Intelligent mechanism carbon production whole-process regulation method

Abstract

The application relates to the technical field of industrial intelligent control, and discloses a full-process intelligent mechanism carbon production regulation method, which comprises the following steps: constructing a digital twin model, and inversely deriving a real-time effective permeability tensor of a bed layer; the application can accurately estimate the change of an airflow channel in a carbonization process, and further deduce the dynamic evolution of a main channel skeleton; the total heating power of a furnace and the bed layer pressure difference are accurately adjusted, the carbonization process is accurately controlled, the airflow distribution and the temperature field are optimized, and the quality fluctuation of products and energy consumption are reduced; and the application can significantly improve the production efficiency, quality consistency and resource utilization rate of mechanism carbon in actual production.

Description

Technical Field

[0001] This invention relates to the field of industrial intelligent control technology, and more specifically, to an intelligent method for controlling the entire process of machine-made charcoal production. Background Technology

[0002] Mechanized charcoal production uses biomass waste (such as sawdust, straw, bamboo shavings, etc.) as raw materials, and produces charcoal materials with high specific surface area and low ash content through high-temperature pyrolysis or carbonization. Traditional mechanized charcoal production uses screw extrusion molding technology to form hollow fuel rods from biomass raw materials. The fuel rods are precisely stacked in a carbonization furnace and subjected to pyrolysis and carbonization under controlled temperature, atmosphere, and residence time conditions to remove organic components and increase the specific surface area and porosity of the charcoal.

[0003] Currently, the production process of machine-made charcoal faces a series of technical challenges, particularly in terms of temperature field, airflow distribution, and the uniformity of the carbonization process. With the continuous expansion of production scale, traditional methods are no longer sufficient to meet the requirements of efficient, stable, and uniform carbonization, especially the carbonization differences caused by different briquettes in the same furnace and the increasing inhomogeneity of carbonization during the production process. Local temperature differences, bed shrinkage, crack formation, and pore evolution during the carbonization process significantly affect the final quality of the charcoal product.

[0004] With the rapid development of intelligent technologies, machine learning, data acquisition, and real-time feedback control are playing increasingly important roles in the carbonization process. Particularly in biochar production, more and more research is exploring the use of sensor networks, digital twin technology, and artificial intelligence algorithms to improve the accuracy and automation of the production process. However, most existing intelligent controls focus on the macroscopic level, relying heavily on one- or two-dimensional monitoring parameters such as temperature, humidity, and feed rate.

[0005] However, the unique characteristics of machine-made charcoal production lie in the complexity of its raw material particle morphology and the high coupling between airflow and heat transfer within the furnace. Traditional intelligent solutions often fail to fully perceive and control the anisotropic channel structure within the bed and the local heterogeneity during the carbonization process. For example, the carbonization process of briquettes at different levels within the stacked bed varies significantly, and the evolution of airflow channels and crack formation during carbonization have a substantial impact on carbonization quality. Existing intelligent systems lack real-time monitoring and regulation of these inherent structural factors. Summary of the Invention

[0006] This invention provides an intelligent method for controlling the entire process of machine-made charcoal production, which solves the technical problems mentioned in the background art.

[0007] This invention provides an intelligent method for controlling the entire process of machine-made charcoal production, comprising:

[0008] A stacking structure flow guidance model based on the stacking method of fuel bars was constructed, and normalized thermal monitoring data sets were generated by collecting production process data.

[0009] Using a digital twin model and constrained by the stacked structure flow guidance model, the permeability distribution state field inside the bed is inverted from the normalized thermal monitoring data set;

[0010] The flow field analysis was performed on the air permeability distribution state field inside the bed to extract the airflow channelization intensity and the dominant airflow path direction;

[0011] Calculate the carbonization non-uniformity index along the dominant airflow path direction, and fuse the carbonization non-uniformity index with the airflow channelization intensity to construct a flow field carbonization coupling evaluation index.

[0012] Based on the deviation between the flow field carbonization coupling evaluation index and the target value, the total heating power of the furnace and the total pressure difference of the furnace bed are calculated by the continuous control law to regulate the carbonization process.

[0013] The beneficial effects of this invention include: by utilizing digital twin technology to invert the real-time effective permeability tensor and combining it with the self-organizing characteristics of the bed structure, the dynamic changes of the main channel skeleton within the bed are accurately quantified, thereby achieving intelligent sensing and regulation of the carbonization process. By adjusting the total heating power of the furnace and the bed pressure difference, the airflow distribution can be effectively optimized, carbonization non-uniformity can be reduced, and product quality consistency and production efficiency can be improved. This invention can significantly improve the precision and stability of machine-made charcoal production, reduce energy consumption and carbon emissions, and has significant engineering application prospects. Attached Figure Description

[0014] Figure 1 This is a flowchart of an intelligent method for controlling the entire process of charcoal production according to the present invention;

[0015] Figure 2 This is a graph showing the evolution of key indicators throughout the entire process of charcoal production according to the present invention.

[0016] Figure 3 This is a local physical field distribution diagram of the present invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0018] like Figure 1 As shown, an intelligent method for controlling the entire process of machine-made charcoal production includes:

[0019] A stacking structure flow guidance model based on the stacking method of fuel bars was constructed, and normalized thermal monitoring data sets were generated by collecting production process data.

