Aluminum ash resource treatment method based on high-aluminum material

Abstract

The application relates to the technical field of aluminum ash resource digital twinning, and discloses an aluminum ash resource processing method based on high-aluminum material. The method constructs a digital twin of the processing process, sets a virtual probe array corresponding to key reaction sites of a real device in the digital twin, and continuously captures dynamic evolution information of aluminum ash processing. Based on the information, an evolution chain of physical fields and material flows is constructed, and then a processing path is deduced in reverse to identify key evolution nodes and constraint conditions. The nodes and conditions are converted into a control strategy set to drive a real processing system to perform closed-loop regulation and control. The method realizes deep perspective of the internal evolution mechanism of the aluminum ash processing process, realizes accurate regulation and control based on causal analysis, and improves the stability and resource recovery rate of the processing process.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology for aluminum ash resource utilization, specifically to a method for aluminum ash resource utilization based on high-alumina materials. Background Technology

[0002] Current methods for the resource utilization of high-alumina aluminum ash mainly rely on physical sensors monitoring macroscopic parameters and manual, experience-based adjustments. In terms of digitalization, existing technologies largely employ simulations based on fixed models or statistical analysis of production data to optimize process parameters. The limitation of these methods is that sensor data only reflects the state at the device boundaries or a limited number of points, failing to reveal the dynamic evolution details of aluminum ash within key regions of the reactor, including real-time changes in phase transformation and component distribution. This superficial information acquisition leads to a lack of understanding of the process mechanism and low process visibility.

[0003] Due to the lack of a detailed characterization of the coupling evolution chain between material flow and physical field during the processing, existing control strategies are mostly based on parameter adjustments or trial and error based on the "input-output" correlation. This approach is essentially empirical and result-feedback-based, unable to trace the root causes affecting resource utilization efficiency and product quality from within the process. When raw material composition or operating conditions fluctuate, it is difficult to quickly locate the root cause of the problem and intervene precisely, resulting in insufficient processing stability, bottlenecks in recovery rate, and a high risk of secondary pollution. This invention aims to achieve in-depth insight into the internal dynamics of the processing process by constructing a virtual probe array dynamically mapped to key physical sites; and to perform reverse deduction based on the constructed evolution chain to identify key evolution nodes and constraints, thereby driving precise closed-loop control based on process causal relationships. Summary of the Invention

[0004] The purpose of this invention is to provide a method for the resource utilization of aluminum ash based on high-alumina materials, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for the resource utilization of aluminum ash based on high-alumina materials, the method comprising:

[0006] In the resource-based processing system, a high-fidelity digital model is established for the processing of high-alumina aluminum ash. Multidimensional parameter data of high-alumina aluminum ash in the actual processing device are collected synchronously. The multidimensional parameter data is imported into the high-fidelity digital model for iterative calibration to generate a digital twin of the processing system.

[0007] A virtual probe array is set in the digital twin of the processing system. The position of the virtual probe array corresponds to the key reaction sites in the real processing device. The dynamic evolution information of high-alumina aluminum ash in the processing system is continuously captured by the virtual probe array.

[0008] Based on the dynamic evolution information captured by the virtual probe array, an evolution chain of physical fields and matter flow is constructed;

[0009] Based on the evolution chain of physical field and material flow, the processing path of high-alumina aluminum ash is reversed to identify the key evolution nodes and constraints in the processing path.

[0010] The identified key evolution nodes and constraints are transformed into a set of control strategies, which are then used to drive the processing system to perform closed-loop control on the high-alumina aluminum ash.

[0011] Preferably, a high-fidelity digital model is established for the processing of high-alumina aluminum ash, including:

[0012] Analysis of the phase composition and chemical reaction network of high-alumina aluminum ash;

[0013] Based on phase composition and chemical reaction networks, a core mechanism model library including thermodynamic equilibrium, reaction kinetics and multiphase transport processes is established.

[0014] Based on the geometric configuration and operational logic of the actual processing device, construct the device structure model and process logic model around the core mechanism model library;

[0015] The core mechanism model library, device structure model and process logic model are coupled to form the high-fidelity digital model.

[0016] Preferably, multi-dimensional parameter data of high-alumina aluminum ash from the actual processing device are collected synchronously, and the multi-dimensional parameter data is imported into the high-fidelity digital model for iterative calibration, including:

[0017] Multiple types of sensors are deployed at different depths and key reaction sites in the actual processing device to obtain real-time data streams of temperature, pressure, gas composition and material morphology.

[0018] Timestamp alignment and outlier cleaning are performed on real-time data streams to form structured multidimensional parameter data;

[0019] Structured multidimensional parametric data are used as boundary conditions and observations, and then input into a high-fidelity digital model.

[0020] Drive a high-fidelity digital model to perform forward simulation, compare the simulation output with the corresponding real observations, and generate a residual sequence.

[0021] Based on the residual sequence, the dynamic parameters and transfer coefficients in the high-fidelity digital model are adjusted in reverse using a parameter estimation algorithm until the degree of agreement between the simulation output and the actual observations meets the preset convergence criterion, thus completing the generation of the digital twin.

[0022] Preferably, based on the dynamic evolution information captured by the virtual probe array, an evolutionary chain of physical fields and matter flow is constructed, including:

[0023] Spatiotemporal distribution data of temperature gradient field, concentration gradient field and reaction rate field are extracted from the dynamic evolution information captured by the virtual probe array.

[0024] The coupling relationship and hysteresis effect between the spatiotemporal distribution data of temperature gradient field, concentration gradient field and reaction rate field were analyzed.

[0025] Based on the coupling relationship and hysteresis effect, a physical field state transition sequence with time as the axis is established;

[0026] In the physical field state transition sequence, mark the key events that lead to phase transformation or reaction path bifurcation;

[0027] By using key events as anchor points and linking them to changes in the composition of the material flow, an evolutionary chain is formed in which the evolution of the physical field and the changes in the composition of the material flow are interconnected.

[0028] Preferably, based on the evolution chain of physical fields and material flow, the processing path of high-alumina aluminum ash is deduced in reverse, including:

[0029] Starting from the final target phase state of the evolutionary chain, trace back to its immediate physical field and the preceding state of matter flow;

[0030] Analyze the transition process from the predecessor state to the current state to identify the main control variables and reaction conditions driving the transition process;

[0031] Continue tracing back to the predecessor state of the predecessor state, repeating the above identification process, until the initial material state is reached;

[0032] Along the complete backtracking path, the difficulty and energy consumption of all transformation processes are marked, and the transformation process nodes with relatively low transformation efficiency are selected and identified as key evolution nodes.

[0033] We analyze the internal factors and external conditions that limit the transformation efficiency of key evolution nodes and summarize them as constraints.

[0034] Preferably, the identified key evolution nodes and constraints are transformed into a set of control strategies, including:

[0035] For each key evolution node, multiple virtual intervention schemes are preset in the digital twin of the processing system;

[0036] Simulate and run each virtual intervention scheme in a digital twin to obtain the improvement in transformation efficiency of each intervention scheme on key evolution nodes under corresponding constraints;

[0037] All virtual intervention schemes were evaluated and ranked based on both the magnitude of improvement and energy consumption cost.

[0038] Select virtual intervention schemes whose evaluation results meet the preset thresholds, and compile their specific operation parameters, execution sequence and action points into instruction units that can be recognized by the control system of the real treatment device.

[0039] The instruction units corresponding to all key evolution nodes are combined and serialized to generate the control strategy set.

