Magnetic performance in-situ monitoring and control method and system in permanent magnet sintering process

Abstract

The invention discloses a magnetic performance in-situ monitoring and control method and system in a permanent magnet sintering process, and belongs to the technical field of magnetic material manufacturing, and the method comprises the steps: obtaining a magnetic response signal of a permanent magnet blank in a sintering furnace, and generating in-situ monitoring data; extracting a dynamic trend parameter reflecting the change rate of the internal state of the permanent magnet blank; based on the in-situ monitoring data and the dynamic trend parameters, carrying out real-time calibration on the dynamic digital twin model simulating the dynamic relationship between the magnetic performance and the process parameters to obtain a calibrated model; rolling prediction operation is executed, and a magnetic performance evolution prediction track representing the magnetic performance change direction in the future time period is generated; according to an optimization function for optimizing future comprehensive magnetic performance and based on a magnetic performance evolution prediction track, solving and generating a comprehensive magnetic field regulation and control instruction; the instruction is sent to a vector magnetic field execution mechanism to drive the vector magnetic field execution mechanism to execute corresponding magnetic field intervention operation; the magnetic field environment in the sintering process can be optimized in real time, and the magnetic performance and consistency of a permanent magnet finished product are improved.

Description

Technical Field

[0001] This invention relates to the field of magnetic material manufacturing technology, and in particular to a method and system for in-situ monitoring and control of magnetic properties during the sintering process of permanent magnets. Background Technology

[0002] High-performance rare-earth permanent magnets, such as neodymium iron boron (NdFeB) and samarium cobalt (SMC) magnets, are indispensable key functional materials in modern industry, widely used in new energy vehicles, wind power generation, and consumer electronics. Sintering is one of the core processes in manufacturing these permanent magnets. It densifies the pressed magnetic powder blank under high temperature and a specific atmosphere, forming the final magnetic microstructure. The temperature, atmosphere, and magnetic field environment during sintering directly determine the key magnetic properties of the permanent magnet, such as coercivity, remanence, and energy product.

[0003] In existing permanent magnet sintering processes, quality control is typically achieved through offline testing. This involves cooling the green body to room temperature after the entire sintering process is complete, then taking a sample and measuring its magnetic properties using magnetic performance testing equipment. For controlling the sintering process, the industry generally employs a fixed, programmed control scheme. This involves pre-setting the sintering furnace's heating curve, holding time, and fixed orientation magnetic field parameters based on historical experience. These parameters are executed according to the predetermined program throughout the sintering process and cannot be adjusted based on actual changes in the internal state of the green body.

[0004] The aforementioned existing technical solutions have significant drawbacks. First, offline testing is a post-event inspection, unable to monitor the evolution of the green body's magnetic properties in real time during sintering. Once performance defects are detected, the entire batch of products may already be finalized and irreversible, leading to high scrap rates and production costs. Second, the open-loop control method relying on fixed programs cannot cope with the impact of uncertainties such as batch differences in raw materials and fluctuations in equipment status on the sintering process. This results in poor performance consistency between different batches, and even between different locations within the same batch, with significant fluctuations in magnetic property indicators. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method and system for in-situ monitoring and control of magnetic properties during the sintering process of permanent magnets. By employing a technique of constructing a dynamic digital twin for forward prediction and closed-loop feedback control, the magnetic field environment during the sintering process can be optimized in real time, thereby improving the magnetic properties and consistency of the finished permanent magnet.

[0006] The above objectives can be achieved through the following approach: A method for in-situ monitoring and control of magnetic properties during the sintering process of permanent magnets, the method comprising: Acquire the magnetic response signal of the permanent magnet billet inside the sintering furnace to generate in-situ monitoring data; Dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank are extracted from in-situ monitoring data. Based on in-situ monitoring data and dynamic trend parameters, the dynamic digital twin model simulating the dynamic relationship between magnetic properties and process parameters is calibrated in real time to obtain the calibrated model. By performing rolling prediction calculations using a calibrated model, a magnetic performance evolution prediction trajectory representing the direction of magnetic performance changes in the future is generated; based on the optimization function for optimizing the future comprehensive magnetic performance and the magnetic performance evolution prediction trajectory, a comprehensive magnetic field control command is solved and generated. The integrated magnetic field control command is sent to the vector magnetic field actuator to drive it to perform the corresponding magnetic field intervention operation.