[0020] Using a digital twin model and constrained by the stacked structure flow guidance model, the permeability distribution state field inside the bed is inverted from the normalized thermal monitoring data set;

[0021] The flow field analysis was performed on the air permeability distribution state field inside the bed to extract the airflow channelization intensity and the dominant airflow path direction;

[0022] Calculate the carbonization non-uniformity index along the dominant airflow path direction, and fuse the carbonization non-uniformity index with the airflow channelization intensity to construct a flow field carbonization coupling evaluation index.

[0023] Based on the deviation between the flow field carbonization coupling evaluation index and the target value, the total heating power of the furnace and the total pressure difference of the furnace bed are calculated by the continuous control law to regulate the carbonization process.

[0024] It should be noted that, because the machine-made fuel bars typically employ a hollow cylindrical structure and are arranged regularly within the steel basket or cage car of the carbonization furnace, the resistance to gas flow along the axis of the fuel bar is significantly less than the resistance to flow perpendicular to the stacking direction. Traditional isotropic porous media models cannot describe the main channel effect locked by the geometric structure. Therefore, by introducing a tensor description method, the bed permeability is decomposed into an isotropic baseline component and an incremental component along the geometric principal axes.

[0025] In a preferred embodiment, the system is applied to a horizontal continuous machine-made carbonization production line. First, the system obtains the geometric dimensional parameters of the current batch of machine-made fuel bars, including the outer diameter of the fuel bars, through a human-machine interface or an interface with the production management system (MES). (e.g., 50 mm), inner diameter (e.g., 15 mm) and baton length (For example, 400 mm). Simultaneously, the system reads the stacking layout parameters of the steel baskets inside the furnace, specifically including the number of columns along the furnace width. Number of layers along the height of the furnace and the number of rows along the length of the furnace Based on these discrete geometric parameters, the system first calculates the initial porosity of the bed. Initial porosity The calculation formula is ,in This represents the solid volume of a single firewood stick multiplied by the total number of firewood sticks. ,and The macroscopic envelope volume of the stacked bed is determined by the effective internal volume of the steel basket. Next, the system determines the main design airflow direction in the furnace. Specifically, since the furnace is designed as a horizontal plug flow type, the main design flow direction of the flue gas is parallel to the furnace length axis; therefore, a unit vector of the main design airflow direction is defined. (Assuming the furnace length direction is the x-axis). Subsequently, the system construction is defined in spatial location. Permeability Tensor Model This tensor model is characterized as In the formula, represent The identity matrix is ​​used to characterize the basic permeability of the medium; The tensor product operation, representing vectors, yields a result. The matrix is ​​used to characterize directionality; The benchmark value for isotropic permeability. This represents the incremental share of penetration along the main direction of the designed airflow.

[0026] In a preferred embodiment, the formula is as follows: and The system is based on the ratio of the inner diameter to the outer diameter of the mechanical rod (i.e., the hollow ratio). and initial porosity Perform quantitative calculations. Specifically, The formula is calculated using the modified Kozeny-Carman equation. ,in It is the first flow resistance structure constant, and its value is usually between 0.003 and 0.005. The preset empirical value of 0.0042 is taken. The characteristic radius of the particle is used in the calculation. (This refers to the percentage of incremental penetration.) Its calculation formula is ,in It is the second flow resistance anisotropy enhancement coefficient, used to reflect the short-circuit contribution of the hollow channel to the airflow. Its value is determined by cold pressure drop experiments, and the preferred value is 25.0. The term demonstrates that the larger the hollow aperture, the quadratic increase in axial fluid conductivity. Ultimately, a model incorporating geometric prior information is generated. Diagonally dominant matrix field.

[0027] In some possible embodiments, this is applied to a vertical intermittent carbonization furnace. Specifically, the fuel rods are stacked vertically, and the airflow design direction is from bottom to top; therefore, the determined design airflow main direction unit vector is... (i.e., along the z-axis). In calculating the initial porosity... When stacking firewood vertically, considering the interlocking between the firewood bars, the system adopts a modified packing density formula. ,in The theoretical hexagonal closest packing porosity, This is a correction term caused by the bending of the fuel rod. The percentage of the penetration increment is used when constructing the penetration tensor model. The calculation formula is adjusted to The added items Used to characterize the rectifying effect of aspect ratio on vertical flow channels. The aspect ratio influence factor is a preset value, preferably 5.0.

[0028] In some possible embodiments, this is applied to a moving bed carbonization environment with irregular stacking. Specifically, although the macroscopic airflow direction... The dimensions are still determined by the furnace body geometry, but the ratio of the inner diameter to the outer diameter is no longer fixed; instead, it follows a normal distribution. The system first obtains the statistical average of the fuel bar dimensions. and Calculate the average hollow ratio Instead of using fixed analytical formulas to determine permeability parameters, a pre-built neural network model is invoked. The system will Input this neural network and output directly and The value of is determined by the neural network, which is trained using a large amount of offline computational fluid dynamics (CFD) simulation data and is capable of capturing nonlinear seepage characteristics at high Reynolds numbers. Specifically, The construction still follows It takes the form of a tensor, but the method of obtaining parameters changes from empirical formulas to data-driven models, thereby improving its adaptability to complex working conditions.

