Fire-resistant material production optimization system based on big data analysis

Abstract

The present application relates to the field of refractory material preparation and industrial automation control, and particularly to a refractory material production optimization system based on big data analysis, comprising a manifold space initialization step: constructing a thermodynamic potential energy manifold data model using raw material characteristic data; a state mapping and positioning step: constructing an observation equation and mapping real-time sintering state data to current state particle coordinates using a damped least squares method; a geodesic solution and control step: solving a second-order Euler-Lagrange differential equation set based on a target energy state position to obtain an optimal geodesic path; and a deviation correction and execution step: generating adaptive control instructions according to the vector deviation of the particle evolution trend and the optimal path to adjust equipment operating parameters; the present application effectively solves the problem of unstable sintering quality caused by raw material fluctuations in traditional production.

Description

Technical Field

[0001] This invention relates to the field of refractory material preparation and industrial automation control, specifically a refractory material production optimization system based on big data analysis. Background Technology

[0002] Temperature timing control is a traditional method in refractory sintering production. It can adjust the combustion parameters in the kiln according to a preset curve and monitor macroscopic temperature changes. It is characterized by its ease of operation and versatility.

[0003] With the improvement of automation in high-temperature industries, sintering control has gradually evolved from simple manual observation to automatic loop control based on sensor feedback. However, when dealing with fluctuations in raw material characteristics and complex microscopic phase transformation processes, existing macroscopic temperature control technologies are still limited by the hysteresis of physical fields, resulting in poor consistency in the sintering quality of refractory materials. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a refractory material production optimization system based on big data analysis. Specifically, the technical solution of this invention includes:

[0005] Production data acquisition terminal, optimization computing server, and equipment control terminal;

[0006] The production data acquisition terminal is configured to acquire raw material characteristic data and real-time sintering status data of refractory material production equipment, and send the raw material characteristic data and real-time sintering status data to the optimization calculation server.

[0007] The optimization calculation server is configured to construct a thermodynamic potential energy manifold data model based on the raw material characteristic data, and to map the real-time sintering state data to the current state particle coordinates on the thermodynamic potential energy manifold data model;

[0008] The optimization calculation server is also configured to calculate the optimal geodesic path based on the current state particle coordinates and the preset target energy state position, and generate adaptive control commands based on the vector deviation between the evolution trend of the current state particle coordinates and the optimal geodesic path.

[0009] The equipment control terminal is configured to adjust the operating parameters of the refractory material production equipment based on the adaptive control command.

[0010] Preferably, the optimized computing server is further configured as follows:

[0011] Extract the energy characterization value of the acoustic emission signal from the real-time sintering state data;

[0012] Obtain the preset fracture energy barrier reference threshold from the storage space;

[0013] The energy characterization value is compared with the fracture energy barrier reference threshold;

[0014] If the energy characterization value is greater than the fracture energy barrier reference threshold, an interruption protection control command is generated as the adaptive control command.

[0015] If the energy characterization value is less than or equal to the fracture energy barrier reference threshold, a path tracking compensation command is generated as the adaptive control command.

[0016] Preferably, the optimized computing server is further configured as follows:

[0017] The raw material feature data is feature-encoded to generate a high-dimensional feature vector;

[0018] Combining the preset phase diagram database, the high-dimensional feature vectors are used to generate a virtual phase transition potential energy data surface, which serves as the thermodynamic potential energy manifold data model.

[0019] The geometric topology of the thermodynamic potential energy manifold data model characterizes the reaction energy barrier distribution corresponding to the raw material characteristic data.

[0020] Preferably, the optimized computing server is further configured as follows:

[0021] Calculate the actual evolution velocity vector of the current state particle coordinates;

[0022] Calculate the tangent vector of the optimal geodesic path at the coordinates of the particle in the current state;

[0023] Based on the actual evolution velocity vector and the tangent vector, calculate the magnitude of the covariant derivative;

[0024] If the modulus of the covariant derivative is less than the preset convergence tolerance threshold, then the current operating parameters of the refractory material production equipment are maintained.

[0025] If the magnitude of the covariant derivative is greater than or equal to the convergence tolerance threshold, then the compensation amount of the adaptive control command is determined based on the magnitude of the covariant derivative.

[0026] Preferably, the production data acquisition terminal includes: a component analysis module and a multi-channel sensing module;

[0027] The component analysis module is configured to collect the chemical composition and particle size distribution of the raw materials before they enter the kiln, and generate the raw material characteristic data.

[0028] The multi-channel sensing module is configured to collect temperature timing data, atmosphere composition data, and acoustic emission waveform data during the firing process to generate the real-time sintering status data.

[0029] Preferably, the device control terminal is further used for:

[0030] The adaptive control command is parsed to obtain the PID parameter correction value;

[0031] The PID parameter correction value is written into the programmable logic controller of the refractory material production equipment.

[0032] The operating parameters include the gas valve opening degree, the combustion fan frequency, and the trolley running speed.