[0007] Optionally, acquiring the magnetic response signal of the permanent magnet blank in the sintering furnace and generating in-situ monitoring data includes: applying a short-time pulsed magnetic field sequence containing at least two sub-pulses with different magnetic field characteristics to the permanent magnet blank; acquiring the complete magnetic response waveform generated by the permanent magnet blank in response to the short-time pulsed magnetic field sequence in real time; performing integration and conversion processing on the complete magnetic response waveform to parse out the in-situ monitoring data containing instantaneous magnetic performance parameters.

[0008] Optionally, the step of extracting dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank from the in-situ monitoring data includes: calculating the rate of change of magnetization intensity within a specific time window from the complete magnetic response waveform to obtain a first dynamic trend parameter; quantifying the shape difference of the complete magnetic response waveform in a specific interval between the current measurement cycle and the previous measurement cycle to obtain a second dynamic trend parameter; and combining the first dynamic trend parameter and the second dynamic trend parameter to form the dynamic trend parameter.

[0009] Optionally, the real-time calibration of the dynamic digital twin model based on in-situ monitoring data and dynamic trend parameters to obtain a calibrated model includes: inputting the instantaneous magnetic property parameters from the in-situ monitoring data into the dynamic digital twin model to calculate the predicted value output by the model; applying constraints to the evolution gradient of the internal state variables of the dynamic digital twin model through the dynamic trend parameters; and adjusting the internal variable parameters in the dynamic digital twin model in reverse using an optimization algorithm that minimizes the difference between the predicted value and the instantaneous magnetic property parameters to complete the real-time calibration and obtain the calibrated model.

[0010] Optionally, the step of generating a magnetic property evolution prediction trajectory by performing rolling prediction calculations through a calibrated model includes: obtaining a process parameter plan that defines the planned path of process parameters in the future time period and inputting it into the calibrated model; performing forward simulation based on the calibrated model and the process parameter plan to obtain predicted values ​​of magnetic property parameters at multiple consecutive time points in the future; and connecting the predicted values ​​of magnetic property parameters at multiple consecutive time points in the future in chronological order to construct the magnetic property evolution prediction trajectory.

[0011] Optionally, the step of optimizing the future comprehensive magnetic properties based on the optimization function and solving and generating comprehensive magnetic field control instructions based on the predicted trajectory of magnetic property evolution includes: generating multiple different candidate magnetic field control schemes and inputting each scheme into the calibrated model to obtain multiple candidate predicted trajectories; evaluating the multiple candidate predicted trajectories using the optimization function and selecting the target scheme with the optimal future comprehensive magnetic properties; and parsing the target scheme into specific instructions containing the time-varying program of magnetic field strength, direction, and waveform, forming the comprehensive magnetic field control instructions.

[0012] Optionally, the method further includes an online learning step: after performing the magnetic field intervention operation, new in-situ monitoring data is acquired in the next measurement cycle; the new in-situ monitoring data is compared with the expected data of the magnetic property evolution prediction trajectory at the corresponding time to generate model error data; the model error data is used to fine-tune the learnable parameters of the dynamic digital twin model to continuously improve the model prediction accuracy.

[0013] Optionally, the optimization function is configured to take maximizing the predicted maximum magnetic energy product at the end of the prediction time domain as the main objective, while simultaneously using the stability of the changes in coercivity and remanence in the prediction time domain as constraints for collaborative optimization.

[0014] Optionally, the vector magnetic field actuator is a three-axis magnetic field generator, and the integrated magnetic field control command drives the three-axis magnetic field generator to generate an intervention magnetic field whose intensity and direction can be independently and programmatically controlled in three-dimensional space.