[0029] It should be noted that the physical fields within the carbonization furnace involve temperature (Celsius or Kelvin, typically on the order of 100 to 1000), gas concentration (percentage or ppm, on the order of 0 to 100 or greater), pressure difference (Pascals, on the order of 10 to 5000), and power control (kilowatts, on the order of 10 to 200). These data have vastly different units and magnitudes. If the raw data is directly input into the digital twin inversion model, large variables (such as pressure difference) will mask the gradient information of smaller variables (such as oxygen concentration), causing the inversion algorithm to struggle to converge or get trapped in local minima. Therefore, by synchronously acquiring data and using Z-score linear normalization based on historical statistics, all physical quantities are mapped to a dimensionless feature space that follows a standard normal distribution, thereby eliminating the influence of dimensions and improving the numerical stability of the inversion algorithm.

[0030] In a preferred embodiment, a Supervisory Control and Data Acquisition (SCADA) system integrated with the Industrial Internet of Things (IIoT) is used. The system sets a sampling period. For 5 seconds. At each sampling time The system is installed at different heights on the side wall of the furnace. A K-type thermocouple (e.g.) Simultaneous acquisition of multi-point temperature data within the furnace. to The unit is degrees Celsius; data is collected by an online infrared gas analyzer installed at the flue gas outlet. Concentration data for key components, specifically including carbon monoxide concentration. carbon dioxide concentration and oxygen concentration The unit is volume percentage; the measured total pressure difference data at the furnace inlet and outlet is collected by a high-precision differential pressure transmitter. The unit is Pascal. Simultaneously, the system synchronously reads the current control command status, including the total furnace heating power setpoint, via a PLC communication protocol (such as Modbus TCP). (Unit: kilowatts) and the differential pressure setting value corresponding to the induced draft fan inverter. (Unit: Pascal). The system assembles all collected instantaneous values ​​into a single dimension in a fixed order. Original measurement data column vector Next, the system retrieves historical statistical parameters stored in the database, namely the historical average vector. and historical standard deviation vector These historical parameters are derived from statistical data collected over the past month on a single batch of gold-quality steel bars produced on the same production line. The system analyzes each component of the original measurement data set. Perform linear standardization; the specific calculation formula is as follows: In the formula, Indicates the first The normalized value of each component is a dimensionless real number. This is the number collected at the current moment. The original measured values ​​of each physical quantity; It is the first The historical average value corresponding to each physical quantity is used to eliminate the center offset of the data; It is the first The historical standard deviations of each physical quantity are used to eliminate data dispersion differences. Finally, the system output consists of all... Normalized thermal monitoring data set .

[0031] In some possible embodiments, considering the potential for electromagnetic interference in industrial environments that could cause instantaneous jumps in sensor readings, a data preprocessing step is added between forming the original measurement data set and the linear normalization process. Specifically, before executing the normalization formula, the original measurement vectors are preprocessed. Each component in the filter is subjected to a sliding weighted filter. The filtering formula is as follows: ,in These are the filtered measurement values. This is the raw value currently being collected. It is the filter coefficient (with a value of 0.8). This is the filtered value from the previous time step. Subsequently, the filtered value is used. Alternative Substitute into the standardized formula In this study, by introducing a low-pass filtering mechanism, the interference of high-frequency noise on the normalization result was effectively suppressed, resulting in a more accurate and efficient feature vector. It can better reflect the true thermal trend of the carbonization process.

[0032] In some possible embodiments, dynamically updated historical statistical parameters are used to address potential sensor zero-point drift or slow changes in furnace characteristics that may occur after long-term operation of production equipment. Specifically, and It is not a fixed constant, but a dynamic statistic calculated based on a rolling time window. The system maintains a length of... The historical database (e.g., the most recent 50 production batches) is used to recalculate the average of all data within that window before each new batch of production begins. and standard deviation Within the current batch sampling period, use the updated version. and Substitute into the formula The calculations are performed. This makes the standardization process adaptive, automatically compensating for data baseline drift caused by equipment aging or seasonal environmental changes, ensuring that the normalized data always remains within the ideal distribution range with a mean of 0 and a variance of 1, thereby guaranteeing the long-term effectiveness of the digital twin model.

[0033] It should be noted that, due to the impossibility of deploying dense flow sensors within the high-temperature, sealed bed of the carbonization furnace, the distribution of internal airflow channels can only be indirectly inferred through the response of external macroscopic thermal quantities. This is a typical ill-conditioned inverse Bernoulli problem or inverse problem, where countless combinations of internal states may correspond to the same external observation result. To eliminate this uncertainty and obtain a reliable solution, structural prior constraints are introduced. A flow guidance model based on stacked geometry is used as an anchor point or regularization term, forcing the inversion result to not deviate too far from the framework determined by the stacked structure. Thus, the digital twin model is not merely a black-box predictor, but a solver integrating various physical mechanisms (momentum / energy equations) and geometric constraints (stacked structure), thereby accurately reconstructing the permeability tensor field of the bed with limited sensor data.

