Optimization method and system for graphite sagger compression molding process

Abstract

The invention relates to the technical field of process optimization, and discloses an optimization method and system for a graphite sagger compression molding process. The method comprises the following steps: determining and analyzing particle size distribution, specific surface area, porosity and viscosity-temperature characteristics of graphite powder and a binder to obtain raw material characteristic parameter data; modeling is conducted on the migration behavior of the binder and the deformation characteristic of the graphite powder in the pressing process, a material rheological model is obtained, meanwhile, a multi-point pressure sensor system is arranged on the surface of a pressing mold, and target pressure data are collected; constructing a pressure distribution feature vector and a material flow state prediction result; performing adaptive adjustment on the initial pressing parameters to obtain a segmented pressure curve and a local pressure compensation strategy; and performing multi-objective optimization on the graphite sagger pressing path to obtain an optimal pressing process parameter combination. The method is combined with material flow state prediction, accurate control over the pressing process is achieved, and the intelligent level of the graphite sagger pressing forming technology is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of process optimization, and particularly to an optimization method and system for the pressing and forming process of graphite crucibles. Background Art

[0002] The traditional pressing and forming process of graphite crucibles mainly adopts a single constant pressure or simple segmented pressurization method, which has obvious density non-uniformity problems in practical applications. Especially in the corner parts of complex-shaped crucibles, low-density regions are likely to be formed, resulting in easy cracking and failure of products in high-temperature application environments. In addition, the adaptability of the existing pressing process to the characteristics of raw materials is poor. When facing different batches of graphite powder and binder, fixed pressing parameters are difficult to cope with the fluctuations of material characteristics, resulting in unstable product quality.

[0003] During the pressing and forming process of graphite crucibles, there is a lack of accurate description models for the migration behavior of the binder and the deformation characteristics of graphite powder under pressure. The process parameters mainly rely on empirical settings, lacking systematic data support and theoretical guidance. Existing monitoring systems usually only focus on the total pressure at the top of the press and cannot obtain the local pressure distribution information inside the mold, especially unable to monitor the pressure state at the corner parts in real time, resulting in difficult prediction and control of the material flow state during the pressing process. Summary of the Invention

[0004] The present invention provides an optimization method and system for the pressing and forming process of graphite crucibles. The present invention combines the prediction of the material flow state to achieve precise control of the pressing process and improve the intelligent level of the pressing and forming process of graphite crucibles.

[0005] In the first aspect, the present invention provides an optimization method for the pressing and forming process of graphite crucibles, and the optimization method for the pressing and forming process of graphite crucibles includes: Measure and analyze the particle size distribution, specific surface area, porosity, and viscosity-temperature characteristics of graphite powder and binder to obtain raw material characteristic parameter data; Based on the raw material characteristic parameter data, model the migration behavior of the binder and the deformation characteristics of graphite powder during the pressing process to obtain a material rheological model. At the same time, arrange a multi-point pressure sensor system on the surface of the pressing mold and collect target pressure data; Perform Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and input the target pressure data into the material rheological model for flow state prediction to obtain a material flow state prediction result; According to the pressure distribution feature vector and the material flow state prediction result, adaptively adjust the initial pressing parameters to obtain a segmented pressure curve and a local pressure compensation strategy; Based on the segmented pressure curve and the local pressure compensation strategy, the pressing path of the graphite crucible is optimized in multiple objectives to obtain the optimal combination of pressing process parameters.

[0006] In a second aspect, the present invention provides an optimization system for the graphite crucible pressing and forming process. The optimization system for the graphite crucible pressing and forming process includes: A measurement module for measuring and analyzing the particle size distribution, specific surface area, porosity, and viscosity-temperature characteristics of the graphite powder and the binder to obtain raw material characteristic parameter data; A modeling module for modeling the binder migration behavior and the graphite powder deformation characteristics during the pressing process based on the raw material characteristic parameter data to obtain a material rheological model. At the same time, a multi-point pressure sensor system is arranged on the surface of the pressing die and target pressure data is collected; A prediction module for performing Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and inputting the target pressure data into the material rheological model for flow state prediction to obtain a material flow state prediction result; An adjustment module for adaptively adjusting the initial pressing parameters according to the pressure distribution feature vector and the material flow state prediction result to obtain a segmented pressure curve and a local pressure compensation strategy; A multi-objective optimization module for optimizing the pressing path of the graphite crucible in multiple objectives based on the segmented pressure curve and the local pressure compensation strategy to obtain the optimal combination of pressing process parameters.

[0007] In the technical solution provided by the present invention, by real-time monitoring the pressure distribution through a multi-point pressure sensor system and combining with the prediction of the material rheological model, the pressing parameters in the easily low-density areas such as the corners of the crucible can be accurately identified and adjusted, effectively improving the density uniformity of the product and enhancing the quality stability of the product. Based on the raw material characteristic parameter database and the binder migration model, the system can automatically adapt to the characteristic differences of different batches of raw materials, and dynamically adjust the pressing parameters through an adaptive control algorithm to ensure the consistency of product quality. An improved Kalman filtering algorithm is used to process the pressure data, combined with the prediction of the material flow state, to achieve precise control of the pressing process and break through the limitations of traditional empirical parameter settings. The multi-objective optimization method simultaneously considers four indicators of product density uniformity, strength, energy consumption, and die life, and generates an optimal pressurization path curve through an improved NSGA-II algorithm to achieve the balance of quality, efficiency, and cost. The present invention reduces the dependence on manual experience and improves the intelligent level of the graphite crucible pressing and forming process. Description of the Drawings

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0009] Figure 1 This is a schematic diagram of an embodiment of the optimization method for the graphite crucible pressing and forming process in the embodiments of the present invention; Figure 2 This is a schematic diagram of an embodiment of the optimization system for the graphite crucible pressing and forming process in the embodiments of the present invention. Specific embodiments

[0010] The embodiments of the present invention provide an optimization method and system for the graphite crucible pressing and forming process. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned accompanying drawings of the present invention are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0011] For ease of understanding, the following describes the specific process of the embodiments of the present invention. Please refer to Figure 1 An embodiment of the optimization method for the graphite crucible pressing and forming process in the embodiments of the present invention includes: Step S101: Measure and analyze the particle size distribution, specific surface area, porosity, and viscosity-temperature characteristics of the graphite powder and the binder to obtain raw material characteristic parameter data; It can be understood that the execution subject of the present invention can be an optimization system for the graphite crucible pressing and forming process, or a terminal or a server. Specifically, no limitation is made here. The embodiments of the present invention take the server as the execution subject as an example for illustration.

[0012] Specifically, the particle size distribution of graphite powder is measured. A high-precision laser particle size analyzer is used to scan graphite raw powder samples of different batches to obtain particle size distribution characteristic parameters such as D10, D50, and D90, and based on this, the particle size distribution width coefficient of the powder is analyzed to ensure the controllability and consistency of the physical properties of the raw materials. The specific surface area parameter of the graphite powder is measured using a specific surface area analyzer to quantify the microscopic surface structure of the sample with an accuracy of 0.01 m 2 / g. Furthermore, the mercury intrusion method is combined to analyze the porosity and pore size distribution of the powder to obtain the average pore size and total pore volume of the sample. These microscopic structure parameters have an impact on the subsequent binder migration and densification process during pressing. To fully characterize the rheological properties of the binder, measurements are carried out using a rotational viscometer at multiple temperature points such as 20°C, 40°C, 60°C, 80°C, and 100°C to obtain the viscosity-temperature relationship curve of the binder. A differential scanning calorimeter is used to analyze the thermal properties of the binder sample to obtain its melting point and freezing point data, and provide a theoretical basis for setting the pressing temperature range. At the same time, a universal material testing machine is used to conduct mechanical property tests on the binder at a set strain rate, measure the stress-strain curve, and calculate its elastic modulus and yield strength based on this. The graphite powder and the binder are mixed in different proportions from 5% to 15%. By testing the bulk density, tapped density, and fluidity of samples with each mixing ratio, the physical property data of the mixture as a whole are supplemented to form a comprehensive understanding of the rheological behavior of the raw materials. The graphite powder particle size distribution data, specific surface area values, porosity data, binder viscosity-temperature relationship curve data, melting point and freezing point data, stress-strain curve data, as well as the test results of the bulk density, tapped density, and fluidity of the mixture are input into a relational database system according to the preset fields and batch numbers. This database adopts a structured organization method to classify all the parameters of each batch of materials into the same record, facilitating subsequent efficient retrieval, parameter sensitivity analysis, and multi-batch data comparison, and forming a high-precision raw material characteristic parameter dataset covering the physical, chemical, and rheological properties of the materials.