[0015] Based on the same inventive concept, the present invention also provides an in-situ monitoring and predictive control system for the magnetic properties of permanent magnets during the sintering process to implement the above-mentioned method, the system comprising: The in-situ monitoring module is configured to acquire in-situ monitoring data of the permanent magnet billet in the sintering furnace and extract dynamic trend parameters from it. The digital twin and prediction module has a built-in dynamic digital twin model, configured to perform model calibration and rolling prediction based on the in-situ monitoring data and the dynamic trend parameters, and output the predicted trajectory of magnetic property evolution. The intelligent decision-making module is configured to solve the predicted trajectory and optimization function based on the magnetic property evolution to generate comprehensive magnetic field control commands; The vector magnetic field execution module is configured to receive and execute the comprehensive magnetic field control command; wherein the in-situ monitoring module, digital twin and prediction module, intelligent decision-making module and vector magnetic field execution module are connected in sequence to form a closed-loop control loop.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention achieves predictive control over the evolution of magnetic properties during sintering by acquiring the dynamic response of the permanent magnet blank in real time within the sintering furnace and utilizing digital twin technology for forward-looking prediction and optimization decisions. It enables optimal control measures to be taken before irreversible degradation of magnetic properties occurs, thereby improving the consistency and stability of the final product's magnetic properties and narrowing the fluctuation range of core performance indicators within a batch.

[0017] 2. The decision-making mechanism based on multi-objective collaborative optimization adopted in this invention can automatically find and execute the process path that maximizes the final comprehensive magnetic performance index, rather than focusing solely on improving a single parameter. This ensures that while improving coercivity or remanence, other key performance characteristics are not sacrificed, thereby effectively improving the overall performance level of the product and the first-pass yield, and reducing rework and material loss caused by substandard performance.

[0018] 3. This invention constructs an adaptive closed-loop system with online learning and self-optimization capabilities. The system can automatically adjust its internal model by continuously comparing the differences between predictions and reality to adapt to drifts in equipment status or subtle changes in raw material batches. This reduces reliance on manual experience and fixed process parameters, achieving a high degree of automation and intelligence in the sintering process, and providing a systematic solution for the efficient and high-quality manufacturing of high-performance permanent magnets. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of in-situ monitoring signal acquisition according to the present invention; Figure 3 This is a schematic diagram illustrating the predictive control and optimization decision-making of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 As shown, one embodiment of the present invention proposes a method for in-situ monitoring and control of magnetic properties during the sintering process of permanent magnets, the method comprising: Acquire the magnetic response signal of the permanent magnet billet inside the sintering furnace to generate in-situ monitoring data; Dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank are extracted from in-situ monitoring data. Based on in-situ monitoring data and dynamic trend parameters, the dynamic digital twin model simulating the dynamic relationship between magnetic properties and process parameters is calibrated in real time to obtain the calibrated model. By performing rolling prediction calculations using a calibrated model, a magnetic performance evolution prediction trajectory representing the direction of magnetic performance changes in the future is generated; based on the optimization function for optimizing the future comprehensive magnetic performance and the magnetic performance evolution prediction trajectory, a comprehensive magnetic field control command is solved and generated. The integrated magnetic field control command is sent to the vector magnetic field actuator to drive it to perform the corresponding magnetic field intervention operation.

[0023] Specifically, through specific in-situ detection methods, not only is the current magnetic performance state of the permanent magnet blank obtained during the high-temperature sintering process, but more importantly, dynamic trend information reflecting the evolution rate of its internal microstructure is extracted. This real-time data, containing both static and dynamic information, is injected into a dynamic digital twin model that runs synchronously with the physical sintering process. Real-time calibration ensures that the virtual model and the physical entity are perfectly aligned in terms of their current state and evolution trend. The dynamic digital twin model can be described using a simplified state-space framework. The state variables of the model are defined as follows: ,in Characterizes the effective field strength associated with the magnetocrystalline anisotropic field (strongly correlated with coercivity). Characterizes the degree of directional alignment of magnetic domains (related to remanence). Its state evolution is governed by the following set of ordinary differential equations: , ,in, and It is a nonlinear function that describes temperature based on physical mechanisms (such as Arrhenius-type thermal activation equations) and / or trained using historical data. and external magnetic field The driving effect on state variables. and These are coupling coefficients, used to couple measured dynamic trend parameters. , Introducing the state equation as an additional driving force enables dynamic calibration. The model output (predicted coercivity) and remanence (Through static mapping functions) , get, and Similarly, this can be determined through data fitting. The specific functional form and parameters in the above equations are the "internal variable parameter P" and the "learnable parameter," which are optimized through calibration and online learning steps. Using this calibrated, highly realistic virtual model, the possible evolution trajectory of magnetic properties under a preset process path can be quickly "predicted" over a future period. Based on this predicted trajectory, the multi-objective optimization decision module can proactively select the magnetic field control strategy that optimizes the overall magnetic properties of the final product and ensures the most stable process from various virtual intervention schemes. This strategy is then translated into precise control commands, which are physically implemented by the vector magnetic field actuator within the sintering furnace. This process is repeated cyclically, achieving a fundamental shift from passive, hysteresis-feedback control to proactive, predictive feedforward control.