[0033] Preferably, it also includes: a remote monitoring terminal;

[0034] The remote monitoring terminal is configured to obtain the thermodynamic potential energy manifold data model and the coordinates of the current state particle from the optimization calculation server, and to visualize the motion trajectory of the current state particle on the thermodynamic potential energy manifold data model.

[0035] Preferably, the optimized computing server is deployed in an industrial edge computing gateway;

[0036] The industrial edge computing gateway interacts with the production data acquisition terminal and the equipment control terminal via the industrial bus protocol.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This system constructs a high-dimensional thermodynamic potential energy manifold data model based on raw material characteristic data and maps real-time sintering state data to the coordinates of current state particles on the manifold. This invention transforms the traditional temperature-time control dimension into a reaction path control dimension constrained by Riemannian geometry. By solving the second-order Euler-Lagrange differential equation system determined by the metric tensor field to calculate the optimal geodesic path, it effectively solves the problem of unstable sintering quality caused by raw material fluctuations and macroscopic temperature control lag in traditional refractory material production, ensuring that the material evolution proceeds along the thermodynamically optimal trajectory with the lowest energy consumption and shortest path.

[0039] 2. By introducing a micro-fracture mechanism based on the energy characterization value of acoustic emission signals, this invention achieves real-time monitoring and active intervention of the propagation of microcracks inside the material during sintering. By performing sliding window integral calculation on the acoustic emission signals and comparing it with the fracture energy barrier reference threshold obtained based on the statistics of historical fracture accident samples, the system can automatically switch between interruption protection and path tracking modes according to the energy release state in the high-temperature rapid sintering scenario that pursues density. This effectively prevents macroscopic product cracking caused by excessive thermal stress while ensuring production efficiency, and achieves a dynamic balance between production efficiency and structural integrity.

[0040] 3. This system overcomes the problems of invisible microscopic states and difficult data alignment in continuous production of industrial kilns by adopting a dataset construction method of offline physicochemical calibration and online data alignment, and manifold generation technology based on deep learning. It establishes a high-precision supervised learning dataset with physical truth constraints by using liquid nitrogen quenching sampling and RFID time synchronization technology, and combines variational autoencoders to extract potential features of raw materials and conditional generative adversarial networks to generate virtual phase transition potential energy data surfaces. This ensures that the geometric topology of the thermodynamic potential energy manifold data model can accurately characterize the reaction energy barrier distribution of a specific batch of raw materials, providing complete and multi-dimensional physical model support for upper-level control.

[0041] 4. This system achieves a precise mapping from the abstract manifold geometric space to the physical device control space by using a path tracking algorithm based on the covariant derivative modulus and an adaptive PID gain scheduling strategy. The system uses the connection and covariant derivative in Riemannian geometry to calculate and quantify the vector deviation between the actual evolution trend and the optimal geodesy. Combined with the real-time corrected Jacobian sensitivity matrix, the geometric deviation is analyzed into the PID parameter correction values ​​of actuators such as gas valves and combustion fans. Thus, high-precision closed-loop control of the reaction path is achieved in the nonlinearly strongly coupled sintering process, forcing the production state to always return to the thermodynamically optimal path. Attached Figure Description

[0042] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0043] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0045] Example 1:

[0046] Please see Figure 1The refractory material production optimization system based on big data analysis includes: a production data acquisition terminal, an optimization computing server, and an equipment control terminal;

[0047] Among them, the production data acquisition terminal is configured to acquire raw material characteristic data and real-time sintering status data of refractory material production equipment, and send the raw material characteristic data and real-time sintering status data to the optimization calculation server.

[0048] The computing server configuration is optimized to construct a thermodynamic potential energy manifold data model based on raw material characteristic data, and the real-time sintering state data is mapped to the current state particle coordinates on the thermodynamic potential energy manifold data model;

[0049] The optimization computing server is also configured to calculate the optimal geodesic path based on the current state particle coordinates and the preset target energy state position, and generate adaptive control commands based on the vector deviation between the current state particle coordinates and the optimal geodesic path.

[0050] The equipment control terminal is configured to adjust the operating parameters of the refractory material production equipment based on adaptive control commands.

[0051] This embodiment details the specific architecture definition and execution process of the above system, aiming to solve the problem of unstable sintering quality caused by raw material fluctuations and macroscopic temperature control lag in traditional refractory material production;

[0052] The system executes the manifold initialization steps; the optimization calculation server receives raw material characteristic data sent by the production data acquisition terminal. Server utilization As boundary conditions, a high-dimensional thermodynamic potential energy manifold data model is constructed. ; in this manifold In this context, each coordinate point no longer represents only temperature or time, but rather the microscopic phase transition state of the refractory material at a specific moment, such as lattice constant, liquid content, and density; the height on the manifold represents the Gibbs free energy of the system.