[0040] Preferably, a set of control strategies is used to drive the processing system to perform closed-loop control on the high-alumina aluminum ash, including:

[0041] The instruction units in the control strategy set are sent to the underlying execution mechanism of the actual processing device according to the arrangement sequence;

[0042] The underlying execution mechanism executes instructions to change the operational parameters of the corresponding key evolution nodes;

[0043] During command execution, response data of high-alumina aluminum ash is continuously collected through sensors;

[0044] The response data is fed back to the digital twin of the processing system in real time, triggering the digital twin to perform a rapid simulation;

[0045] Compare the results of rapid simulation with the expected targets of the control strategy set to generate a deviation signal of the control effect;

[0046] Based on the deviation signal of the control effect, the parameters of the subsequent instruction units to be executed are dynamically fine-tuned to form a closed-loop control.

[0047] Preferably, the method further includes a step of determining the final state of the high-alumina aluminum ash treatment:

[0048] During the closed-loop control process, the dynamic evolution information captured by the virtual probe array is continuously monitored;

[0049] When the signal strength representing the characteristics of the target product in the dynamic evolution information reaches saturation and stabilizes for more than a preset time, and the signal strength representing the characteristics of residual impurities or by-products drops below the threshold;

[0050] Virtual sampling and phase analysis are performed on the overall state of the current high-alumina aluminum ash in the digital twin of the processing system.

[0051] If the results of virtual sampling and phase analysis confirm that the preset final state composition and structure requirements have been met, a processing completion command will be generated.

[0052] Preferably, after generating the processing completion instruction, a virtual mapping step of the resource-based products is also included:

[0053] Based on the final state distribution of high-alumina aluminum ash in the digital twin after processing, a three-dimensional composition and structure map of the resource-based product is generated.

[0054] Based on the three-dimensional composition and structure map, a pre-set product database is matched for the resource products, and the performance indicators of the products as recycled materials in different application scenarios are predicted.

[0055] By binding performance indicators, three-dimensional composition and structural maps with corresponding processing data, a full life-cycle digital archive of the high-alumina material aluminum ash resource utilization treatment is generated.

[0056] Preferably, the step of dynamically fine-tuning the parameters of the subsequent instruction unit to be executed based on the deviation signal of the control effect includes:

[0057] Analyze the deviation signal of the control effect to determine whether the main source of the deviation is the deviation of thermodynamic conditions, kinetic conditions, or flow conditions.

[0058] Based on the type of deviation source, the corresponding parameter correction logic is called from the preset compensation rule library;

[0059] The parameter correction logic is used to calculate the required adjustment amount and direction of the current instruction unit parameters;

[0060] Without violating the safety and process boundaries of the processing device, the temperature setting, pressure setting, material flow rate or reaction time parameters in the subsequent instruction units to be executed are superimposed and calculated.

[0061] The new parameters calculated by superposition are updated to the corresponding instruction unit.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] By setting up a virtual probe array dynamically mapped to key physical sites within a digital twin, multidimensional information such as the microstructure, phase composition, temperature gradient, and mass transfer of aluminum ash at corresponding physical locations can be captured directionally and continuously in virtual space. This overcomes the limitations of physical sensors, which can only measure macroscopic parameters at boundaries or limited points, enabling direct, multidimensional, and in-situ observation of invisible processes in key areas inside the reactor. It obtains detailed process data that is unavailable through traditional methods, providing a crucial information dimension for a deeper understanding of the instantaneous reaction mechanism of aluminum ash resource utilization.

[0064] Based on the dynamic information captured by virtual probes, an evolutionary chain of physical fields and material flows is constructed, and reverse deduction is performed. Starting from the final or intermediate state of the process, the evolutionary links and conditions leading to that state are traced back to the previous level. It can accurately locate the decisive steps that restrict reaction rates and affect product purity, as well as the sensitive ranges of process parameters in complex process networks. It can identify bottlenecks and potential risk points that are easily overlooked by traditional forward optimization methods, elevating process cognition from the correlation level to the causal level.

[0065] By transforming the key nodes and constraints identified through reverse engineering into a set of control strategies driving the real system, the generation of control commands directly stems from the analysis of the inherent causal chain of the process. This achieves a paradigm shift in control from "result feedback - experience adjustment" to "process insight - causal intervention." The control actions have clear process orientation and foresight, enabling early and precise intervention against process bottlenecks. This improves the stability of the high-alumina aluminum ash processing process, resource recovery efficiency, and product consistency, while suppressing unexpected side reactions. Attached Figure Description

[0066] Figure 1 This is a schematic diagram illustrating the working principle of a method for resource recovery of aluminum ash based on high-alumina materials according to the present invention.

[0067] Figure 2 Flowchart for building a high-fidelity digital model;

[0068] Figure 3 A flowchart for constructing the evolution chain of physical fields and matter flow;

[0069] Figure 4 This is a comparative analysis chart of intervention programs;

[0070] Figure 5 This is a trend chart of the closed-loop control process. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Please see Figure 1This invention provides a method for the resource-based treatment of aluminum ash from high-alumina materials. The method includes: establishing a high-fidelity digital model for the treatment process of high-alumina aluminum ash in a resource-based treatment system, which operates synchronously with the actual treatment device. During the operation of the actual treatment device, multi-dimensional parameter data of the high-alumina aluminum ash during the treatment process are collected synchronously. This data includes, but is not limited to, real-time changes in temperature, pressure, material composition, and morphology. The collected multi-dimensional parameter data is continuously imported into the established high-fidelity digital model. Iterative calibration is performed through comparison calculations between the data and the model, so that the model output continuously approximates the actual process, ultimately generating a digital twin that is highly consistent with the actual treatment system. In this digital twin, a set of virtual probe arrays is set. The spatial position of each virtual probe in the array corresponds one-to-one with a preset key reaction site in the actual treatment device. Through this virtual probe array, the dynamic evolution information of the simulated treatment of high-alumina aluminum ash in the digital twin can be continuously captured, such as the spatiotemporal changes in local temperature, component concentration, and reaction rate. Based on the dynamic evolution information captured by the virtual probe array, the correlation between the physical field and the material flow is constructed, forming a chain-like sequence describing the changes in material composition and phase driven by the physical field. According to the constructed evolution chain of the physical field and material flow, starting from the target state of the processing, the processing path of high-alumina aluminum ash is deduced backward along the time axis. During this backward deduction, key evolution nodes that decisively influence the final processing result and the constraints limiting the efficiency of these nodes are identified. All identified key evolution nodes and their corresponding constraints are transformed into a series of specific, executable operational instructions, forming a set of control strategies. Finally, this set of control strategies is used to drive the real processing system to perform real-time, adaptive, closed-loop control of the high-alumina aluminum ash processing process.

[0073] Example 1: See Figure 2To establish a high-fidelity digital model for the processing of high-alumina aluminum ash, the initial phase composition of the high-alumina aluminum ash was first analyzed, and the chemical reaction network occurring during processing was clarified. Based on the analyzed phase composition and chemical reaction network, a core mechanism model library was established, including modules for thermodynamic equilibrium calculation, reaction kinetics calculation, and multiphase transport process calculation. According to the actual geometric configuration and operational control logic of the real processing device, a device structure model describing the physical structure of the device and a process logic model describing the material flow and operation sequence were constructed around the core mechanism model library. The core mechanism model library, device structure model, and process logic model were coupled and integrated to form a high-fidelity digital model. Multidimensional parameter data of high-alumina aluminum ash in the real processing device were simultaneously collected and imported into the high-fidelity digital model for iterative calibration. Specifically, temperature sensors, pressure sensors, gas analyzers, and image sensors were deployed at different depths of the reaction chamber and at preset key reaction sites in the real processing device to obtain real-time data streams of temperature, pressure, gas composition, and material morphology. Real-time data streams from different sensors are timestamped and outliers are removed to create structured multidimensional parametric data. This structured multidimensional parametric data is then used as boundary conditions and observations, input into a high-fidelity digital model. The high-fidelity digital model is driven to perform forward simulation calculations. The simulated data output is compared with the corresponding real observations to generate a residual sequence. Based on the residual sequence, a parameter estimation algorithm is used to reverse-engineer the reaction kinetic parameters and mass heat transfer coefficients in the high-fidelity digital model until the simulation output and the real observations meet a preset convergence criterion, thus completing the generation of the digital twin.

