Doping control method of mixed roasting material based on large model

Abstract

The invention discloses a mixing control method for a mixed roasting material based on a large model, and the method comprises the steps: obtaining a paste dispersion state image collected by a camera, and obtaining a time queue of stirring speed values collected by a rotating speed sensor; carrying out semantic coding on the paste dispersion state image by adopting an image analysis and data analysis technology based on AI (Artificial Intelligence) and a large model, and simultaneously carrying out time sequence correlation on the stirring speed value; therefore, intelligent optimization of stirring speed control parameters is realized according to semantic flow field fine-grained alignment response representation between the paste dispersed state image semantic coding features and the stirring speed time sequence correlation features. Therefore, the time sequence change in the dispersion process can be monitored in real time, and the stirring speed can be quickly and adaptively adjusted according to the time sequence change, so that the optimal dispersion effect can be maintained, and the stability and reliability of the production process can be ensured even in a changeable industrial production environment.

Description

Technical Field

[0001] This application relates to the field of intelligent control, and more specifically, to a doping control method for mixed calcined materials based on large models. Background Art

[0002] Mixed calcined materials are widely used in many industrial fields such as metallurgy, chemical engineering, and materials science, especially in the preparation of high-performance composite materials. Such materials are usually composed of multiple metal oxides (such as zinc oxide, manganese oxide, titanium oxide) and rare earth elements (such as lanthanum, cerium), and their physical and chemical properties are crucial for the performance of the final product.

[0003] High-energy dispersion is a key step in the preparation of mixed calcined materials, which ensures the uniform distribution of each component in the mixture at the micro level. This process is particularly important for the production process of rare earth antibacterial / purification materials, because the material properties largely depend on the uniformity of its components.

[0004] Traditional dispersion control methods rely on simple mechanical stirring and fixed process parameters, lack real-time monitoring capabilities, and cannot respond in time to the temporal dynamic changes during the dispersion process, which easily leads to unstable dispersion effects and affects the quality of the final product. In addition, due to the lack of high-precision monitoring means, traditional methods are difficult to capture the uniformity at the micro level, especially in high-performance composite materials (such as rare earth antibacterial / purification materials) that require high-precision dispersion, and problems of local composition non-uniformity are likely to occur, seriously affecting the material properties. These limitations not only increase the production difficulty but also may lead to a decline in product performance and cannot meet the application requirements with extremely high requirements for composition uniformity.

[0005] Therefore, an optimized doping control scheme for mixed calcined materials is desired. Summary of the Invention

[0006] To solve the above technical problems, this application is proposed. Embodiments of this application provide a doping control method for mixed calcined materials based on large models, which acquires the images of the paste dispersion state collected by a camera, and acquires the time queue of the stirring speed values collected by a rotational speed sensor, and uses image analysis and data analysis techniques based on AI and large models to perform semantic encoding on the images of the paste dispersion state, and at the same time perform temporal correlation of the stirring speed values, so as to realize the intelligent optimization of the stirring speed control parameters according to the semantic flow field fine-grained alignment response representation between the semantic encoding features of the paste dispersion state images and the temporal correlation features of the stirring speed. In this way, it is possible to monitor the time series changes during the dispersion process in real time and quickly and adaptively adjust the stirring rate accordingly, which helps to maintain the best dispersion effect and ensure the stability and reliability of the production process even in a changing industrial production environment.

[0007] According to one aspect of the present application, a doping control method for a mixed calcined material based on a large model is provided, which includes: S1: Configure a mixed calcined material, which is formed by mixing a mixed metal oxide containing zinc oxide / manganese oxide / titanium oxide with a rare earth mixture containing lanthanum / cerium.

[0008] S2: Mechanically disperse the mixed calcined material to obtain a paste-like mixture.

[0009] S3: Perform high-energy dispersion on the paste-like mixture to obtain a suspension material with solid-liquid separation.

[0010] S4: Mechanically disperse the suspension material with solid-liquid separation to obtain a paste-like mixture.

[0011] Compared with the prior art, a doping control method for a mixed calcined material based on a large model provided by the present application acquires an image of the dispersion state of the paste collected by a camera, and acquires a time queue of the stirring speed values collected by a rotational speed sensor, and uses image analysis and data analysis techniques based on AI and large models to perform semantic encoding on the image of the dispersion state of the paste, and at the same time perform temporal correlation of the stirring speed values, so as to realize intelligent optimization of the stirring speed control parameters according to the semantic flow field fine-grained alignment response representation between the semantic encoding features of the paste dispersion state image and the temporal correlation features of the stirring speed. In this way, it is possible to monitor the time series changes during the dispersion process in real time, and accordingly quickly and adaptively adjust the stirring rate, which helps to maintain the best dispersion effect and ensure the stability and reliability of the production process even in a changing industrial production environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 It is a flowchart of a doping control method for a mixed calcined material based on a large model according to an embodiment of the present application.