[0034] In a preferred embodiment, execution takes place in the real-time computing unit of an industrial control computer. First, the system constructs a digital twin model describing the momentum and energy transfer between the gas and solid phases within the furnace. Considering the stringent computational speed requirements (millisecond-level response) of real-time control, this digital twin model is not directly run using time-consuming computational fluid dynamics (CFD) mesh simulations, but rather a pre-trained reduced-order surrogate model (ROM). This reduced-order surrogate model is formalized as a nonlinear vector function. In the formula, It is the normalized thermal data vector predicted by the model, and its dimensions are the same as those of the measured normalized data set. same; This represents a function mapping relationship built based on a deep neural network (DNN). These are the network weight parameters determined through training with a large amount of CFD simulation data during the offline phase. It is the unknown quantity to be solved at the current moment, namely the effective permeability tensor field inside the bed; and The current control input is known. Next, the system defines the inversion objective function. This function is composed of a weighted sum of two parts: a data fidelity term and a structural constraint term. The specific calculation formula is as follows: In the formula, It is the first one after collection and normalization. Each measured data component; The output of the digital twin model is the first One predicted component; These are preset weighting coefficients used to adjust the importance of different physical quantities (such as temperature and pressure) in error calculation. For example, they can give higher weight to differential pressure data. (The rest are 1.0), because pressure difference is most sensitive to changes in the flow field; It is the regularization coefficient, with a value of 0.01, used to balance observation error and structural bias; It is the static stacking structure flow guidance model tensor; The Frobenius norm, which is the square root of the sum of the squares of all elements in the matrix, is used to quantify the permeability tensor retrieved in the current inversion. With theoretical template The overall structural distance between them is calculated. Finally, the gradient descent iterative algorithm is used to solve for this distance. System initialization. Then iterate at each step In, according to the formula An update will be performed. Among other things, This is the learning rate (step size), with a value of 0.001; It is the gradient of the objective function with respect to the permeability tensor, calculated using automatic differentiation. When the change in the objective function... Less than the preset convergence threshold (For example When the iteration terminates, the final output is given. This represents the current state of the air permeability distribution within the bed layer.

[0035] In some possible embodiments, the digital twin model is constructed using a simplified physical mechanism model rather than a neural network proxy model. Specifically, the furnace and bed are discretized into a thermal-fluid network consisting of several nodes. The momentum transfer process follows the flow path equations constructed using Kirchhoff's laws, and the energy transfer follows the nodal thermal balance equations. Digital Twin Model The output is predicted by quickly solving this system of algebraic equations. When defining the inversion objective function, the structural deviation constraint term adopts the form of a weighted Mahalanobis distance, i.e. ,in It uses a pre-defined covariance matrix based on the local variance of the stacking structure, allowing the inversion results to deviate more from the template in loosely stacked regions (with high variance) and less in densely stacked regions. This makes the inversion process more interpretable and does not require a large amount of offline training data, making it suitable for small sample scenarios.

[0036] In some possible implementations, to address the problem that the gradient descent algorithm may get trapped in local optima, an adaptive moment estimation optimization algorithm with a momentum term (Adam) is used instead of the standard gradient descent method. In the update solution step, the system maintains the first-order moment estimate. and second-order moment estimation The updated formula has been adjusted to .in, and These are the first and second moments after bias correction. It is a small constant to prevent the denominator from being zero. This improvement allows the algorithm to decrease rapidly in the early stages of inversion, and automatically reduce step size oscillations when approaching the optimal solution. This improves the accuracy of the inverted permeability distribution state field while ensuring real-time performance, and enables the capture of more subtle channelization precursor features.

[0037] It should be noted that the permeability tensor It is a second-order symmetric positive definite tensor, which describes the ease or difficulty of fluid flow in various directions within a porous medium. However, a The matrix contains six independent components. Directly using it as a feedback signal to input the control loop would lead to the curse of dimensionality and lack of intuitive interpretability. Therefore, spectral decomposition is used to find the principal axis system of this tensor. The direction corresponding to the largest eigenvalue is the direction of minimum fluid flow resistance (i.e., the main channel direction), while the degree of difference in eigenvalues ​​(anisotropy ratio) reflects whether the fluid is biased, i.e., whether it flows only through a single channel while ignoring other regions. Channelization intensity is extracted... and deflection angle The system can compress the complex internal flow field state into two key pieces of information: how much the flow is deflected and in which direction it is deflected.

[0038] In a preferred embodiment, the linear algebra operation module is deployed in an edge computing gateway. The system receives the state field of the air permeability distribution inside the bed. Specifically, to simplify calculations and obtain global features, the system first performs a volume-weighted average of the local permeability tensors across the entire field, obtaining a value representing the average state of the entire bed. Global effective penetration tensor matrix Next, the system calls the symmetric matrix eigenvalue solver in a numerical algebra library (such as LAPACK) to solve the problem. Perform eigenvalue decomposition. The decomposition process satisfies the mathematical relation... ,in For scalar eigenvalues, Let be the corresponding eigenvector. The system solves for three real eigenvalues ​​and sorts them in descending order, denoted as . .at this time, This is the maximum principal eigenvalue, representing the fluid conductivity in the direction of greatest permeability; The minimum principal eigenvalue represents the capability in the most difficult penetration direction; and... corresponding unit eigenvector (i.e., after normalization) This is identified as the main feature vector. Subsequently, the system calculates the airflow channelization intensity. The calculation formula is: In the formula, It is a dimensionless real number greater than or equal to 1; when The time indicates that the bed is isotropic and has no channels; The larger the value, the greater the penetration rate along the main channel, indicating a more significant channelization effect. Simultaneously, the system extracts the dominant airflow path direction and directly takes... Finally, the system calculates the spatial deflection angle. The system retrieves the unit vector of the main airflow direction in the furnace design. Calculate using the vector dot product formula: In the formula, Represents the vector dot product. The Euclidean norm (modulus) of a vector. The function is an inverse cosine function, and the calculation result is converted into an angle value (0 to 90 degrees). The absolute value is taken because the fluid flowing in either the forward or reverse direction along the channel is collinear on the geometric path; this angle... Quantify the degree to which the actual self-organized airflow skeleton deviates from the design axis.