[0074] In practical implementation, a high-fidelity digital model is established for the treatment process of high-alumina aluminum ash. The phase composition of high-alumina aluminum ash includes metallic aluminum, alumina, aluminum nitride, and salt components. The treatment process takes place in a rotary kiln to recover metallic aluminum and stabilize aluminum nitride. Analyzing the phase composition and chemical reaction network of high-alumina aluminum ash requires determining the melting of aluminum, the hydrolysis of aluminum nitride, and the phase transformation of alumina. Based on the phase composition and chemical reaction network, a core mechanism model library is established, including thermodynamic equilibrium, reaction kinetics, and multiphase transport processes. The thermodynamic equilibrium module uses the minimum Gibbs free energy principle to calculate the equilibrium composition of the system. The reaction kinetics module describes the rate equations of each solid-gas phase reaction. The multiphase transport process module covers material flow, heat transfer, and gas diffusion within the kiln. Based on the geometric configuration and operating logic of a real rotary kiln, a device structure model and a process logic model are constructed around the core mechanism model library. The device structure model defines the length, diameter, inclination angle, and lining material of the rotary kiln, while the process logic model defines the aluminum ash feeding, heating, rotation, and discharging sequence. The core mechanism model library, device structure model and process logic model are coupled to form a high-fidelity digital model. The coupling is achieved through data interface to realize real-time data exchange between modules.

[0075] In some embodiments, multidimensional parameter data of high-alumina material ash from a real rotary kiln are synchronously acquired and imported into a high-fidelity digital model for iterative calibration. Thermocouples, pressure sensors, mass spectrometers, and industrial cameras are deployed at different depths and key reaction sites in the kiln head, middle, and tail of the real rotary kiln to acquire real-time data streams of temperature, pressure, gas composition, and material morphology. The real-time data streams are timestamped and outlier cleaned. Timestamping is based on a unified clock signal to synchronize all sensor data, and outlier cleanup uses sliding window filtering to remove noise points, forming structured multidimensional parameter data. This structured multidimensional parameter data is input into the high-fidelity digital model as boundary conditions and observations. Boundary conditions include the kiln head feed rate and kiln tail exhaust pressure, while observations include the kiln temperature distribution curve and gas concentration changes. The high-fidelity digital model is driven to perform forward simulation, and the simulation output is compared with the corresponding real observations to generate a residual sequence. The residual sequence includes temperature residuals, pressure residuals, and component concentration residuals. Based on the residual sequence, a parameter estimation algorithm is used to reverse-adjust the kinetic parameters and transfer coefficients in the high-fidelity digital model. The kinetic parameters refer to the reaction rate constant, and the transfer coefficients refer to the thermal conductivity coefficient and the mass diffusion coefficient, until the degree of agreement between the simulation output and the actual observations meets the preset convergence criterion. The convergence criterion is defined as the root mean square error of all residuals being less than a set threshold, thus completing the generation of the digital twin.

[0076] It is understandable that parameter adjustment during iterative calibration is based on optimization algorithms. In specific implementation, the objective function constructed by minimizing the residual sequence is adopted using the gradient descent method. The objective function has the following form:

[0077]

[0078] in: Represents the objective function value. This represents the vector of dynamic parameters and transfer coefficients to be adjusted. Indicates the number of observation points. Indicates the index of the observation point. Indicates the first The weighting coefficients for each observation point Indicates the high-fidelity digital model in the first... Simulation output values ​​for each observation point Indicates the first The actual observed values ​​at each observation point. The gradient descent method iteratively updates the parameter vector. Until the objective function It reaches the minimum value.

[0079] In some embodiments, the data comparison is reflected in the quantitative difference between the simulation output and the actual observed values. In a specific implementation, for the temperature observation point in the kiln, the actual observed value is 1150°C under steady state, while the initial simulation output of the high-fidelity digital model is 1100°C, resulting in a temperature residual of 50°C. By adjusting the thermal conductivity coefficient and reaction activation energy parameters, after multiple iterations, the simulation output approaches 1152°C, and the temperature residual is reduced to 2°C, satisfying the convergence criterion. For gas composition observation, the actual mass spectrometer detected a peak nitrogen concentration of 12%, while the initial simulation output nitrogen concentration was 8%, resulting in a concentration residual of 4%. By adjusting the kinetic parameters of the aluminum nitride hydrolysis reaction, the simulated nitrogen concentration finally reached 11.8%, and the concentration residual was reduced to 0.2%.

[0080] Optionally, the arrangement of multiple types of sensors needs to cover key reaction sites. In specific implementations, key reaction sites include the aluminum melting zone and the aluminum nitride reaction zone. Thermocouples are embedded at different depths in the kiln wall lining to measure the radial temperature gradient, and industrial cameras capture images of material morphology changes through the kiln's viewing window. Timestamp alignment uses a network time protocol to synchronize the data acquisition times of all sensors. Outlier cleaning uses box plots to identify and remove outlier data points that deviate from the interquartile range. Structured multidimensional parameter data is stored in the form of a time-space matrix, where rows correspond to time series and columns correspond to the measurement variables of different sensors.

[0081] It is understandable that the coupling mechanism of the high-fidelity digital model is implemented through a software platform. In practice, the core mechanism model library is built in chemical process simulation software, the device structure model is built in 3D modeling software, and the process logic model is built in process simulation software. Data coupling among the three is achieved through an application programming interface (API). During forward simulation, the process logic model calls the geometric parameters of the device structure model and the computational kernel of the core mechanism model library to solve for the simulation output values ​​of multidimensional parameter data. The comparison of residual sequences is performed on a unified data processing platform, which calculates the absolute and relative errors between the simulation output values ​​and the actual observed values ​​at each time step.

[0082] Optionally, the parameter estimation algorithm can employ sequential quadratic programming. In practice, a constrained nonlinear optimization problem is constructed based on the residual sequence. The constraints include the physical range of the kinetic parameters and the positive definiteness of the transfer coefficients. The sequential quadratic programming method iteratively solves the optimization problem to update the parameters in the high-fidelity digital model. The preset thresholds for the convergence criteria are set according to process requirements: the root mean square error threshold for temperature residuals is set to 5°C, and the root mean square error threshold for concentration residuals is set to 0.5%. Iteration stops when the root mean square error of all residual sequences is simultaneously below the threshold. The generation of the digital twin is marked by the solidification of the high-fidelity digital model parameters and the output of a calibration report.

[0083] Example 2: See Figure 3Based on the dynamic evolution information captured by a virtual probe array, an evolutionary chain of physical fields and material flows is constructed. Spatiotemporal distribution data of temperature gradient field, concentration gradient field, and reaction rate field are extracted from the dynamic evolution information captured by the virtual probe array. The coupling relationship and hysteresis effect among the spatiotemporal distribution data of temperature gradient field, concentration gradient field, and reaction rate field are analyzed. Based on the analyzed coupling relationship and hysteresis effect, a physical field state transition sequence with time as the axis is established. Key events leading to phase transformation or bifurcation of chemical reaction paths are marked in the physical field state transition sequence. Using these key events as anchor points, the changes in material flow composition before and after their occurrence are correlated, forming an evolutionary chain in which the evolution of physical fields and changes in material flow composition are interconnected. The processing path of high-alumina aluminum ash is reverse-engineered based on the evolutionary chain of physical fields and material flows. Starting from the final target phase state described by the evolutionary chain, the preceding physical field and material flow states that directly generated this state are traced back. The transformation process from the preceding state to the current target state is analyzed, and the main control variables and reaction conditions driving this transformation process are identified. Continue tracing back to the precursor state of the previous state, repeating the above identification process until the initial high-alumina aluminum ash material state is reached. Along the complete tracing path, the difficulty and energy consumption of all transformation processes are marked, and the transformation process nodes with relatively low conversion efficiency are selected as critical evolution nodes. The internal factors and external conditions limiting the conversion efficiency of these critical evolution nodes are analyzed and summarized as constraints.