[0014] Figure 2 It is a schematic diagram of data flow of a doping control method for a mixed calcined material based on a large model according to an embodiment of the present application.

[0015] Figure 3 It is a flowchart of sub-step S3 of a doping control method for a mixed calcined material based on a large model according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0018] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0019] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0020] In the technical solution of the present application, a doping control method for a mixed roasting material based on a large model is proposed.

[0021] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0022] In the technical solution of the present application, a doping control method for a mixed roasting material based on a large model is proposed. Figure 1 It is a flowchart of a doping control method for a mixed roasting material based on a large model according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of a doping control method for a mixed roasting material based on a large model according to an embodiment of the present application. As Figure 1 and Figure 2As shown, the doping control method of the mixed roasting material based on the large model according to the embodiments of the present application includes the steps of: S1, configuring the mixed roasting material, which is formed by mixing a mixed metal oxide containing zinc oxide / manganese oxide / titanium oxide with a rare earth mixture containing lanthanum / cerium; S2, mechanically dispersing the mixed roasting material to obtain a paste-like mixture; S3, performing high-energy dispersion on the paste-like mixture to obtain a suspension material with solid-liquid separation; S4, performing mechanical dispersion on the suspension material with solid-liquid separation to obtain a paste-like mixture.

[0023] Specifically, in S1, the mixed roasting material is configured, which is formed by mixing a mixed metal oxide containing zinc oxide / manganese oxide / titanium oxide with a rare earth mixture containing lanthanum / cerium. Among them, the mass ratio of rare earth to inorganic oxide in the mixed roasting material is 3:7. In the mixed roasting material, cerium metal / lanthanum metal is added in the form of an ethanol solution of cerium nitrate / lanthanum nitrate, and zinc metal / manganese metal / titanium metal is added in the form of powder. In one example, first, in the preparation stage, it is necessary to determine the specific dosage of each component according to the specific requirements of the target product. According to the requirement of the mass ratio of 3:7, that is, the rare earth part accounts for 30% of the total weight, and the inorganic oxide accounts for 70%. Then, enter the preparation process, gradually add the diluted rare earth solution to the pre-weighed metal oxide powder, and continuously stir at the same time to ensure that all components can be fully fused to form a homogeneous paste-like mixture. During the entire mixing process, it is necessary to closely monitor the addition amount of each component and the mixing time to ensure the uniformity and stability of the mixed roasting material. Specifically, it can be verified whether the mixture reaches the expected composition ratio and uniformity through sampling analysis.

[0024] Specifically, in step S2, the mixed calcined material is mechanically dispersed to obtain a paste-like mixture. Here, the mixed calcined material is put into a high-speed dispersing device for mechanical dispersion. Among them, the time of the mechanical dispersion is 20 min - 30 min. In one example, the prepared mixed calcined material is carefully put into the high-speed dispersing device. When feeding the material, attention should be paid to avoiding pouring too much material at one time, so as not to affect the normal operation of the device or cause the material to splash. The correct method is to add it step by step in batches, and the amount of each batch of material should be appropriate so that the device can effectively process it. In addition, keep the container clean during the feeding process to prevent external impurities from mixing into the material and affecting the quality of the final product. Then, set appropriate process parameters. For this case, the key is to control the length of the mechanical dispersion time, that is, 20 minutes to 30 minutes. The selection of this time range is based on the results of experimental verification, which can not only ensure sufficient dispersion effect, but also will not cause unnecessary changes to the material properties due to excessive processing time. In addition to time, an appropriate rotation speed also needs to be set. After completing the above preparations, start the high-speed dispersing device to start the mechanical dispersion process. During this period, the device will apply strong shear force and turbulence to the mixed calcined material, prompting the particles therein to collide and rub against each other, thereby breaking the original agglomerated structure and enabling each component to be more evenly distributed throughout the system. As the dispersion time increases, the material gradually turns into a paste-like mixture with fluidity.