[0053] The system performs state mapping and positioning steps; the system receives real-time sintering status data. It includes temperature, atmosphere, and acoustic emission characteristics; to solve the problem of inverse solution from macroscopic sensor data to microscopic manifold coordinates, the computational server is optimized to construct the observation equations. ,in, To create a nonlinear regression model pre-trained using historical batch data, this embodiment employs a fully connected neural network structure, where the number of nodes in the input layer corresponds to the microstate. Dimensions In this embodiment and subsequent embodiments, a unified setting is provided. The number of output layer nodes corresponds to the real-time sintering status data. The dimension is set to a three-layer structure. The activation function used was LeakyReLU; the training set for historical batch data was constructed using an offline physicochemical calibration-online data alignment method.

[0054] To overcome the challenges of sampling and alignment in continuous production of industrial kilns, this embodiment discloses specific construction steps: Sampling method and location: A pneumatic high-speed robotic arm with a liquid nitrogen cooling chamber is installed at specially designed observation holes reserved at the end of the preheating zone, the middle of the firing zone, and the front end of the cooling zone of the tunnel kiln; Sampling timing and representativeness assurance: A high-temperature resistant RFID positioning tag is installed on the standard sample plate for testing. When the RFID reader on the production line detects that the sample has reached the observation hole position, it triggers the robotic arm to... The sample is grabbed and pulled into the liquid nitrogen chamber instantly to freeze the instantaneous microscopic phase structure of the material at high temperature, preventing secondary phase change during the cooling process, thereby ensuring that the sample can represent the real material state inside the kiln at a specific moment.

[0055] Format and measurement of microstate data: Offline analysis of quenched samples to measure microstate vectors. ,in, The spinel phase lattice constant obtained by XRD refinement is expressed in angstroms. , The volumetric density is determined using the Archimedes method, and the unit is 1. , The liquid volume fraction is calculated using grayscale analysis of SEM images, expressed as a percentage. Other components... to For extended definition of microstructure parameters;

[0056] Time synchronization and alignment: utilizing precise sampling timestamps recorded by RFID Index time windows from historical sensor database The mean values ​​of internal temperature, atmosphere, and acoustic emission data are used as the training input vector. This allows for the labeling of truth values ​​determined offline. Compared with historical sensor data Establish precise input-output alignment and construct a high-precision supervised learning dataset with physical truth constraints to describe microstates. The corresponding macroscopic physical field characteristics; the system solves a constrained nonlinear least squares problem iteratively:

[0057]

[0058] in, The sensor noise covariance matrix is ​​obtained by pre-calibrating the sensor noise floor data collected during the system initialization phase under no-load conditions and calculating its variance matrix. It is used to quantify the inherent measurement uncertainty of the hardware. For the time-series smoothing constraint coefficient, in this embodiment, a fixed constant determined through cross-validation of historical data is used, for example... To balance model fit and trajectory smoothness; to avoid iterative solutions getting stuck in an infinite loop, the system uses damped least squares method and sets a maximum number of iterations. Convergence tolerance When the Euclidean distance between the state points in two consecutive iterations When convergence is determined, the result is output; through this optimization solution, Projected onto manifold The coordinates of the current state particle are obtained from the above. ; It accurately characterized the specific location of the material in the reaction pathway;

[0059] The system performs geodesic calculation and control steps; the server sets a target energy state position. This location corresponds to the optimal compactness and crystal phase structure of the finished product, i.e., the global minimum point on the potential energy surface; the system is based on Riemannian geometry principles, specifically by solving a system of second-order Euler-Lagrange differential equations determined by the metric tensor field:

[0060]

[0061] To ensure the feasibility of the above equations on the specific technical object of this invention, namely the thermodynamic potential energy manifold, this embodiment provides the following physical definitions and mathematical constructions for the key variables: metric tensor It is defined as the height of the potential energy manifold, i.e., the Gibbs free energy. Regarding the coordinates of the microstate The Hessian matrix, i.e. This definition transforms the energy barrier curvature of chemical thermodynamics into a geometrical property, making high-energy-barrier regions geometrically appear as high-curvature, difficult-to-walk regions; Christofel notation. Based on the above metric tensor and its inverse matrix The calculation formula is as follows: Used to quantify the guiding effect of manifold curvature on the path; affine parameters Defined as a normalized reaction process, with a range of values. To guide actual production control, the system was established. With physical time mapping relationship ,in, This represents the pre-defined optimal reaction rate distribution along the path;

[0062] Based on the above definition, the system uses a two-end firing method to iteratively adjust the initial tangent vector until the trajectory satisfies the boundary conditions. and Thus, the accurate solution from arrive Optimal geodesic path Geodesic lines This represents the reaction evolution trajectory with the lowest energy consumption and shortest path under current thermodynamic constraints;

[0063] System calculation of mass point Actual evolutionary trend and optimal path The vector deviation between them; based on this deviation, an adaptive control command is generated and sent to the equipment control terminal to adjust the operating parameters of the refractory material production equipment, such as the gas flow rate and air flow rate, so as to force the mass points... Regress to the optimal path superior.