[0084] In practical implementation, an evolutionary chain of physical fields and material flows is constructed based on the dynamic evolution information captured by a virtual probe array. The virtual probe array is set in the rotary kiln model of the digital twin, and the positions of the virtual probes correspond to key reaction sites in the aluminum melting zone, aluminum nitride reaction zone, and alumina phase transformation zone of the real rotary kiln. Spatiotemporal distribution data of temperature gradient field, concentration gradient field, and reaction rate field are extracted from the dynamic evolution information captured by the virtual probe array. The temperature gradient field data includes the time-varying sequence of temperature distribution cloud maps along the kiln's axial and radial directions. The concentration gradient field data includes the concentration distribution cloud maps of various gas components such as carbon monoxide, nitrogen, and water vapor. The reaction rate field data is calculated from the reaction kinetics model based on local temperature and concentration. In practical implementation, the generation of reaction rate field data relies on local temperature and concentration data captured by a virtual probe array within the rotary kiln model of the digital twin. This data is input to the reaction kinetics module in the core mechanism model library. Based on predefined rate equations for each solid-phase and gas-phase reaction, this module integrates local temperature values, reactant concentration values, and reaction equilibrium concentration parameters to calculate the instantaneous reaction rate value corresponding to each virtual probe location. The spatiotemporal distribution of the reaction rate field data is formed by mapping all these local reaction rate values ​​onto the three-dimensional grid nodes of the digital twin, thereby constructing a dynamic cloud map reflecting the changes in reaction intensity in different regions within the kiln. The coupling relationship and hysteresis effect between the spatiotemporal distribution data of the temperature gradient field, concentration gradient field, and reaction rate field are analyzed. The coupling relationship is manifested in the spatial overlap between the peak region of the reaction rate field and the high-temperature region of the temperature gradient field and the high-concentration region of the reactants in the concentration gradient field. The hysteresis effect is manifested in the time delay between changes in the temperature field and changes in the concentration field. A physical field state transition sequence with time as the axis is established based on coupling relationships and hysteresis effects. Each time slice in the sequence records the complete spatial distribution of the temperature gradient field, concentration gradient field, and reaction rate field. Key events leading to phase transitions or reaction path bifurcation are marked in the physical field state transition sequence. These key events include the melting initiation of metallic aluminum and the explosive acceleration of the aluminum nitride hydrolysis reaction. The corresponding changes in mass flow composition are linked to these key events as anchor points. These changes are reflected through local compositional change data obtained from virtual probes and phase analysis data, forming an evolutionary chain where physical field evolution and mass flow composition changes are interconnected. This evolutionary chain is stored in a directed graph structure, where nodes represent combinations of physical field states and mass flow states, and edges represent transitions between states.

[0085] In some embodiments, the processing path of high-alumina aluminum ash is reverse-engineered based on the evolution chain of physical fields and material flows. Starting from the final target phase state of the evolution chain, the final target phase state is characterized by the aggregation of metallic aluminum, complete reaction of aluminum nitride, and the existence of alumina as a stable α phase. The preceding physical field and material flow states directly preceding the final target phase state are traced back to the state combination where metallic aluminum is in the droplet polymerization stage, the aluminum nitride reaction rate reaches its peak, and alumina begins to transform into the α phase. The transformation process from the preceding state to the final target phase state is analyzed to identify the main control variables and reaction conditions driving the transformation process. The main control variables include the rotary kiln speed and heating power, while the reaction conditions include maintaining the local temperature above 1150°C and the exposed surface area of ​​the aluminum nitride particles. The preceding state of the preceding state is traced back further, and the above identification process is repeated until the initial material state is reached, where solid high-alumina aluminum ash powder enters the rotary kiln. The difficulty and energy consumption of all transformation processes are marked along the complete backtracking path. The difficulty is quantified by the activation energy or driving force required for the transformation, and the energy consumption is calculated by integrating local heat flux and reaction enthalpy change. Transformation process nodes with relatively low conversion efficiency are screened out, and the decomposition initiation node of aluminum nitride particles from being coated with salt to being fully exposed is identified as the critical evolution node. The internal factors and external conditions limiting the conversion efficiency of the critical evolution node are analyzed. The internal factor is the obstruction of mass transfer by the salt coating layer, and the external condition is insufficient local temperature in the kiln. These are summarized as constraints.

[0086] It is understandable that the construction of the evolutionary chain relies on a model that quantitatively describes the interactions between physical fields. In specific implementation, to express the synergistic effect of the temperature gradient field and the concentration gradient field on the reaction rate field, a coupling influence factor function is introduced, with the following form:

[0087]

[0088] in: This represents the coupling effect factor, used to correct local reaction rates. Indicates the location of the virtual probe. and time The captured temperature value, Indicates the location of the virtual probe. and time The concentration value of the captured reactant A, This represents the equilibrium concentration of reactant A at the current temperature. This indicates the initial reference concentration of reactant A. The apparent activation energy of a reaction. Represents the universal gas constant. Indicates the reaction order. The hysteresis effect of the temperature gradient field and the concentration gradient field is compared. and The change in phase difference is used to determine, where It is the lag time.

[0089] In some embodiments, data comparison is reflected in the correlation between the physical field state and actual observations. In a specific implementation, the temperature gradient field data captured by the virtual probe in the aluminum nitride reaction zone in the digital twin shows that the local temperature rises from 900°C to 1180°C within 10 minutes. Simultaneously, the concentration gradient field data shows that the nitrogen concentration begins to rise significantly about 2 minutes after the temperature reaches 1150°C, demonstrating a lag effect where temperature change precedes concentration change. The reaction rate field is calculated based on the coupling influence factor function, and peaks occur when the temperature reaches 1150°C and the aluminum nitride surface concentration reaches a threshold. When identifying key evolution nodes in the reverse simulation, the backtracking path shows that the accumulation efficiency of metallic aluminum is high, while the efficiency of the initial stage of the aluminum nitride reaction is low. Conversion efficiency calculations indicate that the rate of decrease in aluminum nitride content in the material flow during the initial stage is only 30% of that during the peak period. Therefore, the starting node of the aluminum nitride reaction is marked as a key evolution node.

[0090] Optionally, the establishment of the physical field state transition sequence can employ a time discretization method. In practice, snapshots of the digital twin's simulation output are taken at fixed time intervals. Each snapshot contains the values ​​of the temperature gradient field, concentration gradient field, and reaction rate field on the three-dimensional mesh nodes. Key events are automatically detected by setting thresholds. When the value of the reaction rate field in a certain spatial region exceeds the set threshold and continues to increase, the event is marked as a reaction burst. In the directed graph structure of the evolutionary chain, node attributes include physical field data pointers and matter flow component vectors, while edge attributes include time intervals and predecessor-successor relationships.