[0025] Specifically, in step S3, the paste-like mixture is highly energy-dispersed to obtain a suspension material with solid-liquid separation. In a specific example of the present application, as Figure 3 shown, step S3 includes: S31, obtaining an image of the dispersion state of the paste collected by a camera; S32, obtaining a time queue of the stirring speed values collected by a rotation speed sensor; S33, extracting the encoded features of the paste dispersion state image from the paste dispersion state image; S34, extracting the temporal features of the stirring speed from the time queue of the stirring speed values; S35, inputting the encoded features of the paste dispersion state image and the temporal features of the stirring speed into a doping control optimization engine based on a large model to obtain a control instruction, and the control instruction includes the recommended stirring speed value at the next time point.

[0026] Specifically, in step S31, an image of the dispersion state of the paste collected by a camera is obtained. Among them, the image of the paste dispersion state refers to a visual representation that reflects the internal component distribution of the material during the dispersion process of the mixed calcined material, which shows the paste-like mixture in a flowing state and contains information such as the relative positions between different components, particle sizes, and their aggregation degrees. In particular, considering the requirements of the application scenario, an industrial-grade camera with high resolution, fast response time, and good optical performance should be selected to clearly and intuitively see the state of the paste during the dispersion process to obtain the image of the paste dispersion state.

[0027] Specifically, in S32, obtain the time queue of the stirring speed values collected by the rotational speed sensor. Among them, the time queue of the stirring speed values provides important information about the real-time operating state of the stirring device. In particular, considering the requirements of the application scenario, an industrial-grade rotational speed sensor with high precision, fast response time, and good stability should be selected. Common types include optical encoders, Hall effect sensors, or magnetoelectric sensors, etc. These sensors can accurately measure the rotational speed of the motor shaft or other rotating components and convert the data into electrical signals for output. The sensor should be installed at key positions of the high-speed dispersion equipment, such as directly connected to the stirring shaft or near the transmission mechanism, to ensure that the changes in the stirring speed can be accurately captured. It should be understood that the time queue of the stirring speed values provides important information about the real-time operating state of the stirring device.

[0028] Specifically, in S33, the paste dispersion state image encoding features are extracted from the paste dispersion state image. In the technical solution of this application, the paste dispersion state image is input into a dispersion state feature extractor based on the Mobile-Former model to obtain a paste dispersion state image semantic encoding feature vector as the paste dispersion state image encoding features. Considering that the paste dispersion state image exhibits multiple levels of dispersion characteristics. This includes the specific morphology of particles in local regions (such as circular or irregular shapes), local changes in color, and the uniformity and consistency of the dispersion state observed from an overall perspective. Therefore, in order to simultaneously capture and refine the dispersion state features at different scales and better understand the semantic relevance between different regions, in the technical solution of this application, the paste dispersion state image is input into a dispersion state feature extractor based on the Mobile-Former model to capture the local image feature information and the semantic dependencies between individual local image regions, obtaining a paste dispersion state image semantic encoding feature vector. It can be understood that Mobile-Former is a hybrid model that combines the advantages of convolutional neural networks (CNNs) and Transformer architectures, designed specifically for mobile devices and other resource-constrained environments. It achieves powerful feature extraction capabilities while maintaining efficient computation by introducing a lightweight structure. Specifically, the convolutional layer is used to capture local features in the paste dispersion state image, such as edges, textures, etc., which are crucial for identifying the specific morphology of the paste. The Transformer layer is good at handling global dependencies and helps to understand the complex dispersion patterns and long-range spatial correlations in the image. Therefore, rich semantic information can be extracted by Mobile-Former and encoded into feature vectors. These feature vectors not only represent the content of the image but also implicitly contain key semantic information about the paste dispersion state, such as particle size distribution, aggregation degree, etc.

[0029] Specifically, in S34, the stirring speed time series features are extracted from the time queue of the stirring speed values. In the technical solution of the present application, the time queue of the stirring speed values is input into the stirring speed time series encoder based on the RNN model to obtain the stirring speed time series correlation feature encoding vector as the stirring speed time series features. Considering that the stirring speed values have time series features in the time dimension, such as continuous increase within a short time interval or periodic time series changes, etc., and there are mutual correlation relationships within each local time period. And considering that the RNN is particularly good at processing data with time series properties and can effectively capture the changing trends and patterns of the stirring speed over time. Based on this, in the technical solution of the present application, the time queue of the stirring speed values is processed using the stirring speed time series encoder based on the RNN model to capture and extract the halving speed time series change patterns within different local times, obtaining the stirring speed time series correlation feature encoding vector, so as to better understand the dynamic behaviors during the dispersion process, such as acceleration, deceleration, and steady state, etc.