[0064] Example 2:

[0065] The optimized computing server is also configured as follows:

[0066] Extract the energy characterization value of acoustic emission signal from real-time sintering state data;

[0067] Obtain the preset fracture energy barrier reference threshold from the storage space;

[0068] The energy characterization values ​​are compared with the reference threshold of the fracture energy barrier;

[0069] If the energy characterization value is greater than the fracture energy barrier reference threshold, an interruption protection control command is generated as an adaptive control command.

[0070] If the energy characterization value is less than or equal to the fracture energy barrier reference threshold, a path tracking compensation command is generated as an adaptive control command.

[0071] This embodiment further defines the exception handling logic of the optimized computing server and introduces an acoustic emission energy criterion mechanism;

[0072] The system performs the energy characterization value extraction step; during the sintering process, the generation and propagation of microcracks inside the material release elastic waves; the optimization calculation server separates the acoustic emission signal from the real-time sintering state data and calculates its energy characterization value. The calculation formula is as follows:

[0073]

[0074] in, The source is the real-time acquisition of the multi-channel sensing module, and its physical meaning is the instantaneous voltage amplitude of the acoustic emission signal; : Defined as the start and end times of the sliding integral window; specifically, This is the current system sampling time. Let be the starting time of integration, and satisfy . ,in, The preset integration window length is set to [value] in this embodiment. This sliding window mechanism ensures that the system can capture transient crack propagation signals in real time, rather than calculating the cumulative energy over the entire process.

[0075] The system performs threshold comparison and bimodal control steps; the server pre-stores a fracture energy barrier reference threshold. To avoid the complex calibration errors in the conversion coefficients between electrical signals and physical mechanical energy, Defined as the equivalent voltage squared integral threshold obtained based on historical fracture accident sample statistics, in units of Its value is taken as the lower quartile of the energy distribution of historical fracture samples; system calculation bias ; in response to This indicates that irreversible macroscopic crack propagation is occurring within the material; at this point, the system immediately generates an interruption protection control command, such as emergency cooling or the introduction of inert gas, prioritizing the preservation of the product's structural integrity rather than continuing to pursue density; in response to This indicates that the current microstructure evolution is within a safe range; the system generates path tracing compensation instructions to continue guiding the particle evolution along the geodesic;

[0076] This embodiment establishes a microscopic melting mechanism for the sintering process; in the scenario of high-temperature rapid sintering that pursues density, this mechanism acts as a fuse for the safety of the material structure; by monitoring the release of microscopic energy in real time, it effectively prevents macroscopic product cracking caused by excessive thermal stress when pursuing the optimal path, and achieves a dynamic balance between production efficiency and structural integrity.

[0077] Example 3:

[0078] The optimized computing server is also configured as follows:

[0079] Feature encoding is performed on raw material characteristic data to generate high-dimensional feature vectors;

[0080] By combining a pre-set phase diagram database, a virtual phase transition potential energy data surface is generated using high-dimensional eigenvectors, which serves as a thermodynamic potential energy manifold data model.

[0081] The geometric topology of the thermodynamic potential energy manifold data model characterizes the reaction energy barrier distribution corresponding to the raw material characteristic data.

[0082] This embodiment describes in detail the construction process of the thermodynamic potential energy manifold data model. In view of the problem that the explanation of the core model construction method is not sufficient, specific implementation details and data sources are added here.

[0083] The system performs a high-dimensional feature vector generation step; the optimized computing server encodes the raw material feature data, including chemical composition and particle size; specifically, to address the problem of heterogeneous raw material data dimensions and nonlinear coupling, such as the nonlinear influence of trace impurities on the main crystal phase formation temperature, the system employs a variational autoencoder (VAE) to extract latent features; the system constructs the original physical parameter vector. :

[0084]

[0085] in, The source is the component analysis module, and its physical meaning is the molar concentration of each oxide component; in this embodiment, it is set according to the actual number of monitored components. These correspond to the concentration indices of the four key components: MgO, Al2O3, SiO2, and Fe2O3, respectively. The source is a particle size analyzer, and its physical meaning is the distribution probability of different particle size ranges; in this embodiment, it is set according to the number of particle size classification channels. The distribution proportions of the three characteristic particle size intervals, D10, D50, and D90, respectively;

[0086] vector Input a pre-trained VAE encoder network and output a continuous high-dimensional feature vector. Latent variables:

[0087]

[0088] To clarify, the superscript in the formula... A unified representation of the transpose operation of a vector or matrix is ​​defined; in this embodiment, it is set as follows: Dimensions To match the degree of freedom requirements of the manifold topology;

[0089] The system executes the virtual phase transition potential energy data surface generation step; combined with a preset phase diagram database, which contains... The system utilizes multi-element phase diagram data. Generate virtual phase transition potential energy data surface To enable those skilled in the art to reproduce the construction of this data surface, the specific construction method of the phase diagram database and the method for generating the training set are disclosed here: This embodiment does not directly use phase diagrams in static image format, but instead uses an offline thermodynamic calculation engine based on CALPHAD, i.e., the phase diagram calculation method, such as Thermo-Calc or FactSage API interface, as the data source; the system pre-performs gridded sampling in the multidimensional composition space and temperature space, calculates the Gibbs free energy value of each grid point under different microscopic phase structures, and forms a structured dataset. ;in, This refers to the thermodynamic truth value label;