[0091] It can be understood that the reverse deduction process is implemented as a backtracking algorithm in the computing system. In specific implementation, starting from the terminal node of the evolution chain, the algorithm traverses the directed graph to find all incoming edges pointing to the current node. The source node corresponding to the incoming edge is the direct predecessor state. The algorithm analyzes the directed edges from the predecessor state to the current state and extracts the control variables and condition parameters stored on the edges. The algorithm is executed recursively until the root node representing the initial material state is accessed. The difficulty of each transition process on the complete backtracking path is obtained by calculating the energy barrier value of the state transition, and the energy consumption is calculated by integrating the total heat input to the system during the transition process. The constraints are summarized as a description of the allowable operating range of the control variables and a qualitative description of the intrinsic properties of the material.

[0092] Optionally, the determination of relatively low conversion efficiency is based on comparative analysis. In specific implementation, the backtracking path includes five main transformation process nodes. The target product generation or reactant consumption per unit time at each transformation process node is calculated. The comparison reveals that the unit consumption at the aluminum nitride reaction initiation node is the lowest, with a value only 30% of the unit consumption in the subsequent high-efficiency reaction stages. Therefore, this node is selected as the key evolution node. Internal factors and external conditions are obtained by analyzing the microscopic simulation data of the digital twin at this node. The microscopic simulation data shows that the salt melt covers the surface of the aluminum nitride particles, and at the same time, the virtual probe temperature reading in the corresponding area of ​​the kiln at this node is lower than the full flow temperature of the salt melting point.

[0093] Example 3: The identified key evolution nodes and constraints are transformed into a set of control strategies. Multiple virtual intervention schemes are pre-set for each key evolution node in the digital twin of the processing system. Each virtual intervention scheme is simulated in the digital twin to obtain the improvement in the transformation efficiency of the key evolution node under the corresponding constraints. All virtual intervention schemes are evaluated and ranked using both the improvement and the energy consumption cost estimated by the simulation as dual indicators. Virtual intervention schemes whose evaluation results meet a preset threshold are selected, and the specific operating parameters, execution sequence, and point of action information in these schemes are compiled into instruction units that can be recognized by the control system of the real processing device. The instruction units corresponding to all key evolution nodes are combined and serialized according to the processing flow to generate a set of control strategies.

[0094] In practical implementation, the identified key evolution nodes and constraints are transformed into a set of control strategies. The key evolution node is the aluminum nitride reaction initiation node, and the constraints include the obstruction of mass transfer by the salt coating layer and insufficient local temperature within the kiln. Multiple virtual intervention schemes are preset for the aluminum nitride reaction initiation node in the digital twin of the rotary kiln processing system. These virtual intervention schemes include adjusting the rotary kiln speed in the corresponding region of the node, changing the power distribution in the corresponding heating zone, introducing additives to break down the salt coating layer, and improving material pretreatment to increase particle size. Each virtual intervention scheme is simulated in the digital twin using a calibrated high-fidelity digital model for dynamic simulation. The improvement in conversion efficiency of the aluminum nitride reaction initiation node under the constraints of salt coating and insufficient local temperature is obtained, and the improvement is quantified by comparing the percentage change in the rate of decrease in aluminum nitride content before and after implementing the intervention scheme.

[0095] In some embodiments, all virtual intervention schemes are evaluated and ranked using both the magnitude of improvement and energy cost as indicators. Energy cost is calculated from energy consumption data obtained through digital twin simulations, including additional electrical and fuel consumption. The evaluation and ranking are performed using a multi-objective evaluation function, which has the following form:

[0096]

[0097] in: This represents the overall assessment score of the intervention program. The weighting coefficients represent the magnitude of the increase. Indicates the first The rate of decrease in aluminum nitride content after the implementation of a virtual intervention program, The baseline rate of decline represents the rate at which no virtual intervention was implemented. The weighting coefficient representing energy consumption costs. Indicates the implementation of the first The increased energy consumption of such virtual intervention schemes This represents the baseline energy consumption. The energy consumption for each virtual intervention protocol is calculated. The values ​​are then sorted, and virtual intervention programs whose evaluation results meet a preset threshold are selected. The preset threshold is set as follows: Furthermore, the improvement exceeded 20%. Virtual intervention schemes that met the preset thresholds were selected. The specific operational parameters, execution timing, and application points of these schemes were compiled into instruction units recognizable by the real rotary kiln control system. The operational parameters included increasing the rotary kiln speed from 1.5 revolutions per minute to 2.2 revolutions per minute; the execution timing was to begin 30 seconds before the material entered the reaction zone; and the application points were the rotary kiln's drive motor and heating belt control module. All instruction units corresponding to key evolution nodes were combined and serialized to generate a control strategy set. This set was stored in an Extensible Markup Language (EXPLAIN) file format, containing a list of instruction units, trigger conditions, and execution logic.

[0098] It is understandable that the pre-set virtual intervention schemes are based on the mechanistic analysis of the constraints. In specific implementation, for the constraint of the salt coating layer hindering mass transfer, the virtual intervention scheme design focuses on destroying the coating layer or increasing the reaction interface. The additive introduction scheme simulates the injection of a small amount of sodium carbonate to reduce the viscosity of the salt melt, and the improved material pretreatment scheme simulates adding a mechanical grinding process before feeding. For the constraint of insufficient local temperature in the kiln, the virtual intervention scheme design focuses on enhancing local heat transfer. The rotary kiln speed adjustment scheme aims to change the residence time of the material in the high-temperature zone, and the scheme to change the power distribution in the heating zone simulates increasing the heating power of the area corresponding to the reaction initiation node by 15%. When simulating each virtual intervention scheme in the digital twin, the simulation boundary conditions are set to be consistent with the real constraints, the salt coating layer thickness distribution and thermophysical properties are set according to the material analysis data, and the initial distribution of the local temperature field is synchronized with the real-time monitoring data.

[0099] In some embodiments, data comparison is reflected in the evaluation results of different virtual intervention schemes. In specific implementation, for the aluminum nitride reaction initiation node, simulation operation shows that adjusting the rotary kiln speed increases energy consumption by 25%, while increasing energy cost by 5%; changing the power distribution in the heating zone increases energy consumption by 35%, while increasing energy cost by 18%; introducing additives increases energy consumption by 40%, while increasing energy cost by 8%; and improving material pretreatment increases energy consumption by 30%, while increasing energy cost by 12%. Substituting into the evaluation function, the adjustment of the rotary kiln speed... The value is 0.65, indicating a change in the power distribution scheme of the heating zone. The value is 0.48, indicating the introduction of an additive scheme. The value is 0.85, indicating an improvement in the material pretreatment scheme. The value is 0.55. Based on a preset threshold, the power distribution scheme of the heating zone is changed. The value is below 0.5, therefore the condition is not met, and the other three schemes are selected. When compiling the instruction unit, the operation parameter corresponding to the rotary kiln speed adjustment scheme is to linearly increase the speed setpoint from 1.5 rpm to 2.2 rpm, and the point of action is the kiln drive frequency converter; the operation parameter corresponding to the additive introduction scheme is that the sodium carbonate powder feed rate is one percent of the total material mass, and the point of action is the powder injector located at the kiln head; the operation parameter of the improved material pretreatment scheme has been executed before feeding, and its instruction unit mainly contains status confirmation instructions.

[0100] Optionally, the evaluation and ranking process can incorporate more indicators. In practice, in addition to the improvement and energy cost, evaluation indicators can also include the complexity of the implementation and its impact on subsequent process stages. A comprehensive score is obtained by weighting and summing different indicator weights. The serialization of instruction units needs to consider process logic and time dependencies. In practice, the instruction unit for improving material pretreatment must be executed first. Instructions for adjusting rotary kiln speed and introducing additives overlap in time; therefore, clear start and end timestamps and interlocking logic must be defined to prevent operational conflicts. The control strategy set file includes a version number, applicable material batch identifier, and digital signature information.