[0030] Specifically, in S35, the paste dispersion state image encoding features and the stirring speed time series features are input into the doping control optimization engine based on the large model to obtain the control instruction, and the control instruction includes the recommended stirring speed value at the next time point. In a specific example of the present application, S35 includes: S351, performing semantic flow field modulation feature response on the paste dispersion state image semantic encoding feature vector and the stirring speed time series correlation feature encoding vector to obtain the dispersion state - stirring speed fine - grained alignment response encoding vector; S352, adding a prompt word to the tail of the dispersion state - stirring speed fine - grained alignment response encoding vector to obtain the question description, and the prompt word is "optimize the stirring speed control parameters based on the paste dispersion state features and stirring time series features"; S353, inputting the question description into the doping control optimization engine based on the large model to obtain the control instruction.

[0031] More specifically, in S351, a semantic flow field modulation feature response is performed on the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed to obtain a fine-grained alignment response encoding vector of the dispersion state - stirring speed. Considering that the semantic encoding features of the paste dispersion state image reflect the details of the paste dispersion state and provide static image information. The temporal correlation features of the stirring speed capture the changing trend and correlation of the stirring speed over time, providing a kinetic view of the dispersion process and revealing the dynamic changes in the time dimension. Directly fusing these features may lead to information loss or inaccurate representation. Therefore, in the technical solution of this application, a semantic flow field modulation feature response mechanism is introduced to process the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed to obtain a fine-grained alignment response encoding vector of the dispersion state - stirring speed. In particular, the semantic flow field modulation feature response mechanism ensures accurate matching of features extracted even in different contexts and captures the subtle connections between features by creating a semantic flow field and implementing a feature alignment strategy, thereby enhancing the quality and representational ability of the fused features.

[0032] Specifically, the specific process of performing a semantic flow field modulation feature response on the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed includes: First, construct a dispersion state - stirring speed semantic flow field between the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed. That is, by constructing a dispersion state - stirring speed semantic flow field between the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed, to capture the dispersion state - stirring speed semantic relationship space between the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed. In a specific implementation, the process of constructing the semantic flow field between the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed generally includes dimension modulation and multi-scale convolution operations to extract rich context information and construct a comprehensive flow field reflecting the semantic association between features. More specifically, the following semantic flow field construction formula is used to construct the dispersion state - stirring speed semantic flow field between the semantic encoding feature vector of the paste dispersion state image and the temporal correlation feature encoding vector of the stirring speed; where the semantic flow field construction formula is; ; where, is the semantic encoding feature vector of the paste dispersion state image, is the temporal correlation feature encoding vector of the stirring speed, is point convolution encoding, is the activation function, is the semantic encoding activation feature vector of the paste dispersion state image, is the stirring speed time series correlation activation feature encoding vector, is the transposed vector of, is matrix multiplication, is the length of, is the convolutional encoding with a convolutional kernel of 5x5, is the convolutional encoding with a convolutional kernel of 3×3, is the upsampling operation, is the dispersion state - stirring speed semantic flow field.

[0033] Next, based on the dispersion state - stirring speed semantic flow field, feature alignment is performed on the semantic encoding feature vector of the paste dispersion state image and the time series correlation feature encoding vector of the stirring speed to obtain the aligned semantic encoding feature vector of the paste dispersion state image and the aligned time series correlation feature encoding vector of the stirring speed. In this process, the goal is to enable features from different sources or modalities to be precisely matched within the same semantic space. Through feature alignment, the positions and directions of the semantic encoding feature vector of the paste dispersion state image and the time series correlation feature encoding vector of the stirring speed can be dynamically adjusted, so that even in the presence of local deformations or scale changes, fine - grained feature matching can be achieved. More specifically, based on the dispersion state - stirring speed semantic flow field, the following feature alignment formula is used to perform feature alignment on the semantic encoding feature vector of the paste dispersion state image and the time series correlation feature encoding vector of the stirring speed to obtain the aligned semantic encoding feature vector of the paste dispersion state image and the aligned time series correlation feature encoding vector of the stirring speed; where the feature alignment formula is: ; where, is the aligned semantic encoding feature vector of the paste dispersion state image, is the aligned time series correlation feature encoding vector of the stirring speed.

