Online experimental device and modeling method for modeling relationship of multiple influence factors of moisture content of tobacco material

Abstract

This invention discloses an online experimental device and modeling method for modeling the relationship between multiple influencing factors of moisture content in tobacco materials, belonging to the field of tobacco processing technology and quality inspection technology. The device includes a sampling and sorting unit, an electric dispersion platform, a morphology detection unit, a fluffiness detection unit, an environmental detection unit, a ranging unit, and a control unit. The ranging unit is installed near an online infrared moisture meter to acquire the vertical distance H from the probe end face to the material surface in real time to correct measurement errors; the morphology detection unit simultaneously acquires the three-dimensional morphology and color difference data of the material under completely dark conditions; the fluffiness detection unit acquires the filling value; and the environmental detection unit acquires the ambient temperature and humidity. This invention achieves spatiotemporal synchronous acquisition of material morphology parameters, environmental parameters, and moisture content data, and establishes a multi-factor relationship model based on a neural network, effectively guiding the refined control of process quality in the production process.

Description

Technical Field

[0001] This invention belongs to the field of tobacco processing technology and quality testing technology, specifically relating to an online experimental device and modeling method for detecting the physical properties and environmental parameters of tobacco materials and establishing a model of the relationship between multiple influencing factors and moisture content. Background Technology

[0002] In the cigarette production process, the stability of the moisture content of tobacco materials directly affects the physical properties, sensory quality, and filling capacity of the finished cigarettes. Currently, the industry mainly relies on online infrared moisture meters or microwave moisture meters to detect the moisture content of tobacco materials on the production line.

[0003] However, in actual production, the measurement accuracy of infrared moisture meters and microwave moisture meters is often affected by various factors. On the one hand, the irregular shape of tobacco materials, which are randomly piled on the conveyor belt, results in an uneven surface. For infrared moisture meters, this height variation directly changes the measurement focal length and light intensity of the infrared probe, causing unexpected drift in the measurement reading. For microwave moisture meters, fluctuations in the thickness of the material layer and uneven packing density alter the attenuation and phase shift of microwaves in the material, leading to deviations in the moisture measurement value. Existing research shows a significant negative correlation between the conveying height of tobacco and the moisture meter reading, and changes in the packing density of tobacco also significantly affect the detection results of microwave moisture meters; the more uniform the height distribution of tobacco conveying, the more stable the moisture meter reading. On the other hand, the hygroscopic and desiccant properties of tobacco materials are affected by the coupling of ambient temperature and humidity, the microstructure of the tobacco material itself (such as thickness and bulkiness), color depth (mainly affecting infrared optical characteristics), and bulk density (affecting microwave dielectric response).

[0004] Current testing methods typically process this data in a fragmented manner: moisture content is measured online using infrared or microwave moisture meters, while the morphology, color difference, and filler content of tobacco materials are usually measured offline through manual sampling. This "spatial-temporal asynchrony" data acquisition method makes it difficult for researchers to establish precise mathematical models between various physical indicators and moisture content, and to quantify the specific contribution rates of environmental factors and material properties to moisture measurement deviations. Consequently, it is difficult to guide the refined control of process quality in production. For example, although there are improved calibration methods for infrared or microwave moisture meters in related technologies, most are offline calibrations or only address a single source of error, failing to achieve simultaneous online acquisition and comprehensive modeling of multiple influencing factors.

[0005] Therefore, developing an experimental device capable of acquiring accurate material data and environmental parameters online synchronously, establishing a high-precision multi-factor correlation model of moisture content, and using the modeling method implemented by this experimental device are of great significance for improving the moisture content control level in tobacco processing. Summary of the Invention

[0006] In view of the above, the purpose of this invention is to overcome the shortcomings of the prior art and provide an online experimental device and modeling method for modeling the relationship between multiple influencing factors of moisture content in tobacco materials. This device can synchronously acquire accurate material data and environmental parameters online, thereby establishing a high-precision multi-factor correlation model for moisture content, obtaining accurate moisture content results, and can be used for the correction and prediction of moisture content measurement results.

[0007] The technical solution adopted in this invention is as follows:

[0008] In a first aspect, the present invention provides an online experimental device for modeling the relationship between multiple influencing factors of moisture content in tobacco materials, wherein the device includes: a sampling and sorting unit, an electric dispersion platform, a morphology detection unit, a fluffy state detection unit, an environmental detection unit, a distance measuring unit, and a control unit.

[0009] The sampling and sorting unit has multi-dimensional motion and quantitative grasping functions, enabling it to grasp tobacco materials from the conveyor belt at fixed points and in fixed quantities and place them on an electric dispersing platform. This unit can employ a robotic arm or similar grasping mechanism, possessing three-dimensional spatial motion capabilities and precise grasping control.

[0010] The electric dispersion platform disperses the tobacco material through the vibration of the screen platform, separating the tobacco material into a single layer and allowing it to fall into the support tray below via the screen. The key to this design is that the vibration dispersion enables the originally piled tobacco material to be distributed in a single layer, providing an ideal material arrangement for the subsequent three-dimensional scanning in the morphology detection unit, thereby ensuring that the three-dimensional laser profilometer can collect complete contour data of each individual tobacco material.