[0039] In some possible embodiments, considering the non-uniformity of the bed's internal structure, this method does not perform global averaging of the tensor, but instead employs a worst-case point extraction method. The system processes each spatial grid node obtained from the inversion. Local tensor at Eigenvalue decomposition is performed separately to calculate the local channelization intensity. Then, the system iterates through all nodes and selects... As the current system airflow channelization intensity Specifically, quality bottlenecks in charcoal production often stem from localized overheating or dead zones. Therefore, capturing the most severe localized channelization characteristics within the production process is more effective in ensuring product consistency than simply observing average values. In this case, the dominant airflow path direction... The direction of the maximum principal feature vector corresponding to the point of maximum intensity is also selected accordingly.

[0040] In some possible embodiments, to operate on computationally limited embedded controllers, a simplified anisotropy metric is used instead of the full eigenvalue decomposition ratio. Specifically, the system utilizes tensor invariants to estimate the channelization strength. The specific calculation formula is adjusted as follows: This borrows from the definition of fractional anisotropy (FA). In the formula, It is the arithmetic mean of the three eigenvalues. Although the formula has a different form, it also represents the degree to which the tensor deviates from spherical symmetry (i.e., the degree of anisotropy), and the numerical range is normalized to between 0 and 1, which facilitates the standardization processing of subsequent control algorithms. At this time, the determination of the dominant airflow path direction adopts the power iteration method to quickly solve for the principal vector corresponding to the largest eigenvalue, avoiding the computational overhead of full spectrum decomposition.

[0041] It's important to note that in charcoal production, high temperature does not equate to good carbonization. Carbonization is a cumulative chemical reaction process influenced by both time and temperature. Especially under conditions of gas channeling, the high-temperature hot gas flow primarily follows the main channel framework of least resistance, resulting in extremely rapid heating of the fuel rods located along this channel path, while the surrounding areas experience a lag. Therefore, simply monitoring the average furnace temperature cannot reflect the true differences in product quality. By virtually sampling in a digital twin model, the temperature history of a series of control volumes is traced along the identified dominant gas flow path (i.e., the heat transfer highway), and the cumulative reaction conversion rate is calculated using Arrhenius kinetics. This method transforms static spatial inhomogeneity into dynamic differences in reaction progress, accurately identifying the risk of both over-burning and under-burning, and providing feedback signals with chemical mechanism support for the control system.

[0042] In a preferred embodiment, this is executed by a post-processing module of a digital twin engine embedded in the control system. The system first bases the processing on the dominant airflow path direction vector. A virtual ray is constructed from the geometric center point of the furnace chamber, running through the entire bed. The system then uniformly selects elements along this ray within the grid of the digital twin model. Representative control volumes (RCVs), for example, selecting Each point covers the entire path from the intake to the exhaust end. The system then accesses the historical database of the digital twin model to extract each control volume. (in From the time of loading the furnace up to the current moment Complete temperature history data The unit is Kelvin (K). Next, the system uses a first-order pyrolysis kinetic model to calculate the current local degree of carbonization at each point. The specific calculation formula uses integral form: In the formula, It is a dimensionless scalar between 0 and 1, where 0 represents that the raw material has not reacted at all, and 1 represents that it has been completely converted into charcoal. The reaction frequency factor (pre-exponential factor) is preset based on the characteristics of sawdust processing rods. ; The apparent activation energy is set to [value]. ; Let be the ideal gas constant, and take the value of . ; For integration variables The instantaneous temperature at time t. The system uses the trapezoidal rule or Simpson's rule to numerically solve the above integral. After obtaining all... After determining the local carbonization degree at each point, the system calculates the arithmetic mean. Finally, the system calculates the statistical value of the distribution dispersion, i.e., the standard deviation. As an indicator of carbonization inhomogeneity, the calculation formula is: This indicator The larger the value, the more significant the difference in reaction progress between upstream and downstream along the main airflow channel, indicating a severe unevenness.

[0043] In some possible embodiments, to improve the real-time performance of the computation and adapt to the computing power limitations of embedded chips, a recursive algorithm in the form of differentiation is used instead of integral operations. Specifically, local carbonization... The update no longer requires backtracking through the entire historical data, but is based on the carbonization degree at the previous sampling time. Perform incremental calculations. The calculation formula is adjusted as follows: In the formula, The sampling period for the control system (e.g., 5 seconds). This reflects the fact that the reaction rate gradually decreases as reactants are consumed.

[0044] In some possible embodiments, considering that the dominant airflow path may not be a strictly straight line, but rather a stream tube of a certain width, the method employs a radial weighting method when selecting the representative control volume. The system not only selects the principal eigenvector... Points on the axis are also selected with the axis as the center and a radius of [missing information]. All grid nodes within a cylindrical region (e.g., 100 mm) are used as the control volume set. When calculating the final carbonization inhomogeneity index, the system introduces a flow velocity-related weighting coefficient. For control volumes with higher flow rates, the weighting is... The larger the value, the more significant the impact of the high-speed flow region on the final product quality and the overall pressure drop in the furnace. The formula for calculating the weighted average degree of carbonization is: The weighted variance formula should be adjusted accordingly.