[0034] Furthermore, perform fine-grained response interaction on the semantic encoding feature vector of the aligned paste dispersion state image and the time-series correlation feature encoding vector of the aligned stirring speed to obtain the fine-grained alignment response encoding vector of the dispersion state - stirring speed. In an embodiment of the present application, first, perform a linear transformation on the semantic encoding feature vector of the aligned paste dispersion state image to obtain a paste dispersion state query vector and a paste dispersion state value vector; then, perform a linear transformation on the time-series correlation feature encoding vector of the aligned stirring speed to obtain a stirring speed key vector. In this process, the linear transformation is used here to remap the feature space, so that different parts of the semantic encoding feature vector of the aligned paste dispersion state image can better serve specific tasks, such as querying, keying, or value passing. The paste dispersion state query vector and the paste dispersion state value vector are used to locate the paste dispersion state features, while the value vector carries the specific paste dispersion information content. Subsequently, input the paste dispersion state query vector, the paste dispersion state value vector, and the stirring speed key vector into a fine-grained response encoding module based on a transformer structure to obtain the fine-grained alignment response encoding vector of the dispersion state - stirring speed. Correspondingly, through the multi-head self-attention mechanism and the feed-forward neural network, the transformer structure can handle long-range dependencies without relying on the sequence order and model the interaction between features. In the fine-grained response encoding module, the similarity calculation between the paste dispersion state query vector and the stirring speed key vector guides the allocation of attention weights, thereby determining which parts of the value vector should be emphasized or suppressed. After multiple stacked attention mechanisms and non-linear transformations, the fine-grained alignment response encoding vector of the dispersion state - stirring speed is obtained. More specifically, perform fine-grained response interaction on the semantic encoding feature vector of the aligned paste dispersion state image and the time-series correlation feature encoding vector of the aligned stirring speed with the following fine-grained response interaction formula to obtain the fine-grained alignment response encoding vector of the dispersion state - stirring speed; where the fine-grained response interaction formula is: ; where and are the paste dispersion state query embedding matrix and the paste dispersion state query bias vector respectively, is the paste dispersion state query vector, and are the paste dispersion state value embedding matrix and the paste dispersion state value bias vector respectively, is the paste dispersion state value vector, and are the stirring speed key embedding matrix and the stirring speed key bias vector respectively, is the stirring speed key vector, is the transposed vector of is the length of is a function is the fine-grained alignment response coding vector of the dispersion state - stirring speed

[0035] More specifically, in S352, a prompt is added to the tail of the fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain a question description. The prompt is "Optimize the stirring speed control parameters based on the dispersion state characteristics of the paste and the stirring time series characteristics". In order to clearly indicate the direction of subsequent processing and ensure that the system can propose specific optimization suggestions based on the results of comprehensive analysis, in the technical solution of this application, a prompt is added to the tail of the fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain a question description. The prompt is "Optimize the stirring speed control parameters based on the dispersion state characteristics of the paste and the stirring time series characteristics". In this way, the system can provide more targeted optimization suggestions and ensure that the proposed control strategy is the best choice based on the current dispersion state.

[0036] Specifically, in the technical solution of this application, the dispersion state image semantic coding feature vector of the paste and the stirring speed time series correlation feature coding vector respectively represent the dispersion state image semantic coding feature of the paste dispersion state image and the implicit correlation feature of the stirring speed time series. When performing feature interaction response based on semantic flow field modulation between them, the difference in feature modality and feature expression between the dispersion state image semantic coding feature vector of the paste and the stirring speed time series correlation feature coding vector will cause the fine-grained coverage of the semantic flow field to be unbalanced, resulting in edge sharpening of the feature manifold of the fine-grained alignment response coding vector of the dispersion state - stirring speed, affecting the accuracy of the large model generation of the final control instruction.

[0037] Based on this, adding a prompt to the tail of the fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain a question description includes: optimizing the fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain an optimized fine-grained alignment response coding vector of the dispersion state - stirring speed; adding a prompt to the tail of the optimized fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain a question description.

[0038] Specifically, the process of optimizing the fine-grained alignment response coding vector of the dispersion state - stirring speed includes: applying a random perturbation operator to perform random recombination on the fine-grained alignment response coding vector of the dispersion state - stirring speed to generate a fine-grained alignment feature chaos coding tensor of the dispersion state - stirring speed, denoted as ; where is a random perturbation operator, represents the fine-grained alignment response coding vector of the dispersion state - stirring speed, represents matrix multiplication, represents the fine-grained alignment feature chaotic coding tensor of the dispersion state - stirring speed.

[0039] Calculate the geometric product operation of the fine-grained alignment response coding vector of the dispersion state - stirring speed to construct a normed linear space field coding matrix, denoted as: ; where represents the transpose vector of the fine-grained alignment response coding vector of the dispersion state - stirring speed, represents the normed linear space field coding matrix.