[0024] This approach fundamentally overcomes the technical bottlenecks of traditional sintering processes, which rely on experience and are uncontrollable. By enabling real-time prediction and proactive intervention of magnetic properties during sintering, optimal corrective measures can be taken before irreversible degradation occurs, thereby improving the consistency and stability of the final product's magnetic properties and narrowing the fluctuation range of core performance indicators within the same batch. Through multi-objective collaborative optimization, this method can automatically find and execute the process path that maximizes the final comprehensive magnetic performance indicators (such as magnetic energy product), thereby improving the overall performance level and yield of the product and reducing rework and material waste caused by substandard performance. Furthermore, the self-learning, adaptive closed-loop system constructed by this method reduces reliance on human experience, achieving a high degree of automation and intelligence in the sintering process, and providing a new systematic solution for the high-quality and high-efficiency manufacturing of high-performance permanent magnets.

[0025] Optionally, the magnetic response signal of the permanent magnet blank in the sintering furnace is acquired to generate in-situ monitoring data, including: applying a short-time pulsed magnetic field sequence containing at least two sub-pulses with different magnetic field characteristics to the permanent magnet blank; acquiring the complete magnetic response waveform of the permanent magnet blank to the short-time pulsed magnetic field sequence in real time; performing integration and conversion processing on the complete magnetic response waveform to analyze the in-situ monitoring data containing instantaneous magnetic performance parameters.

[0026] Specifically, by applying a composite pulsed magnetic field containing rich dynamic information, the nonlinear magnetic response of the permanent magnet blank under high-temperature sintering is excited and captured. This method differs from the limitation of a single pulse obtaining only a static magnetic slice; it creates the necessary conditions for extracting dynamic trend parameters through the design of a pulse sequence. The system first applies a preset short-duration pulsed magnetic field sequence to the permanent magnet blank located in the high-temperature zone of the sintering furnace. This sequence is not a simple single pulse, but a designed combination of waveforms, typically including a high-intensity main pulse with an amplitude of 1.5 to 2 Tesla to drive the blank to near-saturation magnetization, followed by a probe sub-pulse with a lower intensity or opposite direction, spaced 5 to 20 milliseconds apart. The total duration of the entire sequence is strictly controlled within 50 milliseconds to ensure negligible impact on the thermal balance of the sintering process.

[0027] Reference Figure 2 As shown, this process captures with high fidelity the extremely weak and rapidly changing magnetic flux variations induced by the excitation sequence and converts them into a raw induced signal suitable for digital processing. To this end, an induction coil positioned adjacent to the permanent magnet blank senses the change in total magnetic flux, thereby generating an induced electromotive force (EMF). A high-speed data acquisition system connected to this induction coil performs analog-to-digital conversion on the EMF signal at a sampling rate of no less than five million times per second, generating a time-resolved digital voltage sequence, which is the raw induced signal. Simultaneously, to eliminate interference from the excitation magnetic field itself, a compensation coil is also included in the system. Its acquired signal is used to cancel out the background signal directly generated by the external excitation field from the main induction coil signal, ensuring that the final raw induced signal only reflects the magnetization response of the permanent magnet blank itself.

[0028] Finally, the raw induced signal at the physical level is interpreted into instantaneous magnetic performance parameters with clear physical meaning through mathematical operations, forming structured in-situ monitoring data. This step first processes the compensated raw induced signal, i.e., the voltage signal. The signal is then integrated to obtain a signal proportional to the change in magnetic flux of the billet. The calculation formula is as follows: , In this formula, This represents the magnetic flux that changes over time. This is a compensated real-time voltage signal. This is a system constant determined through prior calibration, and its value is related to the number of turns and the effective cross-sectional area of ​​the induction coil. Subsequently, using the known geometric dimensions of the blank, the magnetic flux signal is converted into the average magnetization of the material. The calculation formula is as follows: , In this formula, The instantaneous average magnetization of the material. It is a geometric factor associated with the effective magnetization volume of the billet. This is achieved through calculation... Excitation magnetic field recorded synchronously By combining these elements, the system can construct a dynamic hysteresis loop segment within the excitation cycle. Finally, key feature points are extracted from this loop segment, such as the magnetic field strength value when the magnetization is zero, which is the instantaneous coercivity, along with other indicators characterizing magnetic properties. Together, these constitute a structured dataset containing the instantaneous magnetic property parameters at the current moment. This is the in-situ monitoring data, which serves as the input for the next step of model calibration.