[0090] Specifically, the system is constructed using a conditional generative adversarial network (cGAN), which contains a generator. and discriminator generator A multilayer perceptron structure is used, with high-dimensional feature vectors as input. With state space coordinates The concatenated vector; here, the state space coordinates The physical definition remains strictly consistent with that of Example 1, that is, each component These represent microscopic physical properties such as lattice constant, packing density, and liquid phase content, rather than chemical component concentrations; the hidden layer is set to 5 layers, with the following node numbers: The activation function is LeakyReLU, and the output layer has no activation function, directly outputting the scalar potential energy value. The generator After training, it is fixed into a differentiable analytic function. The continuous hypersurface formed by this function within its domain is the thermodynamic potential energy manifold data model.

[0091] Because neural networks are infinitely differentiable, this manifold satisfies the differential manifold conditions required for Riemannian geometric computation; discriminator This is used to determine whether the input potential energy value conforms to the thermodynamic laws in the phase diagram database. The input is a real phase diagram data point. Or generate data points Combining conditions The loss function outputs the probability of authenticity; during the training phase, a physical constraint term is introduced into the loss function. Defined as the first derivative continuity penalty at the state point in the critical region of phase transition, ensuring the generation of the potential energy surface. It is a smooth and continuous manifold surface; to ensure that the model output conforms to the fundamental laws of thermodynamics, physical constraint terms... The specific calculation formula is defined as follows:

[0092]

[0093] in, For example, take the gradient penalty weight coefficient. , These are sampling points in the critical region of phase transition. The theoretical chemical potential gradient vector is calculated from discrete grid point data using a trilinear interpolation algorithm from a phase diagram database. This constraint forces the potential energy surface gradient field output by the generator to be consistent with the theoretical thermodynamic gradient field, thus avoiding the generation of artifacts that violate the principle of minimizing Gibbs free energy.

[0094] The generator after training The output is a scalar field defined on the state space, namely the virtual phase transition potential energy data surface. Its geometric topology directly characterizes the reaction energy barrier distribution of the batch of raw materials in the reaction process. Among them, the peak topology corresponds to unstable high-energy states or insurmountable reaction barriers, while the valley topology corresponds to stable crystalline phase products, such as spinel phase.

[0095] Example 4:

[0096] The optimized computing server is also configured as follows:

[0097] Calculate the actual evolution velocity vector of the current state particle coordinates;

[0098] Calculate the tangent vector of the optimal geodesic path at the current state particle coordinates;

[0099] The magnitude of the covariant derivative is calculated based on the actual evolution velocity vector and tangent vector.

[0100] If the modulus of the covariant derivative is less than the preset convergence tolerance threshold, then the current operating parameters of the refractory material production equipment are maintained.

[0101] If the magnitude of the covariant derivative is greater than or equal to the convergence tolerance threshold, the compensation amount of the adaptive control command is determined based on the magnitude of the covariant derivative.

[0102] This embodiment discloses a path tracking algorithm based on covariant derivatives, which is the mathematical core for achieving high-precision control;

[0103] The system performs vector calculations; the computation server is optimized for manifold operations. Perform vector calculations on the above; calculate the actual evolution velocity vector. This vector is obtained through the formula The calculation yielded that, The preset system scan cycle is set to [value] in this embodiment. The introduction of this time parameter, seconds, ensures... It possesses the correct physical dimensions, thus accurately characterizing the actual direction and rate of the current reaction; simultaneously, it calculates the tangent vector. It is the optimal geodesic path. At the current point The tangent direction at a given point represents the ideal evolutionary direction;

[0104] The system performs a deviation assessment step; to measure whether the actual evolution deviates from a straight line, i.e., a geodesic, on the curved manifold space, the system calculates the modulus of the covariant derivative. Prior to this, the system was based on the generated virtual phase transition potential energy data surface. Constructing the metric tensor The metric tensor is defined as the Hessian matrix of the potential energy surface with respect to the state variables:

[0105]

[0106] Here, we further explain the physical dimension definitions of the key parameters to clarify the physical meaning of the metric tensor: In this embodiment, potential energy It is the dimensionless Gibbs free energy after normalization, i.e. ,in, For Gibbs free energy, The gas constant is... Sintering temperature, state coordinates It is also a normalized dimensionless parameter, such as mole fraction, relative density, etc., with a range of values. Therefore, based on the above second-order partial derivative formula, the metric tensor is calculated. As a dimensionless quantity, this ensures the dimensionality consistency and mathematical rigor of geometric quantities such as covariant derivatives in subsequent Riemannian geometric calculations;

[0107] The system calculates the inverse matrix of the metric tensor. To prevent singularities in the Hessian matrix within flat regions of the potential energy surface, Tikhonov regularization is used to modify the inversion process: ,in, As the numerical stability factor, take ; It characterizes the sensitivity or conductivity of state changes under the influence of potential energy gradient;

[0108] Based on metric tensor and its inverse matrix The system calculates the Christofel symbol. :

[0109]

[0110] Calculate the covariant derivative according to the Riemannian geometric definition. The One component:

[0111]

[0112] Among them, the rate of change term of the non-stationary tangent vector field in the formula The system uses the first-order backward difference method for numerical approximation calculations:

[0113]

[0114] Here These are the geodesic tangent vector components calculated at the current moment. The tangent vector component calculated for the previous control cycle and cached in the register is set to zero if it is the first cycle of system startup; this discretization process ensures the executability of the mathematical definition in a digital computer.