[0101] It is understandable that compiling instruction units into forms recognizable by the control system of a real processing device requires adherence to communication protocols. In practice, instruction units are compiled into data structures conforming to the OPCUA standard, containing the target device address, instruction code, parameter values, and execution time. The generation of the control strategy set is accomplished by a dedicated strategy compiler software module. This module reads the evaluation and ranking results and scheme details, generates instruction sequences according to the process sequence template, and automatically checks the logical consistency between instructions.

[0102] See Figure 4 This chart presents a comparative analysis of four different intervention schemes in terms of conversion efficiency improvement and energy cost. The left bar chart shows the improvement in conversion efficiency at the aluminum nitride reaction initiation node for each scheme, while the right bar chart shows the increase in energy cost resulting from implementing each scheme. The comprehensive evaluation score Φ above each scheme reflects its overall performance, and the red dashed line indicates the preset evaluation threshold. The charts clearly show the trade-offs between efficiency improvement and energy control for different schemes, providing intuitive data support for selecting the optimal control strategy.

[0103] Example 4: A control strategy set is used to drive a processing system to perform closed-loop control on high-alumina aluminum ash. The instruction units in the control strategy set are issued to the underlying execution mechanism of the actual processing unit according to a pre-arranged sequence. The underlying execution mechanism executes the instructions to change the operating parameters of corresponding key evolution nodes, such as heating power or airflow. During instruction execution, sensors deployed in the actual processing unit continuously collect response data of the high-alumina aluminum ash. The response data is fed back to the digital twin of the processing system in real time, triggering a rapid simulation. The rapid simulation results are compared with the expected target of the control strategy set to generate a control effect deviation signal. Based on the control effect deviation signal, the parameters of subsequent instruction units to be executed are dynamically fine-tuned to form closed-loop control. This dynamic fine-tuning process analyzes the control effect deviation signal to determine whether the main source of the deviation is a deviation from thermodynamic conditions, kinetic conditions, or flow conditions. Based on the type of deviation source, the corresponding parameter correction logic is called from a preset compensation rule library. The parameter correction logic is used to calculate the required adjustment amount and direction of the current instruction unit parameters. Without violating the safety and process boundaries of the processing unit, the temperature setting, pressure setting, material flow rate, or reaction time parameters in the subsequent instruction units to be executed are superimposed and calculated. The new parameters calculated by superposition are updated to the corresponding instruction unit.

[0104] In practical implementation, a set of control strategies drives the processing system to perform closed-loop control of high-alumina aluminum ash. The control strategy set is stored in an extensible markup language file format and parsed by the central control system. Instruction units from the control strategy set are sequentially sent to the underlying actuators of the actual rotary kiln unit. These instructions include adjusting the rotary kiln speed and initiating sodium carbonate powder injection. The underlying actuators execute the instructions, changing the operating parameters of the corresponding key evolution nodes. The kiln drive frequency converter receives the speed adjustment instruction and increases the speed from 1.5 revolutions per minute to 2.2 revolutions per minute. The powder injector at the kiln head receives the start instruction and injects sodium carbonate powder at a set rate. During instruction execution, sensors continuously collect response data of the high-alumina aluminum ash, including thermocouples inside the kiln and a gas analyzer at the kiln tail. The collected response data includes temperature changes in the aluminum nitride reaction zone and changes in nitrogen concentration in the exhaust gas. The response data is fed back in real-time to the digital twin of the rotary kiln processing system, triggering a rapid simulation. This rapid simulation performs short-time step prediction calculations based on the current state variables and new boundary conditions. A deviation signal for the control effect is generated by comparing the rapid simulation results with the expected target of the control strategy set. The expected target is that the temperature at the initiation node of the aluminum nitride reaction reaches 1180°C and the nitrogen concentration rises at a rate of 0.5% per minute. The rapid simulation results show that the temperature is 1165°C and the rise rate is 0.3% per minute. Based on the deviation signal for the control effect, the parameters of the subsequent instruction units to be executed are dynamically fine-tuned to form a closed-loop control. The heating power instruction units for the subsequent maintenance phase are also dynamically fine-tuned.

[0105] In some embodiments, the specific process of dynamically fine-tuning the parameters of subsequent instruction units to be executed based on the deviation signal of the control effect includes parsing the deviation signal to determine whether the main source of the deviation is a deviation of thermodynamic conditions, kinetic conditions, or flow conditions, and comparing the differences between the expected target and the rapid simulation results in terms of temperature, concentration, and reaction rate. Based on the type of deviation source, the corresponding parameter correction logic is called from a preset compensation rule library, which is stored in the form of a lookup table, associating the deviation type with the correction algorithm. The parameter correction logic is used to calculate the required adjustment amount and direction of the current instruction unit parameters. For deviations in thermodynamic conditions, i.e., insufficient temperature, the power compensation algorithm is called. Under the premise of not violating the safety of the processing device and the process boundary, the temperature setting, pressure setting, material flow rate, or reaction time parameters in the subsequent instruction units to be executed are superimposed and calculated. The process boundary conditions include a maximum heating temperature of 1250°C and a maximum powder spraying rate. The new parameters after superposition calculation are updated to the corresponding instruction unit, and the updated instruction unit parameters are immediately issued for execution.

[0106] It is understandable that the analysis of the source of deviation relies on pattern recognition of multidimensional signals. In specific implementation, the deviation signal of the control effect is a vector containing multiple components. The compensation rule base stores parameter correction logic functions for different types of deviation. For deviations in thermodynamic conditions, the corresponding correction logic calculates the required supplementary heat and converts it into an adjustment amount for heating power. The calculation formula is as follows:

[0107]

[0108] in: This indicates the amount of heating power that needs to be adjusted. This represents the thermal efficiency coefficient of the kiln. This indicates the average density of the material inside the kiln. Indicates the volume of material in the affected area. This indicates the average specific heat capacity of the material. This represents the calculated temperature deviation. This indicates the planned time window for temperature adjustments. This formula is used to calculate the time window for adjusting the temperature of materials in the target area. Compensation within time The theoretical power required to reach the temperature, combined with the thermal efficiency coefficient. To obtain the actual heating power that needs to be adjusted .

[0109] In a specific implementation, the correspondence between the analysis of deviation sources and parameter fine-tuning is shown in Table 1:

[0110] Table 1: Correspondence between Types of Control Effect Deviation and Parameter Correction Logic

[0111]

[0112] In some embodiments, the data comparison is reflected in the parameter changes before and after the control is executed. In a specific implementation, the initial instruction unit sets the heating power of the aluminum nitride reaction zone to 850kW. The control effect deviation signal generated after rapid simulation shows a temperature deviation of -15K, which is determined to be a deviation of thermodynamic conditions. The power compensation algorithm is called from the compensation rule base to calculate the amount of power adjustment that needs to be increased. The power output is set at +22kW. Within the safe power limit of 1200kW, the heating power command unit parameter for the next time period is updated from 850kW to 872kW. The underlying actuator executes the updated command, and subsequent sensor feedback shows the temperature deviation has decreased to -3K.

[0113] Optionally, the compensation rule base is constructed based on historical data and mechanism analysis. In practice, historical data includes records of various deviations that occurred during previous processing and their successful corrections. Mechanism analysis provides quantitative relationships between different operating parameters and thermodynamic, kinetic, and flow conditions. The dynamic fine-tuning process is real-time; in practice, the entire process from sensor data feedback and rapid simulation to instruction unit parameter updates is completed within seconds, ensuring real-time intervention in the continuous processing. Safety and process boundary checks are performed immediately after parameter superposition calculations. The check logic compares the updated parameter values ​​with the allowable ranges stored in the database; if the limits are exceeded, boundary values ​​are used to replace the calculated values.