[0029] And system constants The physical meaning of is the magnetic flux to voltage conversion coefficient of the induction coil, and its value is related to the effective number of turns of the induction coil. and effective cross-sectional area It is inversely proportional, as determined through offline calibration experiments. The calibration method involves placing a known magnetic moment at the position of the billet inside the sintering furnace. Using standard samples, apply the same pulsed magnetic field sequence and measure the induced voltage. And accumulate points. From the formula The calculation shows that, among which Permeability of free space. Geometric factor. The physical meaning of is the average magnetization intensity corresponding to a unit magnetic flux, and its value is related to the effective magnetized volume of the billet. Inversely proportional, the calculation formula is: ,in It is calculated based on the actual geometric dimensions of the blank.

[0030] Optionally, dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank can be extracted from the in-situ monitoring data, including: calculating the rate of change of magnetization intensity within a specific time window from the complete magnetic response waveform to obtain a first dynamic trend parameter; quantifying the shape difference of the complete magnetic response waveform in a specific interval between the current measurement cycle and the previous measurement cycle to obtain a second dynamic trend parameter; and combining the first dynamic trend parameter and the second dynamic trend parameter to form a dynamic trend parameter.

[0031] Specifically, by extracting deep information from dynamic responses that can predict future performance evolution trends, key inputs are provided for predictive control. Quantifying the rate of change in magnetization of a permanent magnet preform under pulse excitation directly reflects the ease of magnetic domain flipping within the material, serving as a sensitive probe of its microstructural state. This is achieved by starting with the magnetic response waveform from in-situ monitoring data, specifically the processed average magnetization signal. The system will lock. The time window in the curve where the change is most dramatic is typically the response interval at the rising or falling edge of the excitation pulse. Within this time window, the first dynamic trend parameter is obtained by calculating the average rate of change of magnetization. The formula is shown below: , In this formula, This is the first dynamic trend parameter, representing the average rate of change of magnetization. This is the time window The total change in internal magnetization. For example, under the action of a pulse excitation rising edge, It represents the difference between the initial magnetization and the peak magnetization. This is the time required to achieve the change. A higher value indicates a faster magnetization response and easier domain flipping in the material, which may indicate grain refinement or improved orientation. Conversely, a lower value may indicate the presence of factors hindering magnetization. This parameter directly reflects the dynamic characteristics of the material's state.

[0032] Next, by capturing and quantifying subtle changes in the local shape of the dynamic hysteresis loop, the uniformity of the distribution of coercive mechanisms within the material can be reflected. This is achieved by focusing on the dynamic hysteresis loop constructed from in-situ monitoring data, particularly the region near the inflection point of the second quadrant (demagnetization curve) within the loop. This is because the shape of this region is closely related to the material's coercive formation mechanism (such as pinning or nucleation). The system will then use the current measurement cycle (the...) The curve segment obtained within this region during the period (the first period) is compared with the previous measurement period (the second period). The corresponding curve segments of the two curves (within a given period) are compared. The second dynamic trend parameter is obtained by calculating the degree of difference between the two curves. The formula is shown below: , In this formula, This is the second dynamic trend parameter, which represents the degree of difference in the local shape of the demagnetization curve within two consecutive periods. and These represent the demagnetization curve segments within a specific magnetic field strength range for the current and previous cycles, respectively. (Function) It could be an area measurement algorithm, such as calculating the area of ​​the region enclosed by two curves; or it could be the distance between shape descriptors, such as by calculating the curvature function of two curves. The norm difference is obtained. Function A uses the absolute value of the area enclosed by the two curves as a quantitative indicator of shape difference. The specific calculation steps are as follows: within the same magnetic field strength range... The inner (usually selected is the interval near the coercivity point of the demagnetization curve, such as...) ), for the current periodic curve Compared with the previous cycle curve Integrating the difference in magnetization, i.e. Where M_n(H) and curves and upper magnetic field strength The corresponding magnetization value, if A smaller value indicates a stable sintering process and a gradual evolution of the magnetic structure; if... A sudden change or sustained increase may indicate unexpected microstructural changes, such as abnormal grain growth or the precipitation of new phases, which could affect the uniformity of coercivity. These two dynamic trend parameters... and This will serve as the core input for the next step of real-time calibration of the dynamic digital twin model.

[0033] Optionally, based on in-situ monitoring data and dynamic trend parameters, the dynamic digital twin model is calibrated in real time to obtain a calibrated model. This includes: inputting the instantaneous magnetic property parameters from the in-situ monitoring data into the dynamic digital twin model and calculating the predicted values ​​output by the model; applying constraints to the evolution gradient of the internal state variables of the dynamic digital twin model through the dynamic trend parameters; and adjusting the internal variable parameters in the dynamic digital twin model in reverse through an optimization algorithm that minimizes the difference between the predicted values ​​and the instantaneous magnetic property parameters to complete the real-time calibration and obtain the calibrated model.

[0034] Specifically, by injecting real-time measurement data from the physical world into the virtual model, the model is forced to synchronize with reality, ensuring that subsequent model-based predictions reflect the true state of the physical entity. Using currently measured magnetic property parameters, the dynamic digital twin model is instantly "anchored," ensuring its output is perfectly aligned with the current state of the physical prototype. Instantaneous magnetic property parameters from in-situ monitoring data, such as instantaneous coercivity, are also used. The instantaneous coercivity is extracted by inputting it into the dynamic digital twin model. The specific method is as follows: find the magnetization intensity in the dynamic hysteresis loop segment. From the moment of crossing zero The magnetic field strength value at that point is accurately calculated using linear interpolation. ,this That is, the instantaneous coercivity during this measurement period. The linear interpolation formula is: , in and Let M(t) be the two nearest sampling time points on either side of zero, and satisfy the following conditions: , This model is a mathematical model constructed based on a hybrid approach of physical mechanisms and data-driven methods, capable of operating according to a set of internal variable parameters. (e.g., simulated average grain size, orientation distribution function, pinning center density, etc.) and external process inputs (e.g., temperature, magnetic field) are used to calculate the corresponding magnetic properties. At this point, the system runs the model to obtain the results based on the current internal parameters. The first predicted value output by the model below .

[0035] Next, dynamic trend parameters are used to constrain the evolution direction of the model's internal states, preventing the model from getting stuck in non-physical parameter combinations during calibration. Specifically, the first dynamic trend parameter extracted in the previous step... Second dynamic trend parameter The model is input along with the internal variable parameters P. The model contains differential equations describing how these internal variable parameters P evolve over time. The dynamic trend parameters D1 and D2 are used as boundary conditions or constraints for these differential equations, thereby ensuring that the adjustment of the model's internal state not only matches the current static value, but also that its rate of change and direction are consistent with physical reality.

[0036] By optimizing the algorithm, the model's internal variable parameters are adjusted in reverse to minimize the difference between the model output and the measured data, thus achieving real-time calibration. The system will launch an optimization solver whose goal is to minimize a preset cost function. The formula is shown below: , In this formula, Let be the cost function that needs to be minimized. It is the first predicted value currently output by the model. These are the measured instantaneous magnetic properties parameters. This represents the rate of change of the variable parameters within the model. It is a function used to quantify the rate of change of the model's internal parameters. Compared with measured dynamic trend parameters and The degree of conformity between them, the smaller the value, the higher the degree of conformity, the function A specific implementation is used to constrain the evolution trend of the model's internal state variables to match the measured dynamic trend. It is defined as follows: , in, and It is the rate of change of two internal state parameters selected in the dynamic digital twin model that are most relevant to the magnetization response rate and microstructure uniformity, respectively (e.g., the inverse rate of change of the simulated average grain size and the rate of change of the variance of the simulated orientation distribution). and To incorporate dynamic trend parameters , The physical dimensions are converted into conversion coefficients that match the rate