[0011] The morphology detection unit includes a light-shielding enclosure and a rotating conveyor mechanism. The light-shielding enclosure houses a fixed light source, a 3D laser profilometer, and a colorimeter. The light-shielding enclosure eliminates interference from ambient light, ensuring the acquisition of 3D morphology and color difference data of the tobacco material under constant illumination. The fixed light source provides standardized lighting conditions, eliminating the influence of light intensity variations on color difference measurement. The 3D laser profilometer has a dimensional measurement error better than ±0.02mm, and the colorimeter has a measurement error ΔE better than ±0.05. The rotating conveyor mechanism feeds the tobacco material into the enclosure for detection and then tilts the detected tobacco material back onto the conveyor belt via a sliding track.

[0012] The fluffiness detection unit is used to measure the filling value of tobacco material and has an automatic material return function. The filling value is a key physical indicator reflecting the fluffiness of tobacco material, which directly affects the filling capacity of tobacco material during the rolling process and the sensory quality of the final product.

[0013] The environmental monitoring unit can be independently installed on the experimental device's measurement platform to acquire ambient temperature and humidity data. This unit includes temperature and humidity sensors, positioned no more than 1 meter from the tobacco material being tested. The temperature measurement error is better than ±0.5℃, and the humidity measurement error is better than ±3%RH. Its independent installation aims to eliminate interference from the heat generated by the device's operation, ensuring the accuracy of the environmental parameters.

[0014] The ranging unit employs a laser ranging probe installed near the online infrared moisture meter on the production line. This probe is used to acquire the vertical distance H from the probe end face to the surface of the tobacco material layer in real time, thus helping to correct the measurement error of the infrared moisture meter. The measurement error of the laser ranging probe is better than ±0.02 mm.

[0015] The control unit includes a mechanical housing, a communication module, a central processing unit, a display screen, a power supply module, and operation buttons. It has functions such as data acquisition and communication, motor control, modeling of multiple influencing factors of moisture content, and result display. As the nerve center of the entire device, the control unit coordinates the working sequence of each detection unit, aggregates multi-source detection data, and performs data fusion and modeling calculations.

[0016] Secondly, the present invention provides a modeling method for a multi-factor relationship model of tobacco material moisture content using the above-mentioned online experimental device, comprising the following steps:

[0017] Step S1, Infrared Moisture Analyzer Height Deviation Correction: The control unit acquires the original moisture content measurement results from the online infrared moisture analyzer and the vertical distance H of the tobacco material surface measured by the laser ranging probe in the ranging unit. The control unit then corrects the real-time readings of the infrared moisture analyzer for height deviation. Specifically, based on the deviation ΔH between the vertical distance H of the material surface measured by the laser ranging probe and the preset calibration distance, the control unit calls a pre-calibrated height correction function to compensate for the original measurement value of the infrared moisture analyzer, obtaining the height-corrected moisture content data.

[0018] Step S2, Environmental Parameter Acquisition: The control unit acquires the environmental temperature and humidity data collected by the environmental detection unit, including the environmental temperature t and the environmental humidity RH, as one of the rapid change compensation factors in the model.

[0019] Step S3, Automatic Sampling: The control unit sends a sampling command to drive the sampling and sorting unit to grab tobacco material samples from the conveyor belt. The sampling and sorting unit performs quantitative grabbing according to the preset sampling cycle and sampling quantity to ensure that the amount of material sampled each time meets the measurement requirements of each detection unit.

[0020] Step S4: Spatiotemporal Alignment of Moisture Content Data: The control unit performs timestamp synchronization alignment. Using the built-in timestamp synchronization module, the average corrected moisture content within a preset time window before and after the sampling time is extracted as the standard moisture content measurement result data W for the tobacco material. The length of the time window can be dynamically set according to factors such as conveyor belt speed and material residence time in the detection process, typically 3-10 seconds before and after the sampling time. The key to this step is: through timestamp synchronization technology, the portion of tobacco material grabbed from the conveyor belt is correlated with the time period during which it is measured by the infrared moisture meter on the conveyor belt, ensuring precise temporal alignment between the material physical characteristic data obtained from offline detection and the moisture content data measured online.

[0021] Step S5, Material Characteristic Parameter Extraction and Automatic Recovery: The sampled tobacco material is divided into two parts: one part is sent to the morphology detection unit for detection via an electric dispersion platform and a rotary conveyor mechanism to obtain the average length D of the sample. length Average width D width Average thickness D thick and color difference data C color Simultaneously, the control unit records the light intensity L of the fixed light source; another portion is placed into the fluffiness detection unit for detection to obtain the filling value V of the tobacco material. fill Specifically, the rotating conveyor's carrier plate slides along a circumferential sliding track driven by a rotating component, sequentially completing the following actions: material reception (positioned below the electric dispersing platform), delivery into a light-proof box for testing, and material dumping and recycling (sliding above the conveyor belt and flipping over). After all tobacco materials have been tested, they are automatically recycled back to the conveyor belt, achieving a non-destructive cyclic experiment.

[0022] Step S6, Data Recording: The control unit records all detection data, including ambient temperature t, ambient humidity RH, vertical distance from the material surface H, illumination intensity of the fixed light source L, and average length D of the tobacco material. length Average width D width Average thickness D thick and color difference data C color and fill value V fill And the standard moisture content W, recorded as a complete set of data records.