[0013] Step S102: Based on the raw material characteristic parameter data, model the binder migration behavior and the deformation characteristics of the graphite powder during the pressing process to obtain a material rheological model. At the same time, arrange a multi-point pressure sensor system on the surface of the pressing die and collect the target pressure data; Specifically, taking porosity as the core parameter, combined with database information such as particle size distribution and average pore size, the permeability coefficient of graphite powder is calculated based on the porous media seepage theory. Using the mathematical model of pressure gradient driven flow, the permeability coefficient, actual pressure gradient and binder viscosity data determined by temperature are combined to establish the rate equation for the migration of binder from the high-pressure area to the low-pressure area of ​​the mold. This migration model takes into account the microstructure of the raw materials and dynamically reflects the real-time temperature and pressure conditions during the pressing process, making the binder distribution prediction more accurate. At the same time, in order to reflect the mechanical response of graphite powder during the pressing process, the cohesion and internal friction angle of the powder system are calculated based on the experimental data of bulk density and tap density, and the material deformation model under the Drucker-Prager criterion is constructed based on this as the key parameter, realizing a unified description of the macroscopic densification of graphite powder and the slip characteristics between microscopic particles. By discretizing the binder migration model and the Drucker-Prager deformation model in three-dimensional space, using the finite difference or finite element numerical method to divide the grid in the pressing cavity, and solving the model for each grid unit, the material rheological model is obtained, which can dynamically reflect the evolution of density, pressure and binder content at different spatial positions during the pressing process. At the same time, the structure of the pressing mold is simulated by finite element analysis software to identify the monitoring points where density and pressure anomalies are most likely to occur in various complex structures and corner areas. According to the simulation results, multi-point pressure sensors are reasonably distributed on the mold surface to form a pressure monitoring network with uniform spatial distribution and efficient reflection of global and local states. Each pressure sensor is pre-amplified, filtered and digitally isolated by a high-performance signal conditioning circuit, and is uniformly collected in an industrial-grade embedded controller through a multiplexer to build a high-real-time, multi-channel multi-point pressure acquisition system. The system supports zero calibration and full-scale automatic calibration of the sensor array to ensure the accuracy and consistency of pressure data at each point. During the actual pressing process, the multi-point pressure sensor system can collect the real-time pressure values ​​of each monitoring point in the mold with high frequency and low latency, and simultaneously input the target pressure data as feedback into the material rheology model, thereby realizing data-driven process optimization and intelligent control.

[0014] Step S103, performing Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and inputting the target pressure data into a material rheological model to perform flow state prediction to obtain a material flow state prediction result; Specifically, a collaborative analysis system of dynamic filtering and physical modeling based on the target pressure data of the pressing process is established. Taking the actually collected multi-point pressure data as the input, a state equation describing the dynamic changes of the system is set, and the measured pressure values of each sensor are used as the output of the observation equation. The state vector includes the current pressure values of each monitoring point and also includes its pressure change rate to reflect the dynamic characteristics of the system, while the observation vector directly corresponds to the real-time output value of the sensor, thus establishing the structure of the Kalman filter model. The Kalman filter model is processed by cyclic iteration according to the two calculation stages of prediction and update. In the prediction stage, the prior state estimate and prior error covariance matrix of the system are calculated based on the posterior state estimate and state transition equation of the previous moment. Then, in the update stage, the Kalman gain is calculated using the residual between the current observation data and the prediction result, and then the prior state is corrected to obtain the posterior state estimate. At the same time, the error covariance is updated to obtain pressure state data with excellent filtering effect. On this basis, to cope with the uncertainty of the noise covariance in the actual process environment such as working condition fluctuations and measurement anomalies, the noise covariance matrix of the Kalman filter is dynamically and adaptively adjusted. By real-time analyzing the residual sequence, when the system detects multiple residual over-threshold or abnormal jumps, the observation noise covariance is automatically increased, thereby enhancing the robustness of the model to sudden disturbances and ensuring that the filtered pressure state data is more real and reliable. Based on the above adaptive Kalman filter results, information such as the pressure mean value, standard deviation, extreme value, and gradient of each region is statistically analyzed to form a high-dimensional feature vector describing the pressure distribution of the pressing cavity, reflecting the spatial inhomogeneity of the material stress and flow. The latest target pressure data and the pressure distribution feature vector after Kalman filtering are imported into the material rheology model established in the early stage. The initial state of material flow is set according to the functional relationship between the current pressure and the die displacement, and each key parameter is used as the input of a three-dimensional finite element or finite difference model, thus realizing the high-precision initial value setting of flow analysis. The material rheology model is discretized into three-dimensional space grids using numerical calculation methods, and the dynamic change process of the binder concentration and graphite powder density over time is synchronously solved on each grid unit to obtain the material density distribution data at the current and several future moments. On this basis, by comparing and analyzing the density fields at different moments, key parameters such as the slope, curvature, and inflection point position of the displacement-pressure curve are calculated. These characteristics can reflect the rate of material flow and the densification stage, and are used to identify abnormal states or bottleneck links in the process. Based on the above density distribution data and curve characteristics, the model predicts the material distribution state at the next time step and even in the subsequent stages of the entire pressing process, realizing the dynamic monitoring and trend prediction of the entire process flow.

[0015] The cavity space is divided by three-dimensional mesh discretization technology, and the input parameters of material flow analysis such as pressure distribution, initial density, initial binder content, etc. are mapped to each mesh node to obtain a meshed pressing cavity model that reflects the actual process state. Based on this model, according to the physical boundary characteristics of the mold, solid wall no-slip boundary conditions are applied to all wall nodes of the cavity to ensure that the material will not slip relative to the wall during the pressing process. At the boundary nodes corresponding to the pressing direction, pressure or displacement boundary conditions consistent with the actual process are set to more accurately simulate the transmission and distribution of the pressing load in space. Relying on the binder migration model established in the early stage, the pressure-driven binder migration rate equation is discretized in three-dimensional space using the finite difference method, thereby converting the continuous partial differential equations into a discrete equation that can be solved on the computer, and using the grid unit as the basic calculation unit, providing an efficient path for the subsequent numerical simulation of the evolution of binder concentration and spatial distribution. At the same time, in order to achieve an accurate description of the macroscopic deformation of graphite powder and the contact and slip of local particles, the constitutive parameters such as material cohesion and internal friction angle obtained under the Drucker-Prager criterion are mapped to each spatial grid unit with the help of tensor transformation method, so that each unit has independent deformation calculation parameters reflecting local mechanical behavior. Based on the localized tensor parameters, for each grid unit under its own force and flow conditions, the discretized binder migration equation and powder deformation equation are solved jointly, and the change process of the binder concentration and graphite powder density in the unit is dynamically calculated in each time step of the pressing, and the original calculation results of the space-time distribution of the whole pressing process are gradually deduced. In order to avoid the sudden change of density or concentration caused by grid division, the original calculation results of each time step are smoothed overall by using the volume weighted average method, and then the discrete grid data is improved by the three-dimensional interpolation algorithm. The generated density distribution data can truly and delicately reflect the densification of the materials in each area inside the entire cavity and the distribution state of the binder.

[0016] Step S104, adaptively adjusting the initial pressing parameters according to the pressure distribution characteristic vector and the material flow state prediction result to obtain a segmented pressure curve and a local pressure compensation strategy; Specifically, taking the dynamically collected pressure distribution feature vectors and the prediction results of the material flow state as the core data sources, a comprehensive evaluation function integrating multiple objectives such as density uniformity, product strength, energy consumption, and die life is constructed. This comprehensive evaluation function quantitatively measures the global optimality of the process results. At the same time, by scientifically setting quantitative indicators such as density standard deviation, average density, unit energy consumption, and die stress, product quality, production efficiency, and manufacturing cost are linked together. For the characteristics of raw materials in different batches and the actual operating state of the die, an adaptive weight adjustment mechanism based on fuzzy rules is introduced. For example, when the fluidity of the raw material binder is insufficient, the weight of density uniformity in the evaluation function is automatically increased. When the die is approaching the end of its life, the weight coefficient of the die life index is correspondingly increased, so as to dynamically balance the priority of each objective in the pressing process and keep the process always in the optimal state window. On this basis, combined with the established material rheological model, through the model predictive control method, the global pressing trajectory is dynamically planned to achieve the rolling optimization of key process parameters such as pressure and displacement in the entire pressing time sequence. The model predictive control module is based on the physical change law of materials, combines the pressure distribution and flow state feedback in real time, predicts the optimal control path at several future moments, and outputs the target reference information of the global pressing trajectory. To cope with local disturbances and spatial distribution non-uniformity in the pressing process, on the basis of the global trajectory planning, the state space of the system is defined as the current pressure distribution, displacement and their change rates, and the action space is set as the pressure adjustment amount and the holding time adjustment amount of each region. By introducing an enhanced learning algorithm based on the Actor-Critic structure, the system can autonomously learn and iteratively optimize the rapid response strategy for pressure abnormal points and material flow bottlenecks at each moment, so as to strengthen the local adaptability and disturbance robustness of the process while ensuring the optimization of overall density and strength. Based on the local optimization strategy obtained by enhanced learning control, the whole pressing process is automatically divided into four stages with clear physical and technological meanings: low-pressure pre-compaction, medium-pressure main compaction, high-pressure densification, and final holding pressure. A reasonable pressure range and application time are matched for each stage. The obtained segmented pressure curve can maximize the product performance under the process window. For areas such as the corners where low density and forming defects are likely to occur, through a targeted local pressure compensation strategy, the pressure amplitude and action duration of the corner area are finely adjusted on the basis of the above segmented pressure curve to achieve active compensation for key defect points, thereby improving the densification and structural integrity of complex graphite crucible products and obtaining the local pressure compensation strategy.