[0040] Perform spatial embedding on the fine-grained alignment feature chaotic coding tensor of the dispersion state - stirring speed and the normed linear space field coding matrix to generate a fine-grained alignment feature metric correlation coding vector of the dispersion state - stirring speed, denoted as: , where represents the fine-grained alignment feature metric correlation coding vector of the dispersion state - stirring speed.

[0041] Perform morphological fractal analysis on the dual tensor of the fine-grained alignment feature metric correlation coding vector of the dispersion state - stirring speed and the fine-grained alignment response coding vector of the dispersion state - stirring speed to obtain a fine-grained alignment feature non-equilibrium topological structure matrix, denoted as: ; where represents retaining the first k singular value truncated decomposition, represents the fine-grained alignment feature non-equilibrium topological structure matrix.

[0042] Perform spatial fusion on the fine-grained alignment response coding vector of the dispersion state - stirring speed and the fine-grained alignment feature non-equilibrium topological structure matrix to output an optimized fine-grained alignment response coding vector of the dispersion state - stirring speed, denoted as: ; where represents the optimized fine-grained alignment response coding vector of the dispersion state - stirring speed.

[0043] Here, by adopting a random perturbation operator to randomly reorganize the dispersed state - stirring speed fine - grained alignment response encoding vector and project it into a normed linear space constructed by tensor product, the effective quantization of the numerical association of the feature set in the unbalanced topological structure can be achieved. Based on this method system, a feature domain with an unbalanced topological structure is constructed by combining the absolute position representation of the feature vector. Relying on the vector retrieval mechanism, the dimensionality - reduced manifold of the feature vector is subjected to structural quantization within the feature domain. This strategy can avoid weakening the feature likelihood of the feature vector due to the sharpness optimization mechanism, thereby enhancing the robustness of the feature manifold boundary of the dispersed state - stirring speed fine - grained alignment response encoding vector. In this way, the precision of the large - model generation of the final control instruction is improved.

[0044] More specifically, in S353, the question description is input into the large - model - based doping control optimization engine to obtain the control instruction. Among them, the control instruction refers to specific operation suggestions or parameter adjustment commands. These instructions are mainly used to guide the stirring device on how to adjust its operating parameters (such as stirring speed) to optimize the dispersion effect and ensure the uniformity and quality of the final product. In this way, the time - series changes during the dispersion process can be monitored in real - time, and the stirring rate can be quickly adjusted accordingly to maintain the best dispersion effect, ensuring the stability and efficiency of the production process even in a changing industrial production environment. In addition, this solution can accurately identify and control the microscopic uniformity during the dispersion process, which is particularly suitable for the production of high - performance composite materials with extremely high dispersion precision requirements (such as rare - earth antibacterial or purification materials), thereby significantly improving the quality and performance of the final product.

[0045] In an example, assume that the dispersion effect is good in the initial stage, but over time, slight agglomeration occurs in some areas. At this time, the images captured by the camera show local non - uniformity, while the data recorded by the rotational speed sensor shows that the stirring speed tends to be stable. Through the above - mentioned analysis process, the system identifies this change and quickly generates a new control instruction, suggesting slightly increasing the stirring speed to break the agglomeration structure. After executing the control instruction, the dispersion effect will be significantly improved, and the material will become uniform again.

[0046] Specifically, in S4, the solid-liquid separated suspension is mechanically dispersed to obtain a paste mixture. It should be understood that even after high-energy dispersion, some fine particles may still re-aggregate to form tiny aggregates. These aggregates may cause local compositional inhomogeneity and affect the performance of the final product. In addition, in the suspension obtained after high-energy dispersion, although the solid particles are temporarily suspended, during storage or transportation, over time, the particles may gradually settle or re-aggregate, reducing their stability. Through secondary mechanical dispersion, any remaining aggregates can be further dispersed, enabling the various components to be more evenly distributed in the liquid medium. This highly uniform state is crucial for the preparation of high-performance composite materials because it ensures the uniform distribution of all components throughout the system, thereby improving the consistency and reliability of the material. In a specific example of the present application, the solid-liquid separated suspension is introduced into the high-speed dispersion equipment for mechanical dispersion, where the time of mechanical dispersion is 20 min - 30 min.

