humidity dual-parameter neural network model

[0094]

[0095] The data in Table 2 show that the model has high prediction accuracy for all 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, which fully meets the needs of online monitoring. Figure 3 This section details the training process and network structure of a dual-parameter neural network model for heat and humidity. The main figure shows the curves of training loss and validation loss over training epochs. It can be seen that both loss values ​​decrease with increasing training epochs, reaching an early stopping point around 78 training epochs, at which point validation loss no longer decreases.

[0096] The staff performed regional segmentation analysis on the predicted humidity characteristic map, as shown in Table 3.

[0097] Table 3 Humidity zone analysis table during a production process

[0098]

[0099] Table 3 shows that the overall moisture content of this batch of corrugated cardboard was acceptable, but 14.64% of the paperboard contained abnormal areas, primarily concentrated at the edges and in localized areas. This triggered a Level 1 alarm and initiated closed-loop control, automatically adjusting drying parameters. The closed-loop control system automatically adjusted production parameters based on the moisture deviation, as shown in Table 4.

[0100] Table 4 Closed-loop control parameter adjustment table

[0101]

[0102] 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.

[0103] The staff conducted a statistical analysis of the system's long-term operation data. Table 5 shows the quality improvement effects before and after the implementation of the system.

[0104] Table 5 Comparison of quality indicators before and after system implementation

[0105]

[0106] The data in Table 5 show that after the implementation of the system, the humidity control accuracy of corrugated cardboard was significantly improved, the defective product rate decreased by 71.0%, customer complaints decreased by 78.1%, and the energy consumption of the production process decreased by 9.1%. Figure 4 The figure demonstrates the closed-loop control system's automatic adjustment of production parameters and its quality improvement effects. The main curve (red) shows the change in humidity deviation over time, which gradually decreases from an initial 1.8% to a stable level of around 0.6%, below the preset threshold of 1.5%. The figure also shows the drying temperature (blue) and drying speed (green) curves before and after the adjustment. It shows that the system first increases the drying temperature and decreases the drying speed to reduce humidity, then performs fine-tuning to maintain the humidity during the stabilization phase. The control process is divided into three stages: initial adjustment (0-20 minutes), fine adjustment (20-50 minutes), and stabilization (50-90 minutes). The lower left corner of the figure shows the quality improvement results before and after the system implementation, including a 63.7% reduction in average humidity deviation, a 68.1% reduction in humidity standard deviation, a 71.0% reduction in defective product rate, and a 9.1% reduction in energy consumption. The lower right corner shows the controller performance indicators, including a steady-state error of less than 0.5%, an overshoot of less than 1.5%, and a settling time of less than 10 minutes, demonstrating the excellent performance of the control system.

[0107] The time series analysis module creates a historical humidity trend chart and predicts future changes. As shown in Table 6, the prediction accuracy decreases with increasing forecast time, but the prediction error within 2 hours is controlled within ±0.8%, meeting the production control requirements.

[0108] Table 6 Humidity prediction accuracy evaluation table

[0109]

[0110] Traditional corrugated cardboard moisture detection primarily relies on contact humidity sensors or offline sampling. These methods suffer from limited measurement points, incomplete coverage, inability to perform real-time online adjustments, and significant disruption to the production process. Specifically, contact sensors are prone to wear and require frequent replacement; they can only measure humidity at a few fixed points, failing to capture the moisture distribution across the entire cardboard; offline sampling testing exhibits significant lag, making it difficult to identify and resolve issues promptly; and the significant differences in processing performance and finished product quality under varying humidity conditions make it difficult to establish effective quality prediction models using traditional methods.

[0111] The infrared thermal imaging online humidity monitoring method of the present invention overcomes the limitations of traditional methods and achieves non-contact, full-coverage, high-precision, real-time monitoring of the humidity of the entire corrugated cardboard surface. By combining thermal pulse excitation with thermal infrared imaging technology, the differences in thermal response between different humidity zones are captured; a precise mapping relationship between temperature characteristics and humidity is established using non-steady-state heat conduction equations and neural network models; humidity anomaly identification, automatic alarm, and closed-loop control functions are integrated to achieve intelligent adjustment of production parameters; and the introduction of a time series analysis and prediction module allows for the early prediction of humidity change trends, providing support for production decision-making. Through these technological innovations, the system significantly improves the quality stability and production efficiency of corrugated cardboard production, reduces energy consumption and the rate of defective products, and provides an effective solution for the intelligent upgrade of the corrugated cardboard manufacturing industry.