[0045] It should be noted that in the control of machine-made charcoal production, the intensity of airflow channelization is... It is an instantaneous physical field characteristic, reflecting the degree to which the gas takes shortcuts at the current moment, and belongs to the process variable; while the carbonization inhomogeneity index It is the result of a chemical reaction that accumulates over time, reflecting the consistency of product quality; it is an outcome variable. Simply relying on... Control may lead to an overreaction, and reliance on it alone is not advisable. Control exhibits a significant lag. A linear weighted fusion technique is used to construct a scalar indicator. This approach projects structural anisotropy (fluid behavior) and mass distribution (chemical results) onto the same control dimension. By adjusting the weighting coefficients, the priorities of preventing flow field deterioration and correcting mass deviations can be balanced in the control logic, thereby achieving a composite control effect that combines feedforward and feedback.

[0046] In a preferred embodiment, this is performed by the decision synthesis module of the central control unit (CCU). The system first receives the current airflow channelization intensity. and carbonization unevenness index The first step is to calculate the anisotropic deviation components of the system structure. The calculation formula is: In the formula, the number 1 represents the ideal isotropic state reference (i.e., ), The degree to which the current flow field structure deviates from the ideal uniform state is quantified and is a real number greater than or equal to 0. The second step involves the system reading preset weighting coefficients from the parameter configuration storage area. A first weighting coefficient is then set to characterize the degree of attention paid to the anisotropy of the structure. A second weighting coefficient of 2.0 is set to characterize the degree of attention paid to the carbonization quality distribution. It is 15.0. and The process of obtaining the value usually includes: The range of variation is between 1.0 and 3.0. Approximately 0 to 2.0; while As a standard deviation, its value is typically small, between 0 and 0.2. To ensure that the two indicators have comparable contributions (magnitude) to the final evaluation index, a certain value must be given. Larger gain coefficient ( The third step is for the system to perform multiplication and calculate the first product. and the second product The fourth step involves adding the two products together to obtain the flow field carbonization coupling evaluation index. The calculation formula is: The index It is a monotonically non-negative scalar. The larger its value, the more severe the system is in a bad state, such as a severe short circuit in the flow field, severe uneven carbonization, or both. The control system needs to take stronger intervention measures.

[0047] In some possible embodiments, to accommodate the control requirements of different stages of the carbonization process, the first weighting coefficient and the second weighting coefficient are designed to vary with the average degree of carbonization. A dynamic function that changes, rather than a fixed constant. Specifically, the formula for calculating the first weighting coefficient is as follows: The formula for calculating the second weighting coefficient is as follows: Preset parameters , , Specifically, in the early stages of carbonization ( (Approaching 0), the product has not yet formed. At this stage, the core control lies in preventing the formation of a channelized skeleton in the flow field, therefore, it is assigned... Higher weighting; while in the later stages of carbonization ( (Approaching 1) The flow field structure is already established, and the control core's steering ensures the consistency of the final product quality, thus giving... Higher weighting. This dynamic weighting mechanism makes the evaluation index... It can automatically guide the controller to focus on the most critical issues at different stages of the process.

[0048] In some possible embodiments, a tolerance normalization mechanism is introduced to determine weighting coefficients to address the differences in quality sensitivity among different specification mechanism bars. Specifically, the system does not directly set... and Instead, it sets a maximum permissible structural deviation threshold. (e.g., 1.5) and the maximum permissible carbonization inhomogeneity threshold (For example, 0.05). The system then automatically calculates the weighting coefficients: , At this point, the first and second products become normalized risk ratios relative to their respective tolerance limits. The final coupling evaluation index... The meaning of is transformed into a comprehensive risk level. When This means that the system exceeds the process safety boundary in at least one dimension or after combining two dimensions.

[0049] It should be noted that traditional PID controllers only regulate a single variable (such as temperature), neglecting the spatial uniformity constraints during the carbonization process. The proposed control law is an incremental multivariable controller, utilizing the carbonization degree deviation... As the primary driving force, it propels the system toward the predetermined endpoint (i.e., heating and ventilation); simultaneously, it utilizes the flow field carbonization coupling to evaluate the index deviation. As a negative feedback correction force or damping term, when the system experiences severe flow field short circuits or mass inhomogeneity, it automatically reduces heating power and wind pressure, sacrificing some production speed in exchange for the reorganization and equilibrium of the physical field.

[0050] In a preferred embodiment, the control is executed by a real-time control task running on a PLC or industrial PC, with a control cycle of The time is set to 10 seconds. First, the system reads the target value for carbonization production set in the process recipe through the human-machine interface (HMI). The target average degree of carbonization is then set. The value is 0.95 (meaning 95% of the raw material is converted into fixed carbon), and this value depends on the product grade requirements; a target flow field carbonization coupling evaluation index is set. The value is 0.05, a small positive number approaching zero, representing the allowable residual non-uniformity under ideal conditions. In each control cycle... The system receives the arithmetic mean of the current local carbonization degree. Evaluation index coupled with flow field carbonization Next, the system calculates the deviation: carbonization degree deviation. The calculation formula is This value is usually positive and gradually approaches 0 as carbonization progresses; coupling index deviation The calculation formula is Due to actual production Usually greater than the target value ,therefore Typically, these values ​​are negative, with larger absolute values ​​indicating worse system conditions. Simultaneously, the system maintains a time integral variable representing the degree of carbonization deviation. Updated using the discrete trapezoidal integral formula: Subsequently, the system executes continuous control law calculations to determine the control command for the next moment. This is for the total furnace heating power setpoint. Its calculation formula is In the formula, It is the power setting value being executed at the current moment (unit: kilowatts); It is the first proportional gain coefficient, with a value of 50.0 kW / unit deviation, used to provide basic heating power based on the current carbonization gap; It is the first integral gain coefficient, with a value of 0.5kW / (unit deviation·second), used to eliminate steady-state error and ensure that the final carbonization degree meets the standard; This is the first coupling correction gain coefficient, with a value of 20.0 kW / unit deviation. Because... Usually negative, positive value This means that as system non-uniformity increases, this term will generate a negative power increment, thereby suppressing the heating rate and preventing further localized overheating. (Regarding the total differential pressure setpoint for the furnace bed...) Its calculation formula is In this formula, This is the current differential pressure setting (unit: Pascal). It is the second proportional gain coefficient, with a value of 100.0 Pa / unit deviation; It is the second integral gain coefficient, with a value of 1.0 Pa / (unit deviation·second); This is the second coupling correction gain coefficient, with a value of 300.0 Pa / unit deviation. Similarly, positive values... This leads to a deterioration of the flow field. Get bigger When the pressure difference becomes negative, it can significantly reduce the set pressure difference, thereby slowing down the airflow velocity and suppressing the further development of channelization. Finally, the system will calculate... and After being processed for safe amplitude limiting, the signals are sent to the burner controller and the induced draft fan frequency converter through the analog output module to complete closed-loop control.