[0017] Step S105: Based on the segmented pressure curve and the local pressure compensation strategy, perform multi-objective optimization on the graphite crucible pressing path to obtain the optimal combination of pressing process parameters.

[0018] Specifically, the segmented pressure curve is parametrically expressed, and methods such as piecewise functions or cubic B-splines are used to describe parameters such as the pressure level, action time, and pressure change rate in each pressing stage, so as to express the complex pressurization path in the form of a finite-dimensional parameter vector, providing a standardized basis for the iteration and calculation of the optimization algorithm. On this basis, aiming at the core requirements of the graphite crucible pressing process, a multi-objective optimization function group is comprehensively set, including product density uniformity, product strength, unit energy consumption, and maximum equivalent stress of the mold. Among them, density uniformity is measured by the standard deviation of the finished product density, product strength depends on the bending or compressive properties of the final pressed product, unit energy consumption is evaluated by the ratio of the energy consumed in the pressing process to the product quality, and the maximum equivalent stress of the mold reflects the influence of the process on the mold life. The parametric expression of the pressurization path and the multi-objective optimization function group are input into the improved NSGA-II algorithm. This algorithm is guided by the multi-objective Pareto optimality theory and realizes the comprehensive search and survival of the fittest for different combinations of process parameters through population intelligent evolution. In the optimization process, adaptive crossover operators and mutation operators are introduced, and the crossover probability and mutation probability are dynamically adjusted according to the optimization progress and population diversity, so as to effectively prevent falling into local optima, while enhancing the exploration ability of the population for the global solution space and ensuring the comprehensiveness and diversity of the final solution set. For each new solution generated by population iteration, the material rheological model is called for physical layer simulation and evaluation to simulate the density distribution, energy consumption, mold stress, and product strength performance of each parameter combination in the actual pressing process, so as to ensure that the calculation results of all objective functions are based on real physics, enhancing the engineering applicability and reliability of the optimization scheme. With the continuous advancement of optimization iteration, the NSGA-II algorithm finally obtains a set of non-dominated Pareto front solution sets, and these candidate solutions represent different optimization directions under multi-objective trade-offs. Based on specific production requirements and process emphases, the fuzzy decision-making method is used to post-process the candidate solution set, and the membership function is used to normalize the advantages and disadvantages of each solution on different objectives, or the weight distribution is adjusted in combination with the producer's preference, so as to select the optimal pressing process parameter combination that best meets the current enterprise production reality and future development direction from the multi-objective Pareto front.

[0019] For the parameter space of the pressurization path, M initial solutions are randomly generated by methods such as random perturbation or Latin hypercube sampling. Each initial solution corresponds to a set of specific variables such as segmented pressure parameters, action duration, and local compensation strategies, and is respectively input into the material rheological model and the multi-objective optimization function group for physical simulation and multi-index evaluation to obtain a fitness vector covering multiple objectives such as density uniformity, product strength, energy consumption, and die life, and then an initial population required for the algorithm operation is constructed. This initial population undergoes fast non-dominated sorting, and non-dominated front individuals of different levels are automatically divided according to the Pareto dominance relationship, providing a scientific stratification basis for the preferential selection of parent individuals in the subsequent generation. On this basis, according to the sorting results and crowding degree information of the initial population, some excellent parent individuals are selected for breeding the next generation of solution sets. To improve the global exploration ability of the algorithm and prevent premature convergence, the crossover probability is dynamically adjusted according to the diversity and distribution of the current iterative population in the solution space. For example, when the population distribution tends to converge, the crossover probability is appropriately increased to enhance the diversity of solutions, while when the diversity is high, the crossover probability is decreased to strengthen the genetic transmission of the superior solutions. The selected parent individuals are subjected to simulated binary crossover operations, and new generation of offspring individuals are generated in the parameter space by combining the gene information of the parent individuals, realizing the inheritance and innovation of key parameters such as the pressurization path. The generated offspring individuals dynamically adjust the mutation probability based on the difference between their fitness values and the fitness values of the parent individuals, that is, if the new generation of solution sets is significantly different from the previous generation, the mutation probability is decreased to steadily approach the optimal Pareto front, and if the solution sets tend to be consistent, the mutation probability is increased to prevent falling into local optima. During this process, polynomial mutation operations are applied to each offspring individual to randomly perturb some of its parameters in order to discover more potential solutions globally. All mutated offspring individuals are combined with the parent individuals to form a larger intermediate population, and fast non-dominated sorting and crowding distance calculation are performed on this population again. Through the elite selection strategy, individuals with higher rankings and uniform distribution in the objective space are preferentially retained to screen and form the next generation of population. After iterating several generations like this, the NSGA-II algorithm can gradually converge to the non-dominated Pareto optimal solution set covering the core process indicators such as density, strength, energy consumption, and life while ensuring the comprehensive optimization of multiple objectives.

[0020] In the embodiments of the present invention, by real-time monitoring of the pressure distribution through a multi-point pressure sensor system and combining with the prediction of the material rheological model, the pressing parameters of the low-density areas such as the corners of the sagger can be accurately identified and adjusted, effectively improving the density uniformity of the product and enhancing the quality stability of the product. Based on the raw material characteristic parameter database and the binder migration model, the system can automatically adapt to the characteristic differences of different batches of raw materials, and dynamically adjust the pressing parameters through an adaptive control algorithm to ensure the consistency of product quality. The improved Kalman filtering algorithm is used to process the pressure data, and combined with the prediction of the material flow state, the precise control of the pressing process is realized, breaking through the limitations of traditional empirical parameter settings. The multi-objective optimization method is adopted to simultaneously consider four indicators of product density uniformity, strength, energy consumption, and die life, and the optimal pressurization path curve is generated through the improved NSGA-II algorithm to achieve the balance of quality, efficiency, and cost. The present invention reduces the dependence on manual experience and improves the intelligent level of the graphite sagger pressing and forming process.

[0021] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Use a laser particle size analyzer to measure the characteristic particle size of the graphite powder, obtain the particle size distribution data of the graphite powder, and measure the graphite powder using a specific surface area analyzer and mercury intrusion porosimetry respectively to obtain the specific surface area value and porosity data; Use a rotational viscometer to measure the binder to obtain the binder viscosity-temperature relationship curve data; Use a differential scanning calorimeter to perform thermal characteristic analysis on the binder to obtain the melting point and freezing point data, and measure the binder at a preset strain rate to obtain the stress-strain curve data; Integrate the graphite powder particle size distribution data, specific surface area value, porosity data, binder viscosity-temperature relationship curve data, melting point and freezing point data, and stress-strain curve data, as well as the bulk density, tapped density, and fluidity test results of the mixture into a relational data structure to obtain the raw material characteristic parameter data.