[0114] It is understandable that the successful execution of closed-loop control depends on the accuracy of rapid simulation using digital twins. In practice, rapid simulation employs a simplified reduced-order model, which is obtained by linearizing a high-fidelity digital model near a specific operating point. This significantly improves computational speed while maintaining a certain level of accuracy. Generating the control effect deviation signal requires comparing the multi-step prediction results of the rapid simulation with the multi-time-period expected target trajectories set in the control strategy set, and calculating the cumulative deviation vector over several future control cycles.

[0115] See Figure 5 This graph illustrates the dynamic trends of temperature and heating power during closed-loop control. The solid blue line represents the actual temperature change, the dashed red line represents the target temperature baseline, and the solid green line represents the real-time adjustment of heating power. The red filled area indicates the deviation range of insufficient temperature. The graph shows how the control system dynamically adjusts the heating power based on temperature deviations, ultimately stabilizing the actual temperature near the target value. The annotations of key control points show the specific adjustment actions and effects of the system at different time points.

[0116] Example 5: During the closed-loop control process, the final state of high-alumina aluminum ash is determined by continuously monitoring the dynamic evolution information captured by the virtual probe array. When the signal intensity representing the target product characteristics in the dynamic evolution information reaches saturation and stabilizes for more than a preset time, and the signal intensity representing residual impurities or by-product characteristics drops below a threshold, virtual sampling and phase analysis are performed on the overall state of the current high-alumina aluminum ash in the digital twin of the processing system. If the results of virtual sampling and phase analysis confirm that the preset final state composition and structure requirements have been met, a processing completion command is generated. After generating the processing completion command, the virtual mapping step of the resource-based product is executed. Based on the final state distribution of the high-alumina aluminum ash in the digital twin at the time of processing completion, a three-dimensional composition and structure map of the resource-based product is generated. Based on the three-dimensional composition and structure map, a preset product database is matched to predict its performance indicators as a recycled material in different application scenarios. The predicted performance indicators, three-dimensional composition and structure map are bound with the corresponding processing process data to generate a full life cycle digital archive of the processed batch of high-alumina aluminum ash.

[0117] In practical implementation, the final state of high-alumina aluminum ash is determined during the closed-loop control process. Dynamic evolution information captured by a virtual probe array is continuously monitored. This dynamic evolution information includes real-time temperature, nitrogen concentration, and alumina characteristic peak intensity signals acquired by the virtual probe array. When the signal intensity representing the target product characteristics in the dynamic evolution information reaches saturation and stabilizes for more than a preset time, and the signal intensity representing residual impurities or by-product characteristics drops below a threshold, the target product characteristic signal is the simulated thermal conductivity signal of metallic aluminum and the simulated X-ray diffraction signal of stable α-phase alumina. The residual impurity signal is the simulated residual aluminum nitride content signal. The preset time is set to ten minutes according to the process specifications, and the threshold is set to five percent of the initial signal intensity. Virtual sampling and phase analysis are performed on the overall state of the current high-alumina aluminum ash in the digital twin of the processing system. Virtual sampling randomly selects multiple points in the three-dimensional space of the digital twin, extracting temperature, pressure, and component concentration data from these points. Phase analysis is completed based on the extracted data through thermodynamic equilibrium calculations and matching with a crystal structure database. If the results of virtual sampling and phase analysis confirm that the preset final state composition and structure requirements have been met, the preset final state requirements include a metal aluminum recovery rate of more than 95%, an aluminum nitride conversion rate of more than 99%, and alumina as the α phase, then a processing completion command is generated. The processing completion command is sent to the main controller of the actual processing device through the control network to trigger the heating stop and discharge procedures.

[0118] In some embodiments, after generating the processing completion instruction, a virtual mapping step of the resource-based product is executed. Based on the final state distribution of high-alumina aluminum ash in the digital twin at the time of processing completion, a three-dimensional composition and structure map of the resource-based product is generated. The final state distribution is the full-field data output by the calibrated digital twin at the time of instruction generation. The three-dimensional composition and structure map is stored in voxel grid form, with each voxel containing phase composition, elemental concentration, and crystal orientation information. A preset product database is matched to the resource-based product based on the three-dimensional composition and structure map to predict its performance indicators as a recycled material in different application scenarios. The product database contains typical performance data of aluminum-based recycled materials with different compositions and structures in refractory materials, building materials, and steelmaking deoxidizer application scenarios. Performance indicator prediction is achieved by performing similarity calculations and regression analysis between the average composition and phase distribution uniformity of the three-dimensional map and the material data in the database. The performance indicators, three-dimensional composition and structural maps are bound to the corresponding processing data, which includes the initial material batch number, the full process parameter curves, and the control strategy set version. This generates a full life cycle digital archive of the high-alumina material aluminum ash resource utilization treatment of the completed batch. The digital archive is packaged in a standardized format and stored on a cloud server.

[0119] It is understandable that the determination of the final state relies on logical judgment of multi-signal fusion. In specific implementation, signal intensity representing the characteristics of the target product reaching saturation is defined as the signal value entering a high-level plateau period with fluctuation amplitude less than 2%. Signal intensity representing residual impurities or by-product characteristics falling below the threshold is defined as the signal value being below the threshold for twenty consecutive sampling periods. The virtual sampling and phase analysis process is automatically executed by the post-processing module within the digital twin, with no fewer than one thousand sampling points to ensure statistical representativeness. Preset final state components and structural requirements are stored in the system in the form of a technical condition list. The virtual analysis results are compared one by one with the list items, and the judgment is passed only when all conditions are met.

[0120] In some embodiments, data comparison is reflected in the signal changes during the final state determination process. In a specific implementation, the simulated signal of the thermal conductivity of metallic aluminum captured by the virtual probe array gradually increases from 0.5 W / (m·K) to 0.95 W / (m·K) at the end of the reaction and then enters a plateau period, stabilizing for more than ten minutes. Simultaneously, the simulated signal of residual aluminum nitride content decreases from an initial 100% to 0.8% and remains below 1%. Virtual sampling is performed at one thousand spatial points in the digital twin. Phase analysis shows that the phases at 973 points are metallic aluminum and α-alumina, accounting for 97.3%, meeting the requirements for metallic aluminum recovery. No aluminum nitride phase was detected at any sampling point, meeting the conversion rate requirements, and the system therefore generates a processing completion command. In the virtual mapping step of the resource-based product, the generated three-dimensional composition and structure map shows that the aluminum phase in the product is distributed in a connected network, and the alumina phase particle size is mainly distributed in the range of 10-50 micrometers. After matching the product database, it is predicted that the product's refractoriness as a refractory material raw material can reach 1750°C, and its aluminum recovery rate as a steelmaking deoxidizer can reach 92%. Data binding of the entire lifecycle digital archive is achieved by assigning a globally unique identifier to this processing procedure; all related data are linked through this identifier.

[0121] Optionally, performance index prediction can employ a similarity-based weighted calculation. In practice, the performance index prediction model calculates the Euclidean distance between the feature vector of the three-dimensional composition and structure map of the current resource product and the feature vector of each sample in the product database. The k closest samples are selected, and their weighted average performance index values ​​are used as the predicted value. The weighting is inversely proportional to the distance. The formula is expressed as:

[0122]

[0123] in: This indicates the current resource utilization product in the first... Predicted values ​​of performance metrics for each application scenario. This indicates the number of nearest neighbor samples selected from the product database. Indicates the sample index. This indicates the relationship between the current resource product feature vector and the first feature vector in the database. Euclidean distance between the feature vectors of each sample Represents the first in the database The sample at the th Known performance metrics values ​​for each application scenario. The standardized format for a full lifecycle digital archive can adopt a structure including JSON-LD or XML, containing multiple blocks such as metadata, process data, result data, and digital signatures.