[0115] Finally, its modulus length is calculated:

[0116]

[0117] in, The origin is the Riemannian geometric definition, and its physical meaning is the Levi-Civita connection on the Riemannian manifold, which is used to correct translation errors caused by the curvature of space. The source is the differential of real-time state data, and its physical meaning is the actual evolution velocity vector; The source is the geodesic planning algorithm, and its physical meaning is the tangent vector of the optimal path;

[0118] The system executes convergence criteria and compensation steps; the system sets convergence tolerance thresholds. It is hereby clarified that the convergence tolerance threshold is... The specific numerical determination method is as follows: During the offline calibration phase of the system, the state data of the equipment under steady-state ideal operating conditions are collected, the statistical distribution of the covariant derivative modulus under this operating condition is calculated, and its mean plus 3 times the standard deviation is taken as the value. In this embodiment, the specific value is... This threshold setting ensures that the system only responds to disturbances that significantly deviate from the thermodynamically optimal path, avoiding over-adjustment of the sensor's white noise; in response to This indicates that the current reaction completely follows the thermodynamically optimal path, and the system maintains its current parameters; in response to This indicates that a deviation has occurred; at this point, the system must convert the abstract geometric deviation in the manifold space into specific control increments for the physical devices.

[0119] The system extracts the vector components of the covariant derivative. ;

[0120] Call the preset Jacobian sensitivity matrix The matrix is ​​defined as ,in, For the first 3D state manifold coordinates, For the first Controllable parameters of the equipment, such as gas valve opening and combustion fan frequency; to ensure the Jacobian matrix This embodiment can accurately reflect the characteristics of the current nonlinear system. The acquisition method is as follows: During the system initialization phase, a step response calibration experiment is performed for each control channel. Apply small perturbation ,For example Measuring the steady-state response of state variables Through the difference ratio Construct the initial Jacobian matrix During the online operation phase, the Broyden rank-1 update algorithm is used to update the... Perform real-time corrections:

[0121]

[0122] in, To prevent regularization factors with denominators of zero, take , These are the current state increment and control increment, respectively;

[0123] The system uses the damped least squares method to calculate the compensation vector of the control parameters. :

[0124]

[0125] in, To control the gain step size, its value is set to... This value was determined through offline simulation experiments, with the system overshoot being less than [a certain value]. Obtained by searching for constraints. The damping coefficient, also known as the Levenberg-Marquardt parameter, is initially set to... And during runtime, based on the residual decrease rate Dynamic adjustment; defined here The calculation formula is used to quantify the model confidence level:

[0126]

[0127] The system is based on Generate adaptive control commands, forcing It tends to 0 in subsequent time steps.

[0128] Example 5:

[0129] The production data acquisition terminal includes: a component analysis module and a multi-channel sensing module;

[0130] The component analysis module is configured to collect the chemical composition and particle size distribution of the raw materials before they enter the kiln and generate raw material characteristic data.

[0131] The multi-channel sensing module is configured to collect temperature timing data, atmosphere composition data, and acoustic emission waveform data during the firing process, and generate real-time sintering status data.

[0132] This embodiment limits the hardware configuration of the production data acquisition terminal;

[0133] The system is configured with a component analysis module; this module is deployed in the mixing section before the kiln, and mainly includes an online X-ray fluorescence analyzer and a laser particle size analyzer; it is configured to collect the chemical composition of the raw materials, such as the percentage content of Al2O3, MgO, SiO2, and Fe2O3, as well as the particle size distribution, such as particle size characteristic parameters such as D10, D50, and D90; these data together constitute the raw material characteristic data;

[0134] The system is equipped with a multi-channel sensing module; this module is deployed in the high-temperature tunnel kiln or shuttle kiln body and is configured to collect temperature time-series data, which is acquired by an S-type or B-type thermocouple array distributed in the preheating zone, firing zone, and cooling zone; and to collect atmosphere composition data, which is collected by a zirconia oxygen analyzer and an infrared gas analyzer inside the kiln. and Concentration; Acoustic emission waveform data is collected by a high-temperature acoustic emission sensor waveguide rod installed on the outer wall of the kiln to monitor micro-cracks; These data are synchronized in time to generate real-time sintering status data.