[0022] Specifically, the microscopic structure and distribution characteristics of the graphite powder are systematically measured. Taking the laser particle size analyzer as the core tool, instrument equipment with high sensitivity and wide detection range is selected, and the graphite raw powder samples of different batches are fully dispersed and pretreated to ensure that the samples are evenly dispersed and then enter the test system. The volume distribution information of the particles is automatically collected through the multi-angle light scattering principle, and the data results including representative characteristic particle size values such as D10, D50, and D90 are output. A specific surface area analyzer is used to measure the nitrogen adsorption isotherm of the sample under standard conditions by physical adsorption method, and the specific surface area value of the powder is calculated based on the BET theory, and its measurement accuracy needs to reach 0.01m 2 / g. A high specific surface area means that the powder is in more sufficient contact with the binder in the mixture, directly affecting the uniformity of the binder distribution and the mechanical properties of the final pressed product. At the same time, the mercury intrusion method is used to analyze the same batch of samples. By forcing mercury into the pores of the graphite powder under high pressure, the intrusion volume of mercury at different pressures is recorded, and parameters such as pore size distribution, total pore volume, and porosity are deduced. These indexes describe the microstructural characteristics of the powder, play a decisive role in the migration, penetration, and distribution of the binder during the pressing process, and are the basic factors affecting the density uniformity and strength of the finished crucible. Turn the analysis object to the binder material. The rheological properties of the binder are tested by a rotational viscometer at multiple temperature conditions such as 20 °C, 40 °C, 60 °C, 80 °C, and 100 °C. On the premise of ensuring sufficient contact between the rotor and the sample and no bubbles, the shear viscosity values corresponding to different temperatures are obtained to get the viscosity-temperature relationship curve of the binder. The thermal properties of the binder sample are analyzed by a differential scanning calorimeter. The sample is weighed according to the standard amount and sealed in a test crucible, heated at a set rate, and the heat flow signal is recorded. Key data such as the melting point, freezing point, and heat distortion temperature of the binder are extracted through the exothermic and endothermic peaks of the thermal curve. These thermal characteristic parameters help to determine the phase change interval and available temperature range of the binder in the actual process environment, preventing uneven distribution or performance deterioration caused by premature melting or solidification during the pressing process. The binder is subjected to tensile or compressive tests at different strain rates by a universal material testing machine. By recording the relationship curve between the load and the deformation, stress-strain data are obtained to analyze the elastic modulus, yield strength, and plastic deformation range of the binder, which helps to evaluate its adaptability under the pressing stress, flow limit, and the impact on the forming quality of the product. Combining the above basic physical, chemical, and thermodynamic characteristic parameters of the graphite powder and the binder provides high-precision input for the material rheological modeling and binder migration behavior analysis of the entire pressing process. After obtaining all the basic data of the above single components, the graphite powder and the binder are mixed and stirred evenly according to the set ratios (such as 5%, 7.5%, 10%, 12.5%, 15%, etc.) to prepare the mixture to be pressed. The bulk density test, tapped density test, and flowability evaluation are respectively performed on the mixed samples. The bulk density and tapped density are realized by the graduated cylinder measurement method and the standard tapping device. The former reflects the space utilization rate of the free packing of the mixture, and the latter measures the rearrangement and compaction ability of the material under vibration disturbance. Together, they determine the initial uniformity of the material distribution in the mold cavity. The flowability test adopts the standard funnel flow time method or the angle of repose method to quantitatively evaluate the flow rate of the mixture through a specified aperture under the action of gravity, so as to analyze its filling and distribution ability in the mold.After completing the tests and measurements of all individual parameters, collect and organize all the data, including the particle size distribution data of the graphite powder, the specific surface area value, the porosity data, the viscosity-temperature relationship curve of the binder, the melting point and solidification point data of the binder, the stress-strain curve data, as well as the bulk density, tapped density, and flowability test results of each group of mixtures. For the needs of large-scale experiments and multi-batch production, use an efficient relational database system for data management. Classify according to multiple indexing structures such as raw material batch number, detection date, test method, parameter type, etc., and enter the full set of data for each batch combination of graphite powder and binder into an independent record unit of the database to ensure the integrity, accuracy, and traceability of the data.

[0023] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Calculate the permeability coefficient of the graphite powder based on the porosity data in the raw material characteristic parameter data, and establish a rate equation for the binder to flow from the high-pressure area to the low-pressure area under the action of pressure according to the permeability coefficient to obtain a binder migration model; Calculate the cohesion and internal friction angle parameters based on the bulk density and tapped density in the raw material characteristic parameter data, and model the deformation characteristics of the graphite powder according to the cohesion and internal friction angle parameters to obtain a Drucker-Prager deformation model; Discretize and solve the binder migration model and the Drucker-Prager deformation model to obtain a material rheology model; Determine the target monitoring points in the pressing die through finite element analysis, and arrange multi-point pressure sensors on the die surface to obtain a pressure monitoring network; Connect the signal conditioning circuit to the pressure monitoring network and collect it through a multiplexer to an embedded controller to obtain a multi-point pressure sensor system; Use the multi-point pressure sensor system to collect the real-time pressure values of each monitoring point during the pressing process, and obtain the target pressure data through zero calibration and full-scale verification.

[0024] Specifically, based on the porosity data obtained from the previous raw material characteristic parameter test, the porous media seepage theory was adopted, and appropriate empirical formulas (such as the Kozeny-Carman equation) were selected. The average porosity, particle size and specific surface area of ​​the graphite powder were substituted into the model to calculate the permeability coefficient of the graphite powder. The permeability coefficient is a key physical quantity that characterizes the flow capacity of the internal fluid of the material. The larger its value, the faster the migration speed of the binder in the powder pore network. After obtaining the permeability coefficient, it was combined with the pressure gradient in the actual pressing process and the real-time viscosity of the binder to construct a rate equation that describes the migration of the binder from the high-pressure area to the low-pressure area driven by the pressure field. This equation reflects the joint influence of the raw material microstructure and the pressing process parameters on the evolution of the binder distribution. At the same time, according to the bulk density and tap density data of different batches of mixtures in the raw material database, the parameter conversion method commonly used in the fields of soil mechanics and powder metallurgy was used to calculate the cohesion and internal friction angle of the graphite powder system. These two parameters characterize the adhesion ability and shear slip resistance between particles, respectively, and are important constitutive constants that describe the macroscopic mechanical response of graphite powder. Based on the above mechanical parameters and the Drucker-Prager yield criterion as the theoretical basis, the constitutive relationship of the deformation characteristics of graphite powder is established to achieve unified modeling of the elastic-plastic deformation, yield and densification behavior of the powder under the pressing force. In this model, the strength, yield limit and shear failure of the material can be accurately controlled by the cohesive force and the internal friction angle, so that the model can not only adapt to the physical property changes between different raw material batches, but also reflect the local deformation and density distribution characteristics of special parts such as complex corners and variable cross-sections. The binder migration model is integrated with the Drucker-Prager deformation model, and the two are discretized in three-dimensional space using numerical analysis methods. The discretization methods include finite difference method and finite element method, which divide the entire pressing cavity space into uniform grid units or high-precision adaptive units, each unit is assigned initial material parameters, and the state variables such as binder concentration, pressure, density and deformation are solved simultaneously in the full time domain. The material rheology model can dynamically simulate the evolution of the physical state of the material at any position and any time during the pressing process, including key processes such as the non-uniform migration of the binder in space, the local densification of the powder, and the global density gradient. On this basis, the actual pressing mold is digitally modeled using finite element structural analysis software, and the stress and density abnormality prone areas inside the cavity, corners, and extreme deformation areas are identified through mechanical simulation, and these areas are selected as target monitoring points. Sensor installation holes are reserved on the mold surface or appropriate structural nodes, and high-precision pressure sensors are reasonably arranged according to spatial uniformity and process sensitivity to form a multi-point pressure monitoring network covering the entire cavity.To ensure the stability of signals in a complex industrial environment with high temperature and multiple interferences, all pressure sensors need to be equipped with signal conditioning circuits, including pre-amplification, low-pass filtering, digital isolation, and temperature compensation units, to effectively suppress non-ideal factors such as environmental noise, electromagnetic interference, and thermal drift. The signal outputs of multiple sensors are uniformly collected and scheduled through a multiplexer and are all connected to an industrial-grade embedded controller. This controller uses a high-performance ARM or RISC architecture and has a real-time operating system, capable of achieving a data processing cycle in milliseconds. Before data collection, the system automatically performs zero calibration and full-scale verification of the sensor array to ensure the consistency and accuracy of the pressure data at all monitoring points. During the entire pressing process, the embedded controller synchronously collects the real-time pressure values at each monitoring point at a high sampling frequency and conducts real-time screening and early warning on data such as abnormal fluctuations and drift signals. All target pressure data are stored digitally and provide real-time data input for the material rheological model and the adaptive control module, realizing a high-resolution data closed-loop feedback for the entire pressing process.

[0025] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Set the state equation and the observation equation based on the target pressure data, where the state vector includes the pressure values and their change rates at each measurement point, and the observation vector is the measured value of the sensor, to obtain the Kalman filter model; Execute the calculation of the Kalman filter model in two stages of prediction and update, calculate the prior state estimate and the prior error covariance, and then calculate the Kalman gain, the posterior state estimate, and the posterior error covariance to obtain the filtered pressure state data; Dynamically adjust the noise covariance matrix for the filtered pressure state data to obtain the adaptive Kalman filter result, and construct a pressure distribution feature vector based on the adaptive Kalman filter result; Import the target pressure data into the material rheological model, establish the initial state of material flow according to the relationship between pressure and displacement, and obtain the input parameters for flow analysis; Perform three-dimensional grid discretization calculation based on the input parameters for flow analysis, solve the changes of the binder concentration and the graphite powder density over time on each grid cell, and obtain the density distribution data; Calculate the slope, curvature, and inflection point position parameters of the displacement-pressure curve based on the density distribution data, and predict the material distribution state at the next time step to obtain the material flow state prediction result.