[0124] It is understandable that the generation of three-dimensional composition and structure maps relies on the high fidelity of the digital twin. In practice, the final state distribution data of the digital twin is mapped from the computational grid to a finer visualization grid through an interpolation algorithm to generate a high-resolution map. The product database needs to be pre-built, and the database includes the composition and structure data of different aluminum ash resource products obtained through experimental characterization and simulation calculations, as well as their performance data measured under standard testing methods.

[0125] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0126] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for the resource utilization of aluminum ash based on high-alumina materials, characterized in that, The processing procedure includes the following steps: In the resource-based processing system, a high-fidelity digital model is established for the processing of high-alumina aluminum ash. Multidimensional parameter data of high-alumina aluminum ash in the actual processing device are collected synchronously. The multidimensional parameter data is imported into the high-fidelity digital model for iterative calibration to generate a digital twin of the processing system. A virtual probe array is set in the digital twin of the processing system. The position of the virtual probe array corresponds to the key reaction sites in the real processing device. The dynamic evolution information of high-alumina aluminum ash in the processing system is continuously captured by the virtual probe array. Based on the dynamic evolution information captured by the virtual probe array, an evolutionary chain of physical fields and matter flow is constructed, including: Spatiotemporal distribution data of temperature gradient field, concentration gradient field and reaction rate field are extracted from the dynamic evolution information captured by the virtual probe array. The coupling relationship and hysteresis effect between the spatiotemporal distribution data of temperature gradient field, concentration gradient field and reaction rate field were analyzed. Based on the coupling relationship and hysteresis effect, a physical field state transition sequence with time as the axis is established; In the physical field state transition sequence, mark the key events that lead to phase transformation or reaction path bifurcation; Using key events as anchor points, we can link them to changes in the composition of the material flow, forming an evolutionary chain in which the evolution of the physical field and the changes in the composition of the material flow are interconnected. Based on the evolution chain of physical field and material flow, the processing path of high-alumina aluminum ash is reversed to identify the key evolution nodes and constraints in the processing path. The identified key evolution nodes and constraints are transformed into a set of control strategies, which are then used to drive the processing system to perform closed-loop control on the high-alumina aluminum ash.

2. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 1, characterized in that, A high-fidelity digital model was established for the processing of high-alumina aluminum ash, including: Analysis of the phase composition and chemical reaction network of high-alumina aluminum ash; Based on phase composition and chemical reaction networks, a core mechanism model library including thermodynamic equilibrium, reaction kinetics and multiphase transport processes is established. Based on the geometric configuration and operational logic of the actual processing device, construct the device structure model and process logic model around the core mechanism model library; The core mechanism model library, device structure model and process logic model are coupled to form the high-fidelity digital model.

3. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 2, characterized in that, Multidimensional parameter data of high-alumina aluminum ash from the actual processing device are collected synchronously, and the multidimensional parameter data is imported into the high-fidelity digital model for iterative calibration, including: Multiple types of sensors are deployed at different depths and key reaction sites in the actual processing device to obtain real-time data streams of temperature, pressure, gas composition and material morphology. Timestamp alignment and outlier cleaning are performed on real-time data streams to form structured multidimensional parameter data; Structured multidimensional parametric data are used as boundary conditions and observations, and then input into a high-fidelity digital model. Drive a high-fidelity digital model to perform forward simulation, compare the simulation output with the corresponding real observations, and generate a residual sequence. Based on the residual sequence, the dynamic parameters and transfer coefficients in the high-fidelity digital model are adjusted in reverse using a parameter estimation algorithm until the degree of agreement between the simulation output and the actual observations meets the preset convergence criterion, thus completing the generation of the digital twin.

4. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 3, characterized in that, Based on the evolutionary chain of physical fields and material flow, the processing path of high-alumina aluminum ash is deduced in reverse, including: Starting from the final target phase state of the evolutionary chain, trace back to its immediate physical field and the preceding state of matter flow; Analyze the transition process from the predecessor state to the current state to identify the main control variables and reaction conditions driving the transition process; Continue tracing back to the predecessor state of the predecessor state, repeating the above identification process, until the initial material state is reached; Along the complete backtracking path, the difficulty and energy consumption of all transformation processes are marked, and the transformation process nodes with relatively low transformation efficiency are selected and identified as key evolution nodes. We analyze the internal factors and external conditions that limit the transformation efficiency of key evolution nodes and summarize them as constraints.

5. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 4, characterized in that, The identified key evolution nodes and constraints are transformed into a set of control strategies, including: For each key evolution node, multiple virtual intervention schemes are preset in the digital twin of the processing system; Simulate and run each virtual intervention scheme in a digital twin to obtain the improvement in transformation efficiency of each intervention scheme on key evolution nodes under corresponding constraints; All virtual intervention schemes were evaluated and ranked based on both the magnitude of improvement and energy consumption cost. Select virtual intervention schemes whose evaluation results meet the preset thresholds, and compile their specific operation parameters, execution sequence and action points into instruction units that can be recognized by the control system of the real treatment device. The instruction units corresponding to all key evolution nodes are combined and serialized to generate the control strategy set.

6. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 5, characterized in that, A set of control strategies is used to drive the processing system to perform closed-loop control on high-alumina aluminum ash, including: The instruction units in the control strategy set are sent to the underlying execution mechanism of the actual processing device according to the arrangement sequence; The underlying execution mechanism executes instructions to change the operational parameters of the corresponding key evolution nodes; During command execution, response data of high-alumina aluminum ash is continuously collected through sensors; The response data is fed back to the digital twin of the processing system in real time, triggering the digital twin to perform a rapid simulation; Compare the results of rapid simulation with the expected targets of the control strategy set to generate a deviation signal of the control effect; Based on the deviation signal of the control effect, the parameters of the subsequent instruction units to be executed are dynamically fine-tuned to form a closed-loop control.

7. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 1, characterized in that, The method also includes a step for determining the final state of the high-alumina aluminum ash treatment: During the closed-loop control process, the dynamic evolution information captured by the virtual probe array is continuously monitored; When the signal strength representing the characteristics of the target product in the dynamic evolution information reaches saturation and stabilizes for more than a preset time, and the signal strength representing the characteristics of residual impurities or by-products drops below the threshold; Virtual sampling and phase analysis are performed on the overall state of the current high-alumina aluminum ash in the digital twin of the processing system. If the results of virtual sampling and phase analysis confirm that the preset final state composition and structure requirements have been met, a processing completion command will be generated.

8. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 7, characterized in that, After generating the processing completion instruction, the process also includes a virtual mapping step for the resource products: Based on the final state distribution of high-alumina aluminum ash in the digital twin after processing, a three-dimensional composition and structure map of the resource-based product is generated. Based on the three-dimensional composition and structure map, a pre-set product database is matched for the resource products, and the performance indicators of the products as recycled materials in different application scenarios are predicted. By binding performance indicators, three-dimensional composition and structural maps with corresponding processing data, a full life-cycle digital archive of the high-alumina material aluminum ash resource utilization treatment is generated.

9. The method for resource utilization of aluminum ash based on high-alumina materials as described in claim 8, characterized in that, The step of dynamically fine-tuning the parameters of subsequent instruction units to be executed based on the deviation signal of the control effect includes: Analyze the deviation signal of the control effect to determine whether the main source of the deviation is the deviation of thermodynamic conditions, kinetic conditions, or flow conditions. Based on the type of deviation source, the corresponding parameter correction logic is called from the preset compensation rule library; The parameter correction logic is used to calculate the required adjustment amount and direction of the current instruction unit parameters; Without violating the safety and process boundaries of the processing device, the temperature setting, pressure setting, material flow rate or reaction time parameters in the subsequent instruction units to be executed are superimposed and calculated. The new parameters calculated by superposition are updated to the corresponding instruction unit.

Images