[0051] In some possible embodiments, to prevent system overshoot caused by integral saturation (IntegralWindup), conditional integral logic is introduced before calculating the control law. The integral value is calculated over time. When this happens, the system will first determine whether the current actuator is already in a saturated state (e.g., the heating power has reached its rated maximum or the pressure difference has reached the fan limit). If the actuator is already saturated and the current deviation... To move the system further toward saturation, we pause the accumulation of the integral term, that is, we let Conversely, normal integration occurs. Furthermore, for the coupling exponent deviation term, a dead zone is added: if... (If a preset dead zone threshold is set, such as 0.01), then a forced command will be executed. This design avoids unnecessary jitter in the control quantity due to minor computational noise when the system is in good condition, thus extending the service life of the actuator.

[0052] In some possible embodiments, a gain scheduling strategy is employed instead of a fixed gain coefficient to accommodate the nonlinear characteristics of the carbonization process. Specifically, a first coupling correction gain coefficient... Second Coupling Correction Gain Coefficient No longer a constant, but a degree of coupling severity It is an exponential function. The specific calculation formula is as follows: ,in Based on the gain, This is the sensitivity factor (value 2.0). Specifically, when... When the coefficient of friction is small (system uniformity), the coupling correction gain is low, and the control system mainly focuses on increasing the carbonization rate; while when... When the size increases (the system tends to become unstable), the correction gain increases exponentially, making the control law extremely sensitive to non-uniformity and able to apply a very strong braking effect, quickly pulling the system back to the safe flow field region.

[0053] like Figure 2 As shown, Figure 2 This diagram illustrates the dynamic evolution of two control indicators over time during the intelligent charcoal production process. The left subplot reflects the change in airflow channelization intensity, an indicator used to monitor whether non-uniform airflow short-circuiting occurs within the furnace. The right subplot reflects the evolution of the average degree of charcoalization and charcoal non-uniformity indicators, characterizing the overall progress and quality consistency of the charcoal reaction. Based on real-time feedback from these two indicators, the system dynamically adjusts the furnace heating power and the total bed pressure difference through a continuous control law to suppress channelization and ensure uniform charcoal production.

[0054] like Figure 3 As shown, Figure 3 Using digital twin model inversion technology, the physical field distribution within the furnace bed across all time and space is presented. The upper subplot depicts the permeability distribution within the bed, with variations in color intensity revealing the flow resistance distribution within different control volumes, helping to identify potential airflow deviation regions. The lower subplot shows the evolution of local carbonization distribution, clearly reflecting the spatial advancement of the carbonization front over time. These two sets of multidimensional data provide support for calculating the flow field carbonization coupling evaluation index, achieving a precise mapping from local field to global control.

[0055] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for intelligent regulation of the whole process of mechanism carbon production, characterized in that, include: A stacking structure flow guidance model based on the stacking method of fuel bars was constructed, and normalized thermal monitoring data sets were generated by collecting production process data. Using a digital twin model and constrained by the stacked structure flow guidance model, the permeability distribution state field inside the bed is inverted from the normalized thermal monitoring data set; The flow field analysis was performed on the air permeability distribution state field inside the bed to extract the airflow channelization intensity and the dominant airflow path direction; Calculate the carbonization non-uniformity index along the dominant airflow path direction, and fuse the carbonization non-uniformity index with the airflow channelization intensity to construct a flow field carbonization coupling evaluation index. Based on the deviation between the flow field carbonization coupling evaluation index and the target value, the total heating power of the furnace and the total pressure difference of the furnace bed are calculated by the continuous control law to regulate the carbonization process.

2. The intelligent mechanism carbon production whole-process regulation method according to claim 1, characterized in that, Construct a flow guidance model for a stacking structure based on the bar stacking method, including: Obtain the outer diameter, inner diameter, and length parameters of the mechanically manufactured fuel rods, as well as the stacking layer and row / column layout parameters of the fuel rods within the furnace. Determine the main direction of the furnace design airflow, which is determined by the stacking arrangement direction and the furnace geometry. Using the stacked bed as an equivalent anisotropic porous medium, construct a permeability tensor model in spatial location. This permeability tensor model is composed of an isotropic permeability baseline value and the permeability increment share along the main direction of the furnace design airflow. Based on the ratio of the inner diameter to the outer diameter of the mechanically manufactured fuel rods and the initial porosity of the bed, determine the isotropic permeability baseline value and the permeability increment share.