[0047] In summary, the doping control method of the mixed roasting material based on the large model according to the embodiments of the present application is elucidated. It acquires the image of the dispersion state of the paste collected by the camera and the time queue of the stirring speed values collected by the rotational speed sensor, and uses image analysis and data analysis techniques based on AI and the large model to perform semantic encoding of the image of the dispersion state of the paste, while performing temporal correlation of the stirring speed values. Based on this, the intelligent optimization of the stirring speed control parameters is achieved according to the semantic flow field fine-grained alignment response representation between the semantic encoding features of the image of the dispersion state of the paste and the temporal correlation features of the stirring speed. In this way, the time series changes during the dispersion process can be monitored in real time, and the stirring rate can be quickly and adaptively adjusted accordingly, which helps to maintain the best dispersion effect and ensure the stability and reliability of the production process even in a changing industrial production environment.

[0048] In particular, in a specific example, after the solid-liquid separated suspension is mechanically dispersed to obtain a paste mixture, the paste mixture is further placed in an oven at 120° C. for 24-48 hours to thoroughly dry the paste mixture to obtain a block mixture; then, the block mixture is placed in a crucible and placed on a conveyor, and a certain gap is controlled to be left between each crucible through a fully automatic intelligent arrangement device to form a roasting air duct, while the roasting temperature between each crucible is controlled to be consistent; in this process, the automatic conveyor first uniformly conveys the crucible in the crucible, and then the crucible is filled with a mixture of granules and a mixture of granules. The mixed material is transferred to the inner cavity of the tunnel kiln, and is automatically and continuously roasted in the tunnel kiln, and roasted in multiple temperature control zones to obtain high-temperature co-mixed doped materials; the multiple temperature zones include a roasting buffer zone, a high-temperature roasting zone, and a cooling zone; the temperature of the roasting buffer zone is controlled to rise from room temperature to 1500°C, and after 2 hours of heating and roasting, it enters the high-temperature roasting zone with a temperature of 1500°C and a high-temperature roasting time of 3 hours, and then enters the cooling zone with a temperature of 1500°C down to room temperature, and the cooling zone takes 2 hours, and then it is transferred to the external conveyor belt and waits for removal; the overall temperature control accuracy is ≤5°C. Then, the high-temperature co-mixed doped oxides that have been roasted are ground, and the block materials are first placed in the grinding equipment for crushing and initial grinding to disperse into small particles, and then different fine grinding equipment is selected according to the use of the material for secondary fine grinding to achieve the required particle size value, and the particle size range of the material is 0.15mm-200nm.

[0049] In order to verify whether the adaptive stirring rate adjustment using AI and large model technology can effectively improve the dispersion effect of the mixed roasting material, and thus improve the uniformity and quality of the final product, this application designed the following experiment. The experiment includes a control group and an experimental group, using the traditional fixed parameter stirring method and the AI-based adaptive stirring rate adjustment method respectively.

[0050] During the experimental preparation stage, the same batch of mixed roasted materials was prepared according to the recipe in the document to ensure the consistency of each group of experimental materials. The experimental device was also kept consistent, and other conditions were exactly the same except for the stirring speed control method. During the experiment, the stirring speed value at each time point, the image of the paste dispersion state captured by the camera and other data were recorded. After the high-energy dispersion was completed, the suspension was mechanically dispersed and dried, roasted and ground.

[0051] In the control group, the traditional fixed parameter method was used for decentralized processing. The experimental results are as follows:

[0052] In contrast, in the experimental group, an adaptive stirring rate adjustment method based on AI and a large model was applied. The experimental results were:

[0053] By comparing the above two sets of experimental data, it can be found that the adaptive stirring rate adjustment method can bring significant improvements in multiple aspects. First, the particle size is finer and the distribution is more concentrated. Both the average particle diameter and the standard deviation of the experimental group are better than those of the control group. Second, the dispersion uniformity score is significantly higher than that of the control group, indicating that the adaptive stirring rate adjustment method can significantly improve the dispersion uniformity. Finally, the material strength test results also show that the materials prepared in the experimental group have better performance. The material strength reaches 140 MPa, which is 15 MPa higher than that of the control group.

[0054] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, the practical application, or the improvement of the technology in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A doping control method for mixed roasted materials based on a large model, characterized in that: The method comprises the following steps: S1: preparing a mixed calcined material, wherein the mixed calcined material is formed by mixing a mixed metal oxide containing zinc oxide / manganese oxide / titanium oxide and a rare earth mixture containing lanthanum / cerium; S2: mechanically dispersing the mixed roasted material to obtain a paste-like mixture; S3: high-energy dispersing the paste-like mixture to obtain a solid-liquid separated suspension; S4: Mechanically dispersing the solid-liquid separated suspension to obtain a paste-like mixture.