[0026] Specifically, a data dynamic processing model with multi-point pressure monitoring as the core is constructed. According to the actual collected pressure data at each monitoring point in the pressing process, combined with the dynamic characteristics of the process flow, the state equation and observation equation are set. In this model, the state vector contains the pressure value of each pressure measuring point at the current moment, as well as the rate of change of the pressure at each measuring point, reflecting the spatial distribution and time evolution trend of the pressure field under the pressing condition. At the same time, the observation vector is composed of the measured values ​​of high-precision pressure sensors distributed in various areas of the mold, which serves as a window for the model to observe the outside world, ensuring that the external measurement and internal prediction are highly coupled. The Kalman filter algorithm is executed cyclically according to the two core stages of prediction and update. In the prediction stage, the model is based on the posterior state estimation of the previous moment, and the prior state estimation and prior error covariance of the system at this moment are deduced through the state transfer matrix. This deduction not only relies on the historical pressure and change rate, but also integrates the system's statistical estimation of process disturbances and environmental noise. After entering the update phase, the model takes the observation data collected by the sensor as input, compares the residuals between the observation and prediction in real time, calculates the Kalman gain based on this, integrates the dual information of the sensor measurement and the model prediction, obtains the posterior state estimate that is closer to the real physical state, and corrects the error covariance accordingly. In this way, the cycle iteration effectively suppresses the measurement noise and occasional anomalies, realizes the robust filtering and dynamic optimization of multi-point pressure data, and outputs high-reliability pressure state data. The noise covariance matrix is ​​dynamically adjusted during the filtering process. In the specific implementation, the model continuously monitors the statistical characteristics of the observation residual sequence. When it detects that the residual exceeds the preset threshold or abnormal jumps for multiple consecutive times, the observation noise covariance is automatically increased, so that the filter relies more on historical data and reduces the interference of occasional noise on the current estimate; when the system runs smoothly and the residual fluctuation is small, the noise covariance is appropriately reduced to enhance the model's responsiveness to the newly collected data. Through the adaptive adjustment mechanism, it is ensured that the Kalman filter algorithm always maintains the best sensitivity and robustness in a changeable and complex industrial environment. Based on the results of dynamic adaptive Kalman filtering, the high-order information such as the pressure mean, standard deviation, extreme value, gradient, etc. in different spatial regions are statistically analyzed, and the pressure change characteristics of each key position are comprehensively extracted to construct a high-dimensional pressure distribution feature vector reflecting the state of the entire pressing cavity. The target pressure data, filtered state vector, etc. are imported into the aforementioned material rheology model. According to the real-time pressure distribution and mold displacement measurement values ​​of the process stage, the initial boundary conditions and spatial distribution parameters of the material flow are established to obtain the input parameters of the flow analysis. The material rheology model performs high-precision three-dimensional mesh discretization on the pressing cavity, dividing the entire pressing space into thousands or even tens of thousands of grid units, and each unit is assigned state parameters such as initial pressure, binder concentration, and graphite powder density.Based on the finite element method, finite difference method or other numerical calculation means, the binder migration equation and Drucker-Prager deformation equation are solved for grid cells in each time step to dynamically simulate the spatial migration and distribution of the binder, the densification behavior of the powder, and the evolution process of the density field during the pressing process, and obtain the material density distribution data of the entire space at any time. Perform volume-weighted averaging and smooth interpolation on the change process of the density of all grid cells over time to improve spatial continuity and simulation accuracy. Based on the smoothed density distribution, extract the key features of the displacement-pressure curve, such as slope, curvature, and the position of the curve inflection point. The slope reflects the material densification rate, the curvature reflects the turning point in the forming stage, and the inflection point often corresponds to key process nodes such as flow blockage, structural adjustment, or densification completion. Combine the curve features and spatial density gradients in each stage to predict the distribution trend of the material in different regions of the cavity at the next time step, and identify potential problems such as density anomalies, corner defects, or insufficient compaction in advance.

[0027] In a specific embodiment, the process of performing three-dimensional grid discretization calculation based on the flow analysis input parameters and solving for the change of the binder concentration and graphite powder density over time on each grid cell to obtain the density distribution data may specifically include the following steps: Perform three-dimensional grid division on the pressing cavity, and assign initial parameter values to the grid nodes based on the flow analysis input parameters to obtain a meshed pressing cavity model; Set the solid wall no-slip condition including the die wall surface and the pressure boundary condition in the pressing direction for the meshed pressing cavity model, and perform finite difference discretization on the rate equation based on the binder migration model to obtain a discrete equation set; Perform tensor transformation on the cohesion and internal friction angle parameters of the graphite powder based on the Drucker-Prager deformation model to obtain the local deformation calculation parameters for each grid cell; Based on the local deformation calculation parameters, solve the discrete equation set to calculate the change of the binder concentration and graphite powder density over time in each grid cell, and obtain the original calculation results for each time step; Perform volume-weighted averaging and three-dimensional interpolation processing on the original calculation results to generate density distribution data.

[0028] Specifically, according to the spatial geometric characteristics of the actual mold, the pressing cavity is spatially discretized using an automatic mesh generation algorithm, and a structured mesh or adaptive refinement mesh strategy is preferably adopted to balance the computational efficiency and the simulation accuracy of complex corner regions. During the mesh division process, attention is paid to the inner wall, corners, steps, and other regions with geometric mutations or where stress concentration is likely to occur in the material, and the mesh density is appropriately increased to improve the simulation resolution. Each mesh node serves as the basic unit for numerical calculation, and accurate state parameters are assigned at the initial moment, which is directly based on the input parameters of the flow analysis, including the initial pressure value, binder concentration, graphite powder density, etc. at each node. The physical boundary conditions are set. For all the mold wall nodes in the cavity, a no-slip condition for the rigid wall is applied, which means that during the forming process, the material cannot penetrate these walls and there is no tangential slip, conforming to the mechanical constraint of the actual mold on the material. For the boundaries in the pressing direction, such as the upper and lower punches or the side pressing plates, pressure boundary conditions or displacement boundary conditions are set, and the parameters such as the loading pressure and holding time in the actual process are mapped to the boundary nodes to ensure a high degree of coordination between the external energy and the internal state of the material. Based on the previously established binder migration model, its rate equation is processed using the finite difference discretization technique in the three-dimensional spatial grid. The continuous partial differential equation is discretized in space and time, the spatial derivative is replaced by the difference quotient of adjacent mesh nodes, and the time derivative is represented by the discrete time step, obtaining a discrete equation set suitable for numerical iteration. The state of each node in this discrete equation set is jointly determined by the local pressure gradient, permeability coefficient, real-time viscosity, and spatial material distribution, realizing the characterization of the migration behavior of the binder in the porous medium and being able to track the distribution dynamics of the binder in the pressing cavity at different process stages. At the same time, in order to describe the mechanical response of the graphite powder during the forming process, local tensor transformation is performed on the cohesion and internal friction angle of the material based on the Drucker-Prager constitutive model. Since these mechanical parameters change with spatial position and material state under complex geometry, material non-uniformity, or raw material batch fluctuations, the local cohesion tensor and internal friction angle tensor are calculated for each mesh element to ensure that the mechanical constitutive relationship can reflect the actual deformation and densification process. Based on the locally calculated deformation parameters, the binder migration and the plastic deformation process of the graphite powder are strongly coupled, and corresponding physical state equations are established for each mesh element to realize the coupled evolution of the states of each region in space. After all the discretization processes and parameter assignments are completed, the entire discrete equation set is solved in time sequence. Each numerical iteration involves updating the binder concentration and graphite powder density of all the mesh nodes as time progresses. The simulation process is advanced through explicit or implicit time step control methods, gradually deducing the evolution trajectories of material flow, penetration, deformation, and densification during the entire pressing process. As the simulation time steps accumulate continuously, the original calculation results at each moment and each spatial position are obtained, reflecting the three-dimensional distribution and spatio-temporal variation law of the material inside the pressing cavity.Since it is difficult to avoid local mutations or "blocky" distributions caused by spatial discretization in discrete grid simulations, volume-weighted averaging and three-dimensional interpolation are performed on all original calculation results. Volume-weighted averaging takes into account the volume proportion of each grid cell in the overall space, making the statistical density distribution closer to the actual material state and avoiding distortion caused by uneven grid fineness or uneven spatial distribution. The three-dimensional interpolation algorithm is used to establish a smooth and continuous data field between grid nodes, so that the density distribution shows a natural transition in the entire cavity space without obvious faults and jumps. After volume weighting and three-dimensional interpolation, density distribution data are generated.