[0135] Example 6:

[0136] The equipment control terminal is also used for:

[0137] Parse the adaptive control instructions and obtain the PID parameter correction values;

[0138] Write the PID parameter correction value into the programmable logic controller of the refractory material production equipment;

[0139] The operating parameters include the opening degree of the gas valve, the frequency of the combustion fan, and the running speed of the trolley.

[0140] This embodiment defines the execution logic of the device control terminal;

[0141] The system executes instruction parsing and PID correction steps; the device control terminal receives an adaptive control instruction from the optimization calculation server, which is essentially a data packet encapsulating control metadata; the terminal parses this data packet and extracts the covariant derivative magnitude calculated by the optimization calculation server. And the Jacobian matrix calculated in Example 4. and error vector The terminal interprets it as the underlying PID parameter correction value. Specifically, the terminal integrates a gain scheduling calculation module, which does not rely on fuzzy logic but directly executes calculations based on... The analytical function is calculated; to achieve an accurate mapping from manifold geometric deviations to PID parameters, this embodiment specifies a particular gain scheduling function, which directly uses the analytically derived covariant derivative magnitude. As the independent variable, and for different control channels Among them, the speed of the gas valve combustion fan trolley Introducing adaptive inertia weight coefficient Unlike traditional fixed-weight control, this embodiment... It is based on dynamic calculations using sensitivity analysis:

[0142]

[0143] in, This is the global gain scaling factor, which is set to [value] in this embodiment. ; The total number of channels for the actuator is 3 in this embodiment. To prevent the minimum value from being divided by zero, take This formula ensures that the correction amount for the PID parameters is concentrated on the actuators most effective in eliminating the current manifold deviation. For example, if the deviation vector mainly points to the temperature dimension, then the weight of the gas valve... It will increase significantly;

[0144] Based on the above dynamic weights, calculate the PID parameter correction values:

[0145]

[0146] in, The integral enhancement threshold is constrained to the convergence tolerance threshold of Example 4. of times, that is This ensures that strong integral action is triggered only when the deviation is significant; The preset reference gain adjustment amplitude is determined during the equipment commissioning phase by measuring the system's critical gain using the Ziegler-Nichols closed-loop oscillation method. After that, take Sure; The hyperbolic tangent function is used to... The increment is limited to the saturation range to prevent [the problem from worsening]. Excessive size can cause system oscillations; The Sigmoid activation function is used only in biases. Exceeding the threshold The integral time constant is significantly reduced. This involves enhancing the integral action to eliminate steady-state error; and preventing the integral time constant from increasing under strong deviation correction mode. Due to significant attenuation, that is This leads to the risk of integral saturation or divergence in the control system, so the system introduces safety clamping logic:

[0147]

[0148] in, As the preset minimum integration time constant, this embodiment takes... This ensures that the PID controller always operates within the stable domain; The sensitivity coefficient has the following values: and For example, when a high energy barrier needs to be quickly overcome, the system will temporarily increase the scaling factor. To improve the heating response speed;

[0149] The system executes the PLC writing and parameter execution steps; the terminal writes the corrected parameters into the register of the programmable logic controller via industrial Ethernet; the PLC adjusts the operating parameters accordingly, including the opening degree of the gas valve to control the fuel input, the frequency of the combustion fan to control the air-fuel ratio and the gas pressure in the kiln, and the speed of the trolley to control the residence time of the material in each temperature zone.

[0150] Example 7:

[0151] It also includes: remote monitoring terminals;

[0152] The remote monitoring terminal is configured to obtain the thermodynamic potential energy manifold data model and the coordinates of the current state particle from the optimization calculation server, and to visualize the motion trajectory of the current state particle on the thermodynamic potential energy manifold data model.

[0153] This embodiment describes the visualization function of the remote monitoring terminal;

[0154] The system performs data acquisition steps; the remote monitoring terminal maintains a long connection with the optimization computing server via the Websocket protocol, and subscribes in real time to the thermodynamic potential energy manifold data model as the background map, and the current state particle coordinates as the moving cursor;

[0155] The system performs trajectory visualization steps; the terminal draws a 3D surface map on the human-computer interaction interface; the surface represents potential energy terrain, and the color depth represents the energy level; a highlighted curve displays the optimal geodesic path; a dynamic light spot displays the current production status; the trajectory line left by the light spot shows the actual historical sintering process;

[0156] This embodiment provides an intuitive white-box monitoring method; in complex, lights-out factory management scenarios, operators no longer face monotonous temperature curves, but rather experience the monitoring process as if looking at a navigation map. Figure 1 The system monitors production; through three-dimensional visualization of potential energy terrain and trajectory, operators can intuitively see the direction and degree of deviation, greatly reducing cognitive load and improving human-machine collaboration efficiency.

[0157] Example 8:

[0158] The optimized computing server is deployed in the industrial edge computing gateway;

[0159] The industrial edge computing gateway interacts with production data acquisition terminals and equipment control terminals via the industrial bus protocol.