[0023] Step S7, Repeat Sampling: Repeat steps S1 to S6, recording N sets of test data. Changes in test data caused by active or passive conditions under the same tobacco material are considered valid samples. The value of N is set according to the modeling accuracy requirements, and is usually no less than 500 sets.

[0024] Step S8: Establishing a multi-factor relationship model: The N sets of detection data are processed according to influencing factors. Ambient temperature (t), humidity (RH), vertical distance from the material surface (H), and illumination intensity of a fixed light source (L) are used as rapidly changing compensation factors. The average length D of the tobacco material is used as the compensation factor. length Average width D width Average thickness D thick and color difference data C color and fill value V fill For the characteristic factors of slowly changing states, the following relationship model is established:

[0025] W=f(t,RH,H,L,D length D width D thick C color V fill )

[0026] Where W is the standard moisture content of the tobacco material, and f is the nonlinear mapping function to be established.

[0027] In at least one of the possible implementations, the model building process includes the following sub-steps:

[0028] (1) Data preprocessing and segmentation: Preprocess the N sets of tobacco material detection data obtained in steps S1 to S7, including but not limited to input features such as ambient temperature t, humidity RH, vertical distance H from the material surface, illumination intensity L of a fixed light source, and average length D of the tobacco material. length Average width D width Average thickness D thick and color difference data C color and fill value V fill Normalize or standardize the dataset, and then divide the preprocessed dataset into training, validation and test sets according to a preset ratio, such as 60%:20%:20%.

[0029] (2) Neural network model construction: Construct a feedforward neural network model, which has: an input layer with the same number of neurons as the number of preprocessed input features; one or more hidden layers, which use nonlinear activation functions; an output layer containing a neuron using a linear activation function to output the continuous standard moisture content W of tobacco materials; the model selects mean square error or mean absolute error as the loss function and uses optimizers such as Adam and SGD for model training; optionally, Dropout or L1 / L2 regularization techniques are introduced during training to prevent overfitting.

[0030] (3) Model training and evaluation: The constructed neural network model is iteratively trained using the training set, and the loss function is minimized by the optimizer. During the training process, the hyperparameters of the neural network are adjusted according to the performance of the model on the validation set, including but not limited to the number of hidden layers, the number of neurons, and the learning rate. An early stopping strategy can be adopted to prevent overfitting. After training, the final model is evaluated using the test set. The evaluation metrics include but are not limited to mean squared error, root mean square error, mean absolute error, and R-squared.

[0031] (4) Model solidification: Solidify the parameters of the neural network model that has obtained the best performance after training and evaluation to form the multi-factor relationship model.

[0032] Compared with existing technologies, the main design concept of this invention lies in the integration of four functional modules into the online experimental device: infrared moisture meter height deviation correction, morphology detection, fluffiness detection, and environmental monitoring. Through timestamp synchronization technology in the control unit, it achieves precise spatiotemporal alignment of tobacco material morphological parameters (length, width, thickness, color difference, filling value), environmental temperature and humidity, and moisture content data. Furthermore, it categorizes influencing factors into rapidly changing compensation factors combined with slowly changing state characteristic factors, and employs a feedforward neural network to establish a nonlinear mapping model from multidimensional input features to standard moisture content. This model can accurately quantify the specific contribution rate of each influencing factor to the moisture content measurement deviation, providing a scientific basis for the refined control of production processes. Compared with traditional fragmented detection methods such as online moisture content measurement and offline physical parameter measurement, this invention significantly improves the correlation between multi-source data, laying a data foundation for high-precision modeling.

[0033] Furthermore, the ranging unit of this invention employs a high-precision laser ranging probe to acquire the vertical distance H from the probe end face of the infrared moisture meter to the surface of the tobacco material layer in real time. The control unit dynamically compensates for the original reading of the infrared moisture meter based on the deviation ΔH between H and the calibrated distance, effectively solving the problems of measurement focal length variation and measurement reading drift caused by the random accumulation of tobacco material on the conveyor belt and the unevenness of the material layer surface. This dynamic correction mechanism is not yet realized in existing infrared moisture meter calibration methods, and has significant innovation and practicality.

[0034] Furthermore, through the coordinated action of the rotating components and the circumferential sliding track in the rotary conveyor mechanism, the entire process of tobacco material processing—from sampling, dispersion, and morphology detection to dumping and recycling—was automated. The measuring cylinder of the fluffy state detection unit also has a flipping function for material recycling. Throughout the experiment, all tobacco material was automatically returned to the conveyor belt after detection, achieving a non-destructive recycling experiment. This avoided material waste and ensured the continuous operation of the production line.

[0035] Furthermore, the functional modules of this invention have a high degree of integration. The shape detection unit adopts a fully light-proof design to ensure the consistency of the measurement environment. The circumferential sliding track of the rotary transmission mechanism has a reasonable layout and occupies little space, making it easy to add and modify on existing production lines, and has good industrial applicability. Attached Figure Description

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings, wherein:

[0037] Figure 1 This is a three-dimensional schematic diagram of the overall structure of the online experimental device provided in the embodiment of the present invention;

[0038] Figure 2 This is a cross-sectional view of the internal structure of the morphology detection unit provided in an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of a fluffy state detection unit provided in an embodiment of the present invention;

[0040] Figure 4 A flowchart of the modeling method provided in an embodiment of the present invention.

[0041] Explanation of reference numerals in the attached figures:

[0042] 1-Sampling and sorting unit; 2-Electric dispersing platform; 3-Morphological detection unit; 4-Fluffy state detection unit; 5-Environmental detection unit; 6-Distance measuring unit; 7-Control unit; 8-Rotary transmission mechanism; 9-Infrared moisture meter; 10-Fixed light source; 11-3D laser profilometer; 12-Colorimeter; 13-Rotating component; 14-Circumferential sliding track; 15-Carrying plate; 16-Measuring cylinder; 17-Pressure probe; 18-Mass sensor. Detailed Implementation

[0043] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0044] Before proceeding with the embodiments of the present invention, the detection principle and data source proposed in this invention will be introduced first:

[0045] This invention is based on an in-depth analysis of the sources of measurement error in infrared moisture meters. It divides the factors affecting the measurement accuracy of infrared moisture meters into two categories for targeted detection and data acquisition, in order to serve the subsequent establishment of neural network models.

[0046] The first category consists of the inherent state parameters of the tobacco material itself, namely slowly varying / fixed factors. These factors directly affect the reflection and absorption characteristics of infrared light on the surface of the tobacco material, and are fundamental characteristics for moisture content measurement. They mainly include:

[0047] (1) Thickness and dimensions of tobacco material: These affect the penetration depth of infrared light and the signal acquisition area. The thickness of the tobacco material accumulation on the conveyor belt directly affects the working focal length of the infrared probe, while the size of the material particles affects the effective acquisition area of ​​the reflected signal. In the embodiments of the present invention below, a three-dimensional laser profilometer 11 is used for non-contact measurement, and the dimensional measurement error is required to be better than ±0.02mm.

[0048] (2) Color of tobacco materials: Different colored tobacco materials have different absorption rates for infrared light of a specific wavelength, which is one of the important sources of measurement error in infrared moisture analyzers. In the embodiments of the present invention below, a colorimeter 12 is used for quantitative analysis, and the measurement error ΔE is required to be better than ±0.05.

[0049] (3) Loose state (fill value): This affects the bulk density of tobacco materials and the reflective surface characteristics of light signals. Tobacco materials with higher fill values ​​exhibit a looser bulk state on the conveyor belt, resulting in a more complex reflection path for infrared light. In the embodiments of the present invention below, the fill value characteristics are obtained by measuring the material cylinder 16 in combination with the mass sensor 18 and the pressure probe 17.

[0050] The second category is environmental parameters at the production site, namely rapidly changing factors. These factors are the main cause of drift in infrared moisture meter measurement results, including:

[0051] (1) Ambient temperature t and ambient humidity RH: These affect the stability of the internal components of the moisture meter and the absorption characteristics of infrared light in the air. In the embodiments of the present invention below, an environmental detection unit 5 is used for real-time monitoring. The temperature measurement error is required to be better than ±0.5℃, and the humidity measurement error is required to be better than ±3%RH.

[0052] (2) Measurement height H: This is the vertical distance from the probe end face of the infrared moisture meter to the surface of the tobacco material layer. When the tobacco material is randomly piled up on the conveyor belt, causing the layer height to change, the measurement focal length changes, resulting in reading drift. In the embodiments of the present invention below, a high-precision laser ranging probe in the ranging unit 6 is used to accurately measure the layer height, and the measurement error is required to be better than ±0.02mm.

[0053] (3) Illumination intensity L: Changes in ambient light intensity can interfere with the light signal acquisition of the infrared moisture analyzer. In the morphology detection unit 3 of the present invention described below, a fully light-shielding design is adopted, and a fixed light source 10 is set inside to provide standardized illumination. The illumination intensity L can be monitored or set by the control unit 7 to ensure the consistency of illumination conditions during morphology detection.

[0054] In view of the above, the present invention proposes an embodiment of an online experimental device, specifically, as follows: Figure 1 As shown, it includes: an infrared detection correction module, a sampling and morphology detection module, a fluffy state detection module, and an environmental monitoring module.

[0055] The infrared detection correction module is located above the conveyor belt of the production line and is equipped with an online infrared moisture meter 9 and a ranging unit 6. The laser ranging probe in the ranging unit 6 scans the height H of the tobacco material layer in real time and transmits the data to the control unit 7. The control unit 7 calls a preset height correction function to compensate the original reading of the infrared moisture meter 9 in real time according to the deviation between H and the calibrated distance, generating height-corrected moisture content data.

[0056] The sampling and morphology detection module is mainly equipped with a sampling and sorting unit 1 (which can adopt a multi-dimensional motion mechanical gripping mechanism). Under the command of the control unit 7, the tobacco material is grasped at fixed points and in fixed quantities from the conveyor belt to the electric dispersing platform 2. The electric dispersing platform 2 separates the tobacco material into a single layer through a vibrating screen, and the tobacco material falls into the carrier plate 15 of the rotating transmission mechanism 8 below through the screen.