[0029] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Construct a comprehensive evaluation function including density uniformity, strength, energy consumption, and die life based on the pressure distribution eigenvector and the prediction result of the material flow state; According to the binder fluidity and die usage status in the raw material characteristic parameter data, the weight coefficients of the comprehensive evaluation function are adjusted by fuzzy rules to obtain an adaptive weight adjustment mechanism; Execute model predictive control based on the material rheological model and the adaptive weight adjustment mechanism to obtain global compaction trajectory planning information; According to the global compaction trajectory planning information, set the state space as the current pressure distribution, displacement, and their change rates, and set the action space as the pressure adjustment amount and hold time adjustment amount for each region, and execute reinforcement learning control to obtain a local optimization control strategy; Based on the local optimization control strategy, divide the compaction process into a low-pressure pre-compaction stage, a medium-pressure main compaction stage, a high-pressure densification stage, and a hold stage, and assign corresponding pressure ranges and application times to each stage to obtain a segmented pressure curve; Locally adjust the compaction parameters at the corners of the crucible in the segmented pressure curve to obtain a local pressure compensation strategy.

[0030] Specifically, with density uniformity, finished product strength, pressing energy consumption, and die life as the main optimization objectives, a weighted comprehensive evaluation function is constructed. Density uniformity is measured by the standard deviation of density in each spatial region, reflecting the consistency of internal densification of the material; strength is generally indexed by actual test values such as flexural strength or compressive strength, directly reflecting the service performance of the finished sagger; energy consumption is reflected by the integral of power consumption per unit mass of the product during the entire pressing process; and die life depends on the maximum equivalent stress and the number of repeated actions in the high-stress area. A set of adaptive weight adjustment mechanisms based on fuzzy rules is established. For example, when it is detected that the fluidity of the binder in the raw material batch is insufficient or the ambient temperature is low, the weight of density uniformity in the comprehensive evaluation function is automatically increased to ensure the consistency of the internal structure of the product first and prevent deterioration of densification; when the cumulative pressing times of the die approach the life critical point or the wear in the high-stress area intensifies, the system dynamically increases the weight of die life, sacrificing some efficiency or energy consumption indicators moderately, and giving priority to ensuring equipment safety and global operation stability. This fuzzy decision-making mechanism adjusts the weights of each objective in real time based on the input raw material characteristic parameters and die state signals, and the preset expert knowledge and data learning model, to maximize the adaptation to the uncertainty and diversity in actual production. Based on the adaptive target weight configuration, combined with the material rheology model, the model predictive control algorithm is used to perform rolling optimization on the entire pressing process trajectory. The material rheology model takes the current pressure distribution, predicted material flow state, and the combined objective of the adjusted weight as inputs, performs finite-step time-domain rolling optimization, and calculates the optimal pressure-displacement trajectories for the current and several future steps, making the value of the comprehensive evaluation function minimum under controllable constraints. The information output by the model predictive control includes the globally optimal pressure curve trajectory, and also includes dynamic adjustment suggestions for the pressing parameters in each stage and region. After obtaining the global pressing trajectory planning information, to improve local adaptability and microscopic optimization ability, the system state space is refined into a joint expression of the current pressure distribution, displacement, and their change rates, and the action space is extended to the fine-tuning amount of pressure in each region and the adjustment amount of holding time. The enhanced learning (such as the Actor-Critic architecture) control algorithm is used to achieve real-time response and autonomous learning for regional distribution differences, abnormal fluctuations, and process bottlenecks. The enhanced learning controller can dynamically explore and solidify the optimal local adjustment strategy under continuous interaction, feedback, and adaptive reward mechanisms, and realize fine correction and real-time disturbance compensation for the global trajectory. Under the combined action of the global trajectory and the local control strategy, the pressing process is divided into four physically and technologically distinct stages: low-pressure pre-compaction stage, medium-pressure main compaction stage, high-pressure densification stage, and final holding stage.The pressure range and action time corresponding to each stage are given by the adaptive control system. For example, in the low-pressure pre-compaction stage, a relatively low pressure is used to quickly eliminate large pores and promote the preliminary distribution of materials. In the medium-pressure main compaction stage, efficient densification of the main material is achieved. In the high-pressure densification stage, extremely high pressure is used to eliminate micro-pores and achieve the restructuring of the intermolecular structure. In the pressure-holding stage, the final pressure is maintained for a long time to promote the migration of the binder and the full fusion of the interface. According to the results of local optimization control, fine-tuning pressures are assigned to different spatial regions in each stage. Especially for areas such as the corners of the crucible where the compactness is prone to be weak, an adaptive local pressure compensation strategy is formed through the precise control of the pressure amplitude, loading rate, and action duration.

[0031] In a specific embodiment, the process of performing step S105 may specifically include the following steps: Parametrically represent the segmented pressure curve to obtain the parametric expression of the pressurization path; Set a multi-objective optimization function group for the graphite crucible pressing process, including product density uniformity, product strength, unit energy consumption, and maximum equivalent stress of the mold; Input the parametric expression of the pressurization path and the multi-objective optimization function group into the improved NSGA-II algorithm, and perform population evolution calculations using adaptive crossover and mutation operators to obtain the optimization iteration results; Evaluate the optimization iteration results using the material rheological model to obtain the target candidate solution set, and perform fuzzy decision-making on the target candidate solution set to obtain the optimal combination of pressing process parameters.

[0032] Specifically, the segmented pressure curve during the entire pressing process is parametrically represented. The segmented pressure curve is divided into a low-pressure pre-compaction stage, a medium-pressure main compaction stage, a high-pressure densification stage, and a final pressure-holding stage according to actual process requirements. Each stage is parametrically described by physical quantities such as the starting pressure, the ending pressure, the duration, and the loading rate. To enable the pressurization path to adapt to dynamic adjustment and multi-region fine-tuning under complex working conditions, mathematical tools such as piecewise polynomial functions, piecewise linear interpolation, and cubic B-spline functions are used to parameterize the pressure-time trajectories of each stage, converting the continuous process curve into a finite-dimensional parameter vector. On this basis, aiming at the quality objectives and operation bottlenecks of graphite crucible forming, an optimization function group integrating multiple objectives such as density uniformity, product strength, unit energy consumption, and the maximum equivalent stress of the mold is constructed. The density uniformity is measured by the standard deviation of the density at each sampling point in the space of the final product, reflecting the consistency of internal densification of the material; the product strength sets the objective function based on mechanical test data such as the flexural strength and compressive strength of the final product; the unit energy consumption is measured by the ratio of the total energy consumption in the whole process of producing one graphite crucible to the mass of the product; and the maximum equivalent stress function of the mold is based on the principal stress or equivalent stress at the point with the maximum force on the mold in the finite element simulation, reflecting the mold safety, life, and maintenance cost. The parametric expression of the pressurization path and the multi-objective optimization function group are input into the improved NSGA-II algorithm. The improved NSGA-II algorithm is based on the multi-objective evolutionary theory. Each set of parametric pressurization paths is regarded as an individual to construct an initial population, and the population is quickly non-dominated sorted and the crowding distance is calculated to maintain the Pareto optimal diversity of the objective space. To improve the global search ability and avoid falling into local extrema, the algorithm introduces adaptive crossover and mutation operators, and dynamically adjusts the crossover probability and mutation probability according to the population diversity and optimization progress. For example, when the solution space distribution of the population tends to converge and the improvement of the objective function slows down, the mutation probability is automatically increased to enhance the ability to explore new solutions; when the excellent solutions increase rapidly and the diversity is good, the crossover probability is appropriately reduced to improve the genetic efficiency of the winning solutions. Through operator operations such as simulated binary crossover and polynomial mutation, the population continuously evolves and recombines in the multi-objective space to generate a new generation of solution sets. The population individuals of each generation are input into the material rheology model for physical simulation to evaluate their impacts on each objective function, including high-dimensional indicators such as material density distribution, energy consumption, and mold stress response. With multiple rounds of evolution and screening of NSGA-II, the algorithm outputs a set of non-dominated Pareto front candidate solutions, and these solution sets achieve an optimal balance among the core indicators such as density uniformity, strength, energy consumption, and mold life. Facing the diverse solution sets, it is difficult to directly select a single optimal solution based solely on quantitative indicators, so a fuzzy decision-making mechanism is introduced for auxiliary screening.By setting the fuzzy membership functions of each objective, the performance of different solutions in the multi-objective space is normalized and softly graded. Then, combined with the strategic preferences of the actual enterprise production, the comprehensive satisfaction is weighted or fuzzily aggregated, so as to select the optimal pressurization path parameter combination that is most suitable for the current production requirements, process equipment, and market orientation.