[0112] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 7 and 8 below.

[0113] Table 7 Variable Explanation Table (Part I)

[0114]

[0115] Table 8 Variable Explanation Table (Part 2)

[0116]

[0117] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for online automatic monitoring of corrugated cardboard humidity, characterized in that: include: Use a thermal infrared imaging device at a fixed position on the corrugated cardboard production line and adjust the focal length to cover the entire width of the corrugated cardboard; A heat pulse excitation system is used to apply uniform heat pulses to the surface of the corrugated cardboard; a thermal infrared imaging device is used to collect thermal response images of the corrugated cardboard surface; According to the thermal response image, the thermal diffusivity and thermal attenuation characteristics of each pixel of the corrugated cardboard are calculated based on the non-steady-state heat conduction equation to generate a temperature distribution matrix; the temperature distribution matrix is ​​analyzed based on the thermal and humidity dual-parameter neural network model, the humidity feature map is extracted, and the corresponding relationship between the temperature distribution matrix and the moisture content of the corrugated cardboard is established; the region is segmented through the humidity feature map, the humidity abnormal area is identified, and the area ratio of the humidity abnormal area and the humidity deviation value are calculated; the humidity deviation value is compared with the preset threshold, and an alarm is triggered if it exceeds the preset threshold; the steps of generating the temperature distribution matrix specifically include: first, spatial filtering the thermal response image, using a 5×5 Gaussian filter to eliminate random noise, and the filter standard deviation is set to 1.2 pixels; then, the temperature change rate of each pixel between consecutive frames is calculated to construct a temperature change rate field; for each pixel, the initial temperature before the heat pulse is extracted. , peak temperature after heat pulse , the relationship curve between temperature and time during the cooling process ; Substitute into the unsteady heat conduction equation: ,in is the thermal diffusivity, is the Laplace operator of temperature; by numerically solving this equation, the thermal diffusivity of each pixel is obtained and thermal attenuation characteristics ,in The exponential coefficient representing the temperature decay over time satisfies 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 attenuation characteristics of the point.

2. The method for online automatic monitoring of corrugated cardboard humidity according to claim 1, characterized in that: The thermal pulse excitation system specifically forms a uniform heat source by instantly releasing heat energy through a high-power infrared lamp array, which is used to stimulate the surface of the corrugated cardboard being tested to produce a thermal response image.

3. The method for online automatic monitoring of corrugated board humidity according to claim 2, characterized in that: The heat pulse intensity was maintained at 5000 watts per square meter and the heat pulse duration was 0.5 seconds.

4. The method for online automatic monitoring of corrugated board humidity according to claim 1, characterized in that: The resolution of the thermal response image collected by the thermal infrared imaging device is set to no 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 corrugated board humidity according to claim 1, characterized in that: Thermal attenuation characteristics refer to the curve of the temperature drop over time after the corrugated cardboard absorbs heat energy. By analyzing the differences in thermal attenuation characteristics in different areas, the distribution of moisture content inside the corrugated cardboard can be reflected.

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

7. The method for online automatic monitoring of corrugated board humidity according to claim 1, characterized in that: Humidity deviation refers to the difference between the actual measured humidity and the standard humidity.

8. The method for online automatic monitoring of corrugated board humidity according to claim 1, characterized in that: The unsteady-state heat conduction equation is used to describe the dynamic process of heat transfer in corrugated cardboard and calculate the thermal diffusivity and 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 coefficient of the corrugated cardboard material. The output is the thermal diffusivity and thermal attenuation characteristics of each point on the corrugated cardboard.

9. The method for online automatic monitoring of corrugated board humidity according to claim 1, characterized in that: The structure of the thermal and humidity dual-parameter neural network model is a five-layer convolutional neural network structure, which includes 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 thermal diffusivity, and the third convolutional layer is used to extract the time series features of 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.

Images