2. The doping control method of the mixed roasted material based on the large model according to claim 1, characterized in that: The mass ratio of rare earth to inorganic oxide in the mixed calcined material is 3:

7. In the mixed calcined material, metal cerium / metal lanthanum is added in the form of ethanol solution of cerium nitrate / lanthanum nitrate, and metal zinc / metal manganese / metal titanium is added in the form of powder.

3. The doping control method of the mixed roasted material based on the large model according to claim 2, characterized in that: The step S2 comprises: putting the mixed calcined material into a high-speed dispersing device for mechanical dispersion, wherein the mechanical dispersion time is 20 minutes to 30 minutes.

4. The doping control method of the mixed roasted material based on the large model according to claim 2, characterized in that: The step S4 comprises: putting the solid-liquid separated suspension into the high-speed dispersing equipment for mechanical dispersion, wherein the mechanical dispersion time is 20 minutes to 30 minutes.

5. The doping control method of the mixed roasted material based on the large model according to claim 2, characterized in that: The step S3 comprises: obtaining a paste dispersion state image acquired by a camera; obtaining a time queue of stirring speed values ​​acquired by a rotation speed sensor; extracting a paste dispersion state image coding feature from the paste dispersion state image; extracting a stirring speed timing feature from the time queue of stirring speed values; and inputting the paste dispersion state image coding feature and the stirring speed timing feature into a large model-based doping control optimization engine to obtain a control instruction, wherein the control instruction comprises a recommended stirring speed value at the next time point.

6. The doping control method of the mixed roasted material based on the large model according to claim 5, characterized in that: Extracting a paste dispersion state image coding feature from the paste dispersion state image includes: inputting the paste dispersion state image into a dispersion state feature extractor based on a Mobile-Former model to obtain a paste dispersion state image semantic coding feature vector as the paste dispersion state image coding feature.

7. The doping control method of the mixed roasted material based on the large model according to claim 6, characterized in that: Extracting stirring speed time series features from the time queue of the stirring speed values ​​includes: inputting the time queue of the stirring speed values ​​into a stirring speed time series encoder based on an RNN model to obtain a stirring speed time series associated feature encoding vector as the stirring speed time series feature.

8. The doping control method of the mixed roasted material based on the large model according to claim 7, characterized in that: The paste dispersion state image coding features and the stirring speed timing features are input into a large model-based doping control optimization engine to obtain control instructions, and the control instructions include a recommended stirring speed value at the next time point, including: performing semantic flow field modulation feature response on the paste dispersion state image semantic coding feature vector and the stirring speed timing association feature coding vector to obtain a dispersion state-stirring speed fine-grained alignment response coding vector; adding a prompt word at the end of the dispersion state-stirring speed fine-grained alignment response coding vector to obtain a question description, and the prompt word is "based on the paste dispersion state characteristics and stirring timing characteristics, the stirring speed control parameters are optimized"; the question description is input into the large model-based doping control optimization engine to obtain the control instruction.

9. The doping control method of the mixed roasted material based on the large model according to claim 8, characterized in that: The semantic flow field modulation feature response of the semantic coding feature vector of the paste dispersion state image and the stirring speed time series association feature coding vector is performed to obtain a dispersion state-stirring speed fine-grained alignment response coding vector, including: constructing a dispersion state-stirring speed semantic flow field between the semantic coding feature vector of the paste dispersion state image and the stirring speed time series association feature coding vector; based on the dispersion state-stirring speed semantic flow field, feature alignment is performed on the semantic coding feature vector of the paste dispersion state image and the stirring speed time series association feature coding vector to obtain an aligned paste dispersion state image semantic coding feature vector and an aligned stirring speed time series association feature coding vector; fine-grained response interaction is performed on the aligned paste dispersion state image semantic coding feature vector and the aligned stirring speed time series association feature coding vector to obtain the dispersion state-stirring speed fine-grained alignment response coding vector.

10. The doping control method of the mixed roasted material based on the large model according to claim 9, characterized in that: Fine-grained response interaction is performed on the aligned paste dispersion state image semantic coding feature vector and the aligned stirring speed time series associated feature coding vector to obtain the dispersion state-stirring speed fine-grained alignment response coding vector, including: linearly transforming the aligned paste dispersion state image semantic coding feature vector to obtain a paste dispersion state query vector and a paste dispersion state value vector; linearly transforming the aligned stirring speed time series associated feature coding vector to obtain a stirring speed key vector; and inputting the paste dispersion state query vector, the paste dispersion state value vector and the stirring speed key vector into a fine-grained response coding module based on a converter structure to obtain the dispersion state-stirring speed fine-grained alignment response coding vector.

Images