[0033] In a specific embodiment, the process of inputting the pressurization path parameterized expression and the multi-objective optimization function group into the improved NSGA-II algorithm and performing population evolution calculation using the adaptive crossover and mutation operators to obtain the optimization iteration result may specifically include the following steps: Randomly generate M initial solutions for the pressurization path parameterized expression, and evaluate each initial solution based on the multi-objective optimization function group to obtain the initial population. Perform fast non-dominated sorting based on the initial population to obtain the sorting result of the initial population. Select parent individuals according to the sorting result of the initial population, dynamically adjust the crossover probability according to the distribution of the current iterative population, and perform simulated binary crossover operations on the selected parent individuals to obtain offspring individuals. Dynamically adjust the mutation probability based on the difference between the fitness values of the offspring individuals and the parent individuals, and perform polynomial mutation operations on the offspring individuals to obtain the mutated offspring individuals. Merge the mutated offspring individuals and the parent individuals to form an intermediate population, and perform non-dominated sorting and crowding degree calculation on the intermediate population again. Select the next generation population through the elite selection strategy to obtain the optimization iteration result.

[0034] Specifically, to ensure that the algorithm can jump out of the local optimal trap and cover the wide diversity of the pressure application path parameter space, based on the parameterized expression of the pressure application path, methods such as random sampling, uniform Latin hypercube sampling, or Sobol sequence are used to randomly generate M representative initial solutions. Each initial solution corresponds to a set of parameter combinations such as pressure, duration, and loading rate in the low-pressure pre-compaction, medium-pressure main compaction, high-pressure densification, and pressure holding stages. All the initial solutions are uniformly composed into the first-generation population. These initial solutions are then input into the aforementioned multi-objective optimization function group one by one, and their comprehensive performance on objective functions such as product density uniformity, product strength, unit energy consumption, and maximum equivalent stress of the mold is calculated respectively. Through links such as material rheology models, finite element analysis, and process energy consumption statistics, their multi-objective fitness is obtained, forming a fitness vector that characterizes the "overall picture of advantages and disadvantages" of each solution in the multi-objective space. To establish an effective Pareto ranking structure in the multi-objective space, fast non-dominated sorting is performed on the initial population. The algorithm sequentially examines whether there is a relationship of "completely superior to" or "partially superior to" between the multi-objective fitness vectors of each pair of individuals, and determines all solutions that are not dominated by other individuals as the first non-dominated rank (i.e., the Pareto front). The remaining individuals are recursively compared in turn and divided into second, third, and other non-dominated ranks to achieve fine stratification of the initial population. Based on the non-dominated sorting results and the calculation of the crowding degree of individuals in the multi-objective space, according to the elitist selection strategy, parents are preferentially selected from individuals with high ranks and uniform distributions, laying a genetic foundation for the new round of population evolution. To improve the global search ability and population diversity, the algorithm dynamically and adaptively adjusts the crossover probability according to the distribution of the current iterative population in the objective space: when the diversity of the population decreases in a certain objective dimension or the optimal front changes slowly, the crossover probability is automatically increased to enhance the exploration ability of new combinations; otherwise, it is appropriately decreased to strengthen the genetic transmission of excellent genes. The selected parent individuals are subjected to simulated binary crossover operations, and two or more groups of parent parameter vectors are mixed and recombined in a probability-controlled manner to generate a new generation of offspring individuals, enabling innovation and optimization of the process parameter combinations in the global range and significantly enhancing the distribution diversity of the Pareto front. Based on the difference between the fitness values of the offspring individuals and the fitness values of the parent individuals. To prevent premature convergence and maintain the active exploration of the population, the polynomial mutation probability is dynamically adjusted accordingly. If the fitness of the current population offspring is highly close to that of the parent and the overall optimization stagnates, the mutation probability is automatically increased to cause more parameters to be perturbed globally, thereby jumping out of the local optimal trap; if the fitness of the offspring is generally superior to that of the parent or the diversity of the objective space increases significantly, the mutation probability is appropriately decreased to steadily promote the continuous expansion of the Pareto front. The polynomial mutation operation perturbs the offspring parameters with a limited amplitude and distribution control, enabling it to expand new regions based on the original excellent solutions and promoting the balanced traversal and solution set refinement of the algorithm in the complex multi-objective space.All the offspring individuals after mutation processing are merged with the original parent individuals to form the intermediate population of the current generation. The algorithm performs non-dominated sorting and crowding degree calculation on this intermediate population again, quantitatively analyzes the superiority and inferiority levels and distribution density of the solution set in the objective space, and based on this, uses the elite selection strategy to preferentially screen from individuals with high non-dominated ranks and large crowding degrees to construct the next-generation evolutionary population. Elite selection ensures the dual improvement of the optimal solution and distribution uniformity, enabling the algorithm to always aggregate, expand, and continuously explore potential solution domains along the Pareto front in the multi-objective space as the number of generations progresses, and finally output a set of high-quality solution sets on the non-dominated Pareto front under the specified number of evolutionary generations, convergence conditions, or actual engineering tolerances.

[0035] The optimization method for the graphite crucible pressing and forming process in the embodiments of the present invention has been described above. Next, the optimization system for the graphite crucible pressing and forming process in the embodiments of the present invention will be described. Please refer to Figure 2 , an embodiment of the optimization system for the graphite crucible pressing and forming process in the embodiments of the present invention includes: A measurement module 201, configured to measure and analyze the particle size distribution, specific surface area, porosity, and viscosity-temperature characteristics of the graphite powder and the binder to obtain raw material characteristic parameter data; A modeling module 202, configured to model the binder migration behavior and the graphite powder deformation characteristics during the pressing process based on the raw material characteristic parameter data to obtain a material rheological model, and at the same time arrange a multi-point pressure sensor system on the surface of the pressing die and collect target pressure data; A prediction module 203, configured to perform Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and input the target pressure data into the material rheological model for flow state prediction to obtain a material flow state prediction result; An adjustment module 204, configured to adaptively adjust the initial pressing parameters according to the pressure distribution feature vector and the material flow state prediction result to obtain a segmented pressure curve and a local pressure compensation strategy; A multi-objective optimization module 205, configured to perform multi-objective optimization on the graphite crucible pressing path based on the segmented pressure curve and the local pressure compensation strategy to obtain an optimal combination of pressing process parameters.

[0036] Through the collaborative cooperation of the above-mentioned various components, the pressure distribution is monitored in real time through the multi-point pressure sensor system, combined with the prediction of the material rheological model, the pressing parameters of the low-density areas that are prone to appear at the corners of the sagger can be accurately identified and adjusted, effectively improving the density uniformity of the product and enhancing the quality stability of the product. Based on the raw material characteristic parameter database and the binder migration model, the system can automatically adapt to the characteristic differences of raw materials in different batches, and dynamically adjust the pressing parameters through the adaptive control algorithm to ensure the consistency of product quality. The improved Kalman filtering algorithm is used to process the pressure data, combined with the prediction of the material flow state, to achieve the precise control of the pressing process, breaking through the limitations of traditional empirical parameter settings. The multi-objective optimization method is used to simultaneously consider the four indicators of product density uniformity, strength, energy consumption, and die life, and the optimal pressurization path curve is generated through the improved NSGA-II algorithm to achieve the balance of quality, efficiency, and cost. The present invention reduces the dependence on manual experience and improves the intelligent level of the graphite sagger pressing and forming process.

[0037] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0038] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing an optimization device for the graphite sagger pressing and forming process (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0039] The above is the case. The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An optimization method for graphite sagger pressing process, characterized in that: include: The particle size distribution, specific surface area, porosity and viscosity-temperature characteristics of graphite powder and binder were measured and analyzed to obtain the raw material characteristic parameter data; Based on the raw material characteristic parameter data, the binder migration behavior and the graphite powder deformation characteristics during the pressing process are modeled to obtain a material rheological model, and a multi-point pressure sensor system is arranged on the surface of the pressing die to collect target pressure data; Performing Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and inputting the target pressure data into a material rheological model to perform flow state prediction to obtain a material flow state prediction result; Adaptively adjusting initial pressing parameters according to the pressure distribution characteristic vector and the material flow state prediction result to obtain a segmented pressure curve and a local pressure compensation strategy; Based on the segmented pressure curve and the local pressure compensation strategy, the pressing path of the graphite sagger is optimized by multiple objectives to obtain the optimal pressing process parameter combination.

2. The optimization method for graphite sagger pressing and forming process according to claim 1, characterized in that: The particle size distribution, specific surface area, porosity and viscosity-temperature characteristics of the graphite powder and the binder are measured and analyzed to obtain raw material characteristic parameter data, including: The characteristic particle size of graphite powder is measured by laser particle size analyzer to obtain the particle size distribution data of graphite powder, and the specific surface area analyzer and mercury intrusion method are used to measure the graphite powder to obtain the specific surface area value and porosity data; The adhesive is measured using a rotational viscometer to obtain adhesive viscosity-temperature relationship curve data; The thermal characteristics of the binder are analyzed using a differential scanning calorimeter to obtain melting point and solidification point data, and the binder is measured at a preset strain rate to obtain stress-strain curve data; The graphite powder particle size distribution data, the specific surface area value, the porosity data, the binder viscosity-temperature relationship curve data, the melting point and freezing point data and the stress-strain curve data, as well as the bulk density, tap density and fluidity test results of the mixture are integrated into a relational data structure to obtain raw material characteristic parameter data.