[0160] This embodiment limits the system's deployment architecture;

[0161] The system adopts an edge computing deployment strategy. Considering the requirements of low latency and high reliability in industrial sites, the optimized computing server is not deployed in a remote cloud, but is deployed as embedded software in the industrial edge computing gateway in the field. The gateway has a GPU acceleration unit to support real-time matrix operations of manifold construction and covariant derivatives.

[0162] The system executes industrial bus interaction steps; the industrial edge computing gateway connects to the lower-level hardware devices through industrial bus protocols such as ModbusTCP, PROFINET, or OPCUA; as the master station, it polls and reads data from the production data acquisition terminal and sends write commands to the equipment control terminal;

[0163] This embodiment adopts an edge computing architecture, ensuring that the closed-loop latency of the control loop is controlled at the millisecond level; in industrial sites with unstable network environments, localized deployment avoids the impact of network fluctuations on production safety; at the same time, this architecture guarantees the data sovereignty of core process data of the factory, such as raw material formulas and sintering curves, meeting the enterprise's confidentiality requirements for core technology assets.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A refractory material production optimization system based on big data analysis, characterized in that, include: Production data acquisition terminal, optimization computing server, and equipment control terminal; The production data acquisition terminal is configured to acquire raw material characteristic data and real-time sintering status data of refractory material production equipment, and send the raw material characteristic data and real-time sintering status data to the optimization calculation server. The optimization calculation server is configured to construct a thermodynamic potential energy manifold data model based on the raw material characteristic data, and to map the real-time sintering state data to the current state particle coordinates on the thermodynamic potential energy manifold data model; The optimization calculation server is also configured to calculate the optimal geodesic path based on the current state particle coordinates and the preset target energy state position, and generate adaptive control commands based on the vector deviation between the evolution trend of the current state particle coordinates and the optimal geodesic path. The equipment control terminal is configured to adjust the operating parameters of the refractory material production equipment based on the adaptive control command.

2. The refractory material production optimization system based on big data analysis according to claim 1, characterized in that, The optimized computing server is also configured to: Extract the energy characterization value of the acoustic emission signal from the real-time sintering state data; Obtain the preset fracture energy barrier reference threshold from the storage space; The energy characterization value is compared with the fracture energy barrier reference threshold; If the energy characterization value is greater than the fracture energy barrier reference threshold, an interruption protection control command is generated as the adaptive control command. If the energy characterization value is less than or equal to the fracture energy barrier reference threshold, a path tracking compensation command is generated as the adaptive control command.

3. The refractory material production optimization system based on big data analysis according to claim 1, characterized in that, The optimized computing server is also configured to: The raw material feature data is feature-encoded to generate a high-dimensional feature vector; Combining the preset phase diagram database, the high-dimensional feature vectors are used to generate a virtual phase transition potential energy data surface, which serves as the thermodynamic potential energy manifold data model. The geometric topology of the thermodynamic potential energy manifold data model characterizes the reaction energy barrier distribution corresponding to the raw material characteristic data.

4. The refractory material production optimization system based on big data analysis according to claim 1, characterized in that, The optimized computing server is also configured to: Calculate the actual evolution velocity vector of the current state particle coordinates; Calculate the tangent vector of the optimal geodesic path at the coordinates of the particle in the current state; Based on the actual evolution velocity vector and the tangent vector, calculate the magnitude of the covariant derivative; If the modulus of the covariant derivative is less than the preset convergence tolerance threshold, then the current operating parameters of the refractory material production equipment are maintained. If the magnitude of the covariant derivative is greater than or equal to the convergence tolerance threshold, then the compensation amount of the adaptive control command is determined based on the magnitude of the covariant derivative.

5. The refractory material production optimization system based on big data analysis according to claim 1, characterized in that, The production data acquisition terminal includes: a component analysis module and a multi-channel sensing module; The component analysis module is configured to collect the chemical composition and particle size distribution of the raw materials before they enter the kiln, and generate the raw material characteristic data. The multi-channel sensing module is configured to collect temperature timing data, atmosphere composition data, and acoustic emission waveform data during the firing process to generate the real-time sintering status data.

6. The refractory material production optimization system based on big data analysis according to claim 1, characterized in that, The device control terminal is also used for: The adaptive control command is parsed to obtain the PID parameter correction value; The PID parameter correction value is written into the programmable logic controller of the refractory material production equipment. The operating parameters include the gas valve opening degree, the combustion fan frequency, and the trolley running speed.

7. The refractory material production optimization system based on big data analysis according to any one of claims 1-6, characterized in that, It also includes: remote monitoring terminals; The remote monitoring terminal is configured to obtain the thermodynamic potential energy manifold data model and the coordinates of the current state particle from the optimization calculation server, and to visualize the motion trajectory of the current state particle on the thermodynamic potential energy manifold data model.

8. The refractory material production optimization system based on big data analysis according to any one of claims 1-6, characterized in that, The optimized computing server is deployed in an industrial edge computing gateway; The industrial edge computing gateway interacts with the production data acquisition terminal and the equipment control terminal via the industrial bus protocol.

Images