[0057] like Figure 2 As shown, the rotating transmission mechanism 8 drives the carrier plate 15 to rotate along the circumferential sliding track 14 into the shape detection unit 3 via the rotating component 13. The shape detection unit 3 is a fully light-proof box structure, which houses a fixed light source 10, a three-dimensional laser profilometer 11, and a colorimeter 12. In a fully light-proof environment, the fixed light source 10 provides standard illumination conditions, and the three-dimensional laser profilometer 11 scans and acquires the three-dimensional contour data of the tobacco material. The image processing algorithm of the control unit 7 extracts the length, width, and thickness of each individual tobacco material and calculates the statistical average: the average length D. length Average width D width Average thickness D thick Colorimeter 12 acquires color difference data C of tobacco materials. color After the inspection is completed, the carrier plate 15 slides out along the circumferential sliding track 14 to the top of the conveyor belt. The carrier plate 15 is tilted by the further rotation of the rotating component 13, and the inspected tobacco material is poured back to the conveyor belt to achieve automatic recycling.

[0058] Continuing from the previous text, regarding the fluffy state detection module, as follows: Figure 3 As shown, the sampling and sorting unit 1 grabs another portion of tobacco material into the measuring cylinder 16 of the fluffy state detection unit 4. The core components of this unit include the measuring cylinder 16, the pressure probe 17, and the mass sensor 18. During measurement, the pressure probe 17 applies pressure to the tobacco material in the measuring cylinder 16 at a preset standard pressure, and the mass sensor 18 measures the height change of the tobacco material before and after compression. Based on this, the control unit 7 calculates the filling value V of the tobacco material.fill In practice, the fill value can be calculated using the following formula:

[0059] V fill = (m × h0) / (ρ0 × h comp )

[0060] Where m is the mass of the tobacco material, h0 is the initial height, ρ0 is the initial density, and h comp This is the height after compression.

[0061] After the measurement is completed, the measuring cylinder 16 automatically flips over, pouring the tobacco material back onto the conveyor belt, thus completing the automatic recycling.

[0062] The main component of the aforementioned environmental monitoring module is the environmental detection unit 5, which is independently positioned approximately 1 meter away from the device to eliminate interference from the heat generated by the device's operation on the measurement environment. This unit integrates a temperature sensor and a humidity sensor, which collects the ambient temperature (t) and ambient humidity (RH) in real time and transmits the data to the control unit 7 via a communication module.

[0063] Finally, the control unit, serving as the system's central nervous system, comprises a mechanical casing, communication module, central processing unit, display screen, power supply module, and operation buttons. It features data acquisition and communication, motor control, modeling of multiple factors influencing moisture content, and result display. The control unit 7 includes a pre-set timestamp synchronization module to precisely align the sampling time with the measurement time period of the infrared moisture meter 9.

[0064] The process of constructing a model of the relationship between multiple influencing factors of tobacco material moisture content using the above-mentioned experimental apparatus can be referred to in this invention. Figure 4 As shown, the specific steps are as follows:

[0065] Step S1: Control unit 7 acquires the raw moisture content measurement results from the online infrared moisture meter 9 in real time via the communication module, and simultaneously acquires the vertical distance H of the tobacco material surface measured by the laser ranging probe in ranging unit 6. Based on the deviation ΔH = H - H0 between H and the pre-calibrated standard distance H0, control unit 7 calls the height correction function to dynamically compensate for the raw reading of the infrared moisture meter 9. The height correction function can be pre-calibrated through offline experiments and is typically expressed as ΔW = g(ΔH), which represents the functional relationship between the moisture content correction and the height deviation. The corrected moisture content data is recorded in real time and stored in the data cache of control unit 7.

[0066] Step S2: The control unit 7 acquires the ambient temperature and humidity data collected by the environmental detection unit 5, including ambient temperature t and ambient humidity RH. The sampling frequency of the environmental parameters is synchronized with the measurement frequency of the infrared moisture meter 9, typically set to once per second.

[0067] In step S3, control unit 7 sends a sampling instruction to sampling and sorting unit 1 according to a preset sampling cycle (e.g., sampling once every 5 minutes). Sampling and sorting unit 1 performs multi-dimensional motion to grab tobacco material samples from the conveyor belt at fixed points and in fixed quantities, and transfers the samples to electric dispersion platform 2. The sampling amount is set according to subsequent testing requirements, usually grabbing 10~30g of tobacco material each time, of which a portion (about 5~10g) is used for morphology detection and another portion (about 5~10g) is used for fluffiness detection.

[0068] Step S4: Control unit 7 performs timestamp synchronization alignment. In actual operation, the built-in timestamp synchronization module of control unit 7 records the start time T of the grasping action of sampling and sorting unit 1. sample Simultaneously, the infrared moisture meter 9 was extracted from the data cache within the time interval [T]. sample -Δt, T sample All corrected moisture content measurements recorded within +Δt are used to calculate the arithmetic mean of these measurements as the standard moisture content W of the tobacco material sample. The value of the time window Δt is determined based on the conveyor belt speed and the residence time of the material in the detection process, and is usually set to 3 to 10 seconds. This spatiotemporal alignment mechanism ensures a strict correspondence between the physical characteristics of the material obtained by offline detection and the moisture content data measured online.

[0069] Step S5: After being dispersed by vibration on the electric dispersion platform 2, the sampled tobacco material falls through a single layer of mesh into the carrier plate 15 of the rotating conveyor mechanism 8. The control unit 7 drives the rotating component 13 to rotate, and the carrier plate 15 slides along the circumferential sliding track 14 into the light-proof box of the morphology detection unit 3. Under completely light-proof conditions, the fixed light source 10 provides standard illumination, and the three-dimensional laser profilometer 11 performs three-dimensional scanning of the tobacco material, collecting the contour point cloud data of each individual tobacco material; the control unit 7 processes the point cloud data and calculates the average length D of the tobacco material. length Average width D width Average thickness D thick Simultaneously, the colorimeter 12 collects color data of the tobacco material and outputs color difference data C. color Control unit 7 records the current light intensity L of fixed light source 10.

[0070] Another portion of the sampled tobacco material is directly fed into the measuring cylinder 16 of the fluffy state detection unit 4. The pressure probe 17 applies pressure to the tobacco material in the measuring cylinder 16 at a preset standard pressure (e.g., 20N). The mass sensor 18 measures the height change before and after compression, and the control unit 7 calculates the filling value V accordingly. fill .

[0071] After all tests are completed, the carrier plate 15 of the rotary conveyor mechanism 8 slides along the circumferential sliding track 14 to the top of the conveyor belt. The rotating component 13 rotates further to tilt the carrier plate 15, turning the material back onto the conveyor belt. The measuring cylinder 16 of the fluffy state detection unit 4 also automatically flips to complete the material recovery. Throughout the entire process, the tobacco material is always in a closed or controlled flow path, avoiding material spillage and contamination.

[0072] In step S6, the control unit 7 organizes all the data obtained in steps S1 to S5 into a complete record, including: ambient temperature t, ambient humidity RH, vertical distance from the material surface H, illumination intensity of the fixed light source L, and average length D of the tobacco material. length Average width D width Average thickness D thick and color difference data C color and fill value V fill In addition to the standard moisture content W, each set of data is assigned a unique timestamp and stored in the database of the control unit 7.

[0073] In step S7, control unit 7 repeats steps S1 to S6 according to a preset sampling cycle, recording N sets of test data. Those skilled in the art will understand that the value of N is determined comprehensively based on modeling accuracy requirements and production line runtime, typically requiring N ≥ 500 sets to ensure sufficient statistical representativeness of the model. Natural variations in test data caused by subjective and objective factors such as batch size, origin, and process adjustments during normal tobacco material production are considered valid samples, eliminating the need for additional perturbation experiments.

[0074] Step S8: Control unit 7 processes the collected N sets of detection data to establish a multi-factor relationship model for the standard moisture content W of tobacco materials. This model can be expressed as:

[0075] W=f(t,RH,H,L,D length D width D thick C color V fill )

[0076] Where W is the standard moisture content of the tobacco material, f is the nonlinear mapping function to be established, t is the ambient temperature, RH is the ambient humidity, H is the vertical distance from the material surface, L is the illumination intensity of the fixed light source, and D... length D is the average length of the tobacco material. width D is the average width. thick C is the average thickness. color V is a color parameter. fill This is the fill value.

[0077] The specific process of model building includes the following sub-steps:

[0078] Sub-step S8.1: Control unit 7 preprocesses the acquired N sets of detection data. First, outlier detection and removal are performed: box plot method or 3σ principle is used to identify and remove outlier data points that exceed the normal range. Then, the input features (t, RH, H, L, D) are processed. length D width D thick C color V fill Normalization is performed to map the value range of each feature to the [0,1] interval or standardize it to a distribution with a mean of 0 and a standard deviation of 1. Specifically, the normalization method can be, but is not limited to, using the Min-Max normalization formula: x'=(xx min ) / (x max -x min ).

[0079] After preprocessing, the dataset can be divided into a training set (60%), a validation set (20%), and a test set (20%) according to a preset ratio as needed. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning and early stopping judgment, and the test set is used for independent evaluation of the final model performance. This invention will not elaborate on these aspects.

[0080] Sub-step S8.2: Construct a feedforward neural network (FNN) model. The structural design of this model can be referenced as follows:

[0081] The input layer has the same number of neurons as the number of preprocessed input features, a total of 9 neurons, each corresponding to one of the 9 input features.

[0082] Hidden layers, using 1 to 3 hidden layers, each containing 32 to 128 neurons, with the specific number determined through cross-validation based on the dataset size and modeling accuracy requirements; the hidden layers employ the ReLU (Rectified Linear Unit) activation function, mathematically expressed as ReLU(x) = max(0, x), which effectively alleviates the vanishing gradient problem and accelerates model convergence; optionally, a Dropout layer (with a dropout rate typically set to 0.2 to 0.5) is introduced into the hidden layers to prevent overfitting.

[0083] The output layer contains a neuron that uses a linear activation function to output the continuous standard moisture content W of the tobacco material (expressed as a percentage, typically ranging from 8% to 20%).

[0084] Furthermore, the mean squared error (MSE) can be chosen as the loss function, but is not limited to, and the Adam optimizer can be used for model training. The initial learning rate is set to 0.001, which can be adaptively adjusted according to changes in the loss value during training. Optionally, an L2 regularization term can be introduced into the loss function to prevent model overfitting.

[0085] Sub-step S8.3: Iteratively train the constructed neural network model using the training set. In each iteration (Epoch), all training samples are input into the model for forward and backward propagation, updating the network weights and bias parameters. During the training process, after each Epoch, the loss value of the current model is calculated on the validation set to monitor the generalization performance of the model.

[0086] Sub-step S8.4: Solidify the parameters of the neural network model that has obtained the best performance after training and evaluation, including the weight matrix and bias vector of each layer, to form a deployable multi-factor relationship model. The above are all for reference and not limitation. For the solidified model, it is preferred to use a general format for storage so as to facilitate integration into the production control system for real-time moisture content prediction and infrared moisture meter reading correction.

[0087] Finally, it should be added that the deployed model can be used in, but is not limited to, the following scenarios:

[0088] (1) During the production process, the environmental parameters, distance measurement height, light intensity of fixed light source and physical characteristic parameters of tobacco material collected in real time are input into the solidification model, and the predicted standard moisture content value can be output in real time for online quality control.

[0089] (2) The output of the solidification model is used as a reference standard and compared with the real-time reading of the infrared moisture meter. The measurement deviation is dynamically calculated and a correction coefficient is generated to achieve adaptive calibration of the infrared moisture meter.

[0090] (3) By analyzing the sensitivity of each input feature to the predicted moisture content through model analysis, the contribution rate of each factor to the moisture content measurement deviation is quantified, providing data support for the adjustment of process parameters of the production line (such as conveyor belt speed, material spreading thickness control, etc.).

[0091] In summary, the core concept of this invention is embodied in: integrating the sampling and sorting unit, the electric dispersion platform, the morphology detection unit, the fluffiness detection unit, the environmental detection unit, and the ranging unit onto the same experimental platform, and having the control unit uniformly coordinate the timing and data processing, thereby realizing the synchronous online acquisition of multi-source data;

[0092] Although infrared moisture meters are widely used in the tobacco industry, the problem of measurement drift caused by changes in the height of the material pile has long remained unresolved. Furthermore, a feasible solution has yet to emerge in the industry that integrates a high-precision laser ranging probe with an infrared moisture meter and eliminates height deviation through dynamic compensation. Therefore, this invention proposes using a high-precision laser ranging probe to measure H in real time, and using a height correction function called by a control unit for dynamic compensation, thus achieving real-time correction of changes in the measurement focal length caused by the unevenness of the tobacco material surface.

[0093] It should also be noted that the shape (length, width, thickness) and color of tobacco materials have a significant impact on the measurement accuracy of infrared moisture analyzers. In order to improve comparability, this invention proposes to set a fixed light source in a fully light-proof box to eliminate the interference of external ambient light and ensure that the detection of all samples is carried out under the same lighting conditions, thereby making the data collected by the three-dimensional laser profilometer and colorimeter highly consistent and comparable.

[0094] Furthermore, this invention employs a feedforward neural network to learn a nonlinear mapping of standard moisture content from multidimensional input features, which can accurately characterize the complex interaction between environmental factors and the physical properties of the material itself. It can not only quantify the contribution rate of each factor, but also be used for real-time prediction of moisture content and dynamic correction of infrared moisture meters, providing a new technical path for intelligent control of tobacco production processes.

[0095] As can be seen from the above, the online experimental device and modeling method provided by this invention are applicable to the tobacco processing production lines of various cigarette manufacturing enterprises, especially for high-end cigarette brands with high requirements for moisture content control accuracy. It can be retrofitted to existing production lines without significant adjustments to the original equipment layout. The fully automated sampling, detection, and recycling closed-loop design ensures continuous operation of the production line without interfering with normal production. Furthermore, the multi-factor relationship model provides real-time decision support for moisture content control during the production process. Clearly, the overall solution of this invention has excellent application prospects and value in the tobacco industry.

[0096] In this invention, when directional terms are mentioned, they are relative concepts based on the embodiments. Furthermore, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0097] The above description of the structure, features, and effects of the present invention is based on the embodiments shown in the figures. However, the above are only preferred embodiments of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched by those skilled in the art to form a variety of equivalent solutions without departing from or changing the design concept and technical effects of the present invention. Therefore, the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. An online experimental device for modeling the relationship between multiple influencing factors of moisture content in tobacco materials, characterized in that, include: The sampling and sorting unit (1) has multi-dimensional motion function and quantitative grasping function, and is used to grasp tobacco materials at fixed points and in a quantitative manner from the conveyor belt. The electric dispersion platform (2) disperses the tobacco material through the vibration of the screen platform, so that the tobacco material is separated into a single layer and falls into the support plate (15) below through the screen. The morphology detection unit (3) includes a light-shielding box and a rotating transmission mechanism (8). The light-shielding box is equipped with a fixed light source (10), a three-dimensional laser profilometer (11) and a colorimeter (12), which are used to acquire three-dimensional morphology data and color difference data of tobacco materials under constant illumination. The fluffy state detection unit (4) is used to measure the filling value of tobacco material; An environmental monitoring unit (5) is set on the measurement platform of the experimental device to acquire environmental temperature and humidity data; The ranging unit (6) uses a laser ranging probe installed near the online infrared moisture meter (9) on the production line to measure the vertical distance H from the probe end face to the surface of the tobacco material layer; as well as The control unit (7) includes a mechanical housing, a communication module, a central processing unit, a display screen, a power supply module and operation buttons, and has functions of data acquisition and communication, motor control, modeling of multiple factors affecting moisture content and display of results.

2. The online experimental device according to claim 1, characterized in that, The rotary transmission mechanism (8) includes a rotating component (13), a circumferential sliding track (14), and a carrier plate (15). The rotating component (13) drives the carrier plate (15) to slide along the circumferential sliding track (14) under the drive of the motor. The carrier plate (15) is positioned below the electric dispersing platform (2), inside the light-proof box, and above the conveyor belt in sequence to assist in the transportation, measurement, and recycling of tobacco materials.

3. The online experimental device according to claim 1, characterized in that, The fluffy state detection unit (4) includes a measuring cylinder (16), a pressure probe (17), and a mass sensor (18). The pressure probe applies pressure to the tobacco material in the cylinder with a preset standard pressure. The mass sensor is used to measure the height of the tobacco material before and after compression to calculate the filling value. The measuring cylinder has a flipping function for material recycling.

4. The online experimental device according to claim 1, characterized in that, The dimensional measurement error of the three-dimensional laser profilometer (11) is within ±0.02mm, and the measurement error ΔE of the colorimeter (12) is within ±0.

05.

5. The online experimental device according to claim 1, characterized in that, The environmental detection unit (5) includes a temperature sensor and a humidity sensor. The distance between the sensor and the tobacco material being tested is no more than 1m. The temperature measurement error is within ±0.5℃ and the humidity measurement error is within ±3%RH.

6. A method for modeling the relationship between multiple influencing factors of tobacco material moisture content using the online experimental device described in any one of claims 1 to 5, characterized in that, include: Step S1: The control unit (7) obtains the original moisture content measurement results of the online infrared moisture meter (9) and the vertical distance H of the tobacco material surface measured by the laser ranging probe in the ranging unit (6). The control unit (7) corrects the height deviation of the real-time reading of the infrared moisture meter (9). Step S2: The control unit (7) acquires the ambient temperature and humidity data collected by the environmental detection unit (5); Step S3: Grab tobacco material samples from the conveyor belt through the sampling and sorting unit (1); Step S4: The control unit (7) performs timestamp synchronization alignment. Using the built-in timestamp synchronization module of the control unit (7), the average corrected moisture content within the preset time window before and after the sampling time is extracted as the standard moisture content measurement result data W of the tobacco material. Step S5, the sampled tobacco material is divided into two parts: one part is sent into the shape detection unit (3) through the electric dispersion platform (2) and the rotary transmission mechanism (8) for detection, and the length average D length , the width average D width , the thickness average D thick and the color difference data C color of the sample are obtained; the other part is put into the bulk state detection unit (4) for detection, and the filling value V fill of the tobacco material is obtained; after the detection is completed, all the tobacco materials are automatically recycled to the conveyor belt; Step S6: The control unit (7) records all detection data as a group; Step S7: Repeat steps S1 to S6 to record N sets of detection data; Step S8, the N groups of detection data are classified and processed according to influencing factors, taking environmental temperature t, environmental humidity RH, vertical distance H of the material surface and fixed light source illumination brightness L as fast-changing compensation factors, taking length average D length , width average D width , thickness average D thick of the tobacco material and color difference data C color and filling value V fill as slow-changing state characteristic factors, and the following relationship model is established: W = f(t, RH, H, L, D length ,D width ,D thick ,C color ,V fill ) Where W is the standard moisture content of the tobacco material, and f is the nonlinear mapping function to be established.

7. The modeling method according to claim 6, characterized in that, The model building process in step S8 includes the following sub-steps: The N groups of tobacco material detection data obtained in steps S1 to S7 are preprocessed, including normalization or standardization processing on input features, environment temperature t, humidity RH, material surface vertical distance H, fixed light source illumination brightness L, length average D length , width average D width , thickness average D thick , and color difference data C color and filling value V fill ; The preprocessed dataset is then divided into training, validation, and test sets according to a preset ratio. The architecture for constructing a feedforward neural network model specifically includes: an input layer with the same number of neurons as the number of preprocessed input features; one or more hidden layers using nonlinear activation functions; and an output layer containing a neuron using a linear activation function to output the continuous standard moisture content W of tobacco materials. The model selects mean squared error or mean absolute error as the loss function and uses an optimizer for model training. The constructed neural network model is iteratively trained using the training set, and the loss function is minimized by an optimizer. During training, the hyperparameters of the neural network are adjusted based on the model's performance on the validation set. After training, the performance of the final model is evaluated using the test set. The parameters of the neural network model that achieves optimal performance after training and evaluation are solidified to form the multi-influencing factor relationship model.

8. The modeling method according to claim 7, characterized in that, The neural network model incorporates Dropout or L1 / L2 regularization techniques during training to prevent overfitting, and employs an early stopping strategy to optimize the model structure.

9. The modeling method according to claim 6, characterized in that, In step S5, the carrier plate (15) of the rotating transmission mechanism (8) is driven by the rotating component (13) to slide along the circumferential track (14) to complete the action cycle of receiving materials, sending them into the shape detection unit (3) and dumping and recycling materials.

10. The modeling method according to any one of claims 6 to 9, characterized in that, In step S1, when correcting the height deviation of the real-time reading of the infrared moisture meter (9), the control unit (7) calls the preset height correction function to compensate the original measurement value of the infrared moisture meter according to the deviation between the vertical distance H of the material surface measured by the laser ranging probe and the calibrated distance.

Images