3. The optimization method for graphite sagger pressing and forming process according to claim 1, characterized in that: The method of modeling the binder migration behavior and graphite powder deformation characteristics during the pressing process based on the raw material characteristic parameter data to obtain a material rheological model, and arranging a multi-point pressure sensor system on the surface of the pressing mold and collecting target pressure data, includes: Calculating the permeability coefficient of the graphite powder according to the porosity data in the raw material characteristic parameter data, and establishing a rate equation for the flow of the binder from the high-pressure area to the low-pressure area under pressure according to the permeability coefficient to obtain a binder migration model; Calculating cohesion and internal friction angle parameters according to the bulk density and tap density in the raw material characteristic parameter data, and modeling the deformation characteristics of the graphite powder according to the cohesion and the internal friction angle parameters to obtain a Drucker-Prager deformation model; Discretizing and solving the binder migration model and the Drucker-Prager deformation model to obtain a material rheology model; The target monitoring points in the pressing die are determined by finite element analysis, and multiple pressure sensors are arranged on the die surface to obtain a pressure monitoring network; Connecting the pressure monitoring network to a signal conditioning circuit and collecting the signals to an embedded controller through a multiplexer to obtain a multi-point pressure sensor system; The multi-point pressure sensor system is used to collect real-time pressure values ​​of each monitoring point during the pressing process, and target pressure data is obtained through zero-point calibration and full-scale calibration.

4. The optimization method for graphite sagger pressing and forming process according to claim 3, characterized in that: The Kalman filter processing is performed on the target pressure data to obtain a pressure distribution feature vector, and the target pressure data is input into a material rheological model to perform flow state prediction to obtain a material flow state prediction result, including: Based on the target pressure data, a state equation and an observation equation are set, wherein the state vector includes the pressure value of each measuring point and its change rate, and the observation vector is the actual value measured by the sensor, to obtain a Kalman filter model; The Kalman filter model is calculated in two stages: prediction and update, a priori state estimation and a priori error covariance are calculated, and then the Kalman gain, a posteriori state estimation and a posteriori error covariance are calculated to obtain filtered pressure state data; Dynamically adjusting the noise covariance matrix of the filtered pressure state data to obtain an adaptive Kalman filter result, and constructing a pressure distribution feature vector based on the adaptive Kalman filter result; Importing the target pressure data into the material rheological model, establishing the initial state of material flow according to the relationship between pressure and displacement, and obtaining flow analysis input parameters; Perform three-dimensional grid discretization calculation based on the flow analysis input parameters, solve the change of binder concentration and graphite powder density over time on each grid unit, and obtain density distribution data; The slope, curvature and inflection point position parameters of the displacement-pressure curve are calculated according to the density distribution data, and the material distribution state at the next time step is predicted to obtain a material flow state prediction result.

5. The optimization method for graphite sagger pressing and forming process according to claim 4, characterized in that: The three-dimensional grid discretization calculation is performed based on the flow analysis input parameters, and the change of the binder concentration and the graphite powder density over time is solved on each grid unit to obtain density distribution data, including: Performing three-dimensional meshing on the pressing cavity, and assigning initial parameter values ​​to the grid nodes based on the flow analysis input parameters, to obtain a gridded pressing cavity model; The gridded pressing cavity model is set with a solid wall no-slip condition on the mold wall and a pressure boundary condition in the pressing direction, and the rate equation is discretized by finite difference based on the binder migration model to obtain a discrete equation group; Based on the Drucker-Prager deformation model, the cohesion and internal friction angle parameters of the graphite powder are converted into tensors to obtain the local deformation calculation parameters of each grid unit; Based on the local deformation calculation parameters, the discrete equation group is solved to calculate the changes of the binder concentration and the graphite powder density in each grid unit over time, and the original calculation results of each time step are obtained; The original calculation results are processed by volume weighted averaging and three-dimensional interpolation to generate density distribution data.

6. The optimization method for graphite sagger pressing and forming process according to claim 1, characterized in that: The method of adaptively adjusting the initial pressing parameters according to the pressure distribution characteristic vector and the material flow state prediction result to obtain a segmented pressure curve and a local pressure compensation strategy includes: Constructing a comprehensive evaluation function including density uniformity, strength, energy consumption and mold life based on the pressure distribution characteristic vector and the material flow state prediction result; According to the binder fluidity and mold usage status in the raw material characteristic parameter data, the weight coefficient of the comprehensive evaluation function is adjusted by fuzzy rules to obtain an adaptive weight adjustment mechanism; Execute model predictive control based on the material rheological model and the adaptive weight adjustment mechanism to obtain global pressing trajectory planning information; According to the global pressing trajectory planning information, the state space is set to the current pressure distribution and displacement and their change rate, the action space is set to the pressure adjustment amount and the pressure holding time adjustment amount of each area, and the reinforcement learning control is performed to obtain the local optimization control strategy; Based on the local optimization control strategy, the pressing process is divided into a low-pressure pre-compaction stage, a medium-pressure main compaction stage, a high-pressure densification stage and a pressure holding stage, and a corresponding pressure range and application time are allocated to each stage to obtain a segmented pressure curve; The pressing parameters of the corners of the sagger in the segmented pressure curve are locally adjusted to obtain a local pressure compensation strategy.

7. The optimization method for graphite sagger pressing and forming process according to claim 1, characterized in that: Based on the segmented pressure curve and the local pressure compensation strategy, the graphite sagger pressing path is optimized by multiple objectives to obtain the optimal pressing process parameter combination, including: Parameterizing the segmented pressure curve to obtain a parameterized expression of a pressurization path; The graphite sagger pressing process is set up with a multi-objective optimization function group including product density uniformity, product strength, unit energy consumption and maximum equivalent stress of the mold; Inputting the parameterized expression of the pressurization path and the multi-objective optimization function group into the improved NSGA-II algorithm, and performing population evolution calculation using adaptive crossover and mutation operators to obtain an optimization iteration result; The optimization iteration result is evaluated using the material rheology model to obtain a target candidate solution set, and fuzzy decision making is performed on the target candidate solution set to obtain an optimal pressing process parameter combination.

8. The optimization method for graphite sagger pressing and forming process according to claim 7, characterized in that: The parameterized expression of the pressurization path and the multi-objective optimization function group are input into the improved NSGA-II algorithm, and the population evolution calculation is performed using adaptive crossover and mutation operators to obtain the optimization iteration result, including: Randomly generate M initial solutions from the parameterized expression of the pressurization path, and evaluate each initial solution based on the multi-objective optimization function group to obtain an initial population; Performing a fast non-dominated sort based on the initial population to obtain a sorting result of the initial population; Selecting a parent individual according to the sorting result of the initial population, dynamically adjusting the crossover probability according to the distribution of the current iterative population, and performing a simulated binary crossover operation on the selected parent individual to obtain a child individual; Based on the difference between the fitness value of the offspring individual and the fitness value of the parent individual, dynamically adjust the mutation probability, and perform a polynomial mutation operation on the offspring individual to obtain a mutated offspring individual; The mutated offspring individuals are merged with the parent individuals to form an intermediate population, and the non-dominated sorting and crowding degree calculation are performed again on the intermediate population. The next generation population is screened out through the elite selection strategy to obtain an optimized iterative result.

9. An optimization system for graphite sagger pressing process, characterized in that: Used to implement the optimization method for a graphite sagger pressing process according to any one of claims 1 to 8, the optimization system for a graphite sagger pressing process comprises: The measurement module is used to measure and analyze the particle size distribution, specific surface area, porosity and viscosity-temperature characteristics of graphite powder and binder to obtain raw material characteristic parameter data; A modeling module is used to model the migration behavior of the binder and the deformation characteristics of the graphite powder during the pressing process based on the raw material characteristic parameter data to obtain a material rheological model, and at the same time, a multi-point pressure sensor system is arranged on the surface of the pressing mold to collect target pressure data; A prediction module, used for performing Kalman filtering on the target pressure data to obtain a pressure distribution feature vector, and inputting the target pressure data into a material rheological model to perform flow state prediction to obtain a material flow state prediction result; An adjustment module, configured to adaptively adjust initial pressing parameters according to the pressure distribution characteristic vector and the material flow state prediction result, to obtain a segmented pressure curve and a local pressure compensation strategy; The multi-objective optimization module is used to perform multi-objective optimization on the pressing path of the graphite sagger based on the segmented pressure curve and the local pressure compensation strategy to obtain the optimal pressing process parameter combination.

Images