3. The intelligent mechanism carbon production whole-process regulation method according to claim 2, characterized in that, Collect production process data to generate a normalized thermal monitoring data set, including: Within a preset sampling period, multiple temperature data, flue gas component concentration data, and measured total pressure difference data at the furnace inlet and outlet are collected simultaneously. The current furnace total heating power set value and induced draft fan pressure difference set value are recorded simultaneously. The temperature data, concentration data, measured total pressure difference data, power set value, and pressure difference set value are combined to form an original measurement data set. The pre-statistical historical average value and historical standard deviation are called up, and linear standardization processing is performed on each data component in the original measurement data set, that is, the difference between the current measurement value and the historical average value is calculated and divided by the historical standard deviation to obtain the normalized thermal monitoring data set.

4. The intelligent mechanism carbon production whole-process regulation method according to claim 3, characterized in that, Using a digital twin model and constrained by the stacked structure flow guidance model, the permeability distribution state field inside the bed is inverted from the normalized thermal monitoring data set, including: A digital twin model describing the momentum and energy transfer between the gas and solid phases within the furnace is constructed. This digital twin model characterizes the mapping relationship between the permeability distribution state field inside the bed and the output thermal data under given input conditions. An inversion objective function is defined, which consists of the weighted sum of squared errors between the predicted data from the digital twin model and the normalized thermal monitoring data set, as well as the structural deviation constraint term between the permeability distribution state field inside the bed to be inverted and the stacked structure flow guide model. The permeability distribution state field inside the bed is updated and solved using a gradient descent iterative algorithm until the inversion objective function converges to its minimum value, and the current permeability distribution state field inside the bed is output.

5. The intelligent mechanism carbon production whole-process regulation method according to claim 4, characterized in that, The flow field analysis of the permeability distribution state field inside the bed is performed to extract the airflow channelization intensity and the dominant airflow path direction, including: Perform matrix eigenvalue decomposition on the permeability tensor data in the permeability distribution state field inside the bed to obtain the maximum principal eigenvalue, the minimum principal eigenvalue, and the principal eigenvector corresponding to the maximum principal eigenvalue; calculate the ratio of the maximum principal eigenvalue to the minimum principal eigenvalue, and use this ratio as the airflow channelization intensity; use the spatial direction indicated by the principal eigenvector corresponding to the maximum principal eigenvalue as the dominant airflow path direction; and calculate the spatial deflection angle of the dominant airflow path direction relative to the main airflow direction of the furnace design.

6. The intelligent mechanism carbon production whole-process regulation method according to claim 5, characterized in that, Calculating the carbonization non-uniformity index along the dominant airflow path includes: Within the digital twin model, several representative control volumes are selected along the dominant airflow path; temperature history data of each representative control volume over time are acquired; using a pyrolysis kinetic model including a reaction frequency factor and apparent activation energy, the current local carbonization degree of each representative control volume is calculated based on the temperature history data; the arithmetic mean of the local carbonization degrees of all representative control volumes is calculated, and the statistical value of the dispersion of the distribution of all local carbonization degrees relative to the arithmetic mean is taken as the carbonization non-uniformity index.

7. The intelligent mechanism carbon production whole-process regulation method according to claim 6, characterized in that, By fusing the carbonization non-uniformity index with the airflow channelization intensity, a flow field carbonization coupling evaluation index is constructed, including: The difference between the airflow channelization intensity and 1 is calculated as the structural anisotropy deviation component; a first weighting coefficient is set to characterize the degree of attention to structural anisotropy, and a second weighting coefficient is set to characterize the degree of attention to carbonization quality distribution; the first product of the structural anisotropy deviation component and the first weighting coefficient is calculated, and the second product of the carbonization non-uniformity index and the second weighting coefficient is calculated; the first product and the second product are added together to obtain the flow field carbonization coupling evaluation index used to characterize the severity of system coupling. 8.The intelligent mechanism carbon production whole-process regulation method according to claim 7, characterized in that, Based on the deviation between the flow field carbonization coupling evaluation index and the target value, the total furnace heating power and the total pressure difference of the furnace bed are calculated using a continuous control law to regulate the carbonization process, including: Set a target average degree of carbonization for the carbonization production process, and a target flow field carbonization coupling evaluation index approaching zero; calculate the target average degree of carbonization minus the arithmetic mean of local carbonization to obtain the degree of carbonization deviation; calculate the target flow field carbonization coupling evaluation index minus the flow field carbonization coupling evaluation index to obtain the coupling index deviation; multiply the current furnace total heating power by the product of the degree of carbonization deviation and the first proportional gain coefficient, the product of the time integral value of the degree of carbonization deviation and the first integral gain coefficient, and the product of the coupling index deviation and the first proportional gain coefficient. The total furnace heating power setting value for the next time moment is obtained by summing the products of the combined correction gain coefficients; the total furnace bed pressure difference setting value for the current time moment is obtained by summing the products of the carbonization degree deviation and the second proportional gain coefficient, the time integral value of the carbonization degree deviation and the second integral gain coefficient, and the coupling index deviation and the second coupling correction gain coefficient; the total furnace bed pressure difference setting value for the next time moment is obtained by summing the products of the total furnace heating power setting value for the next time moment and the total furnace bed pressure difference setting value for the next time moment as control commands to the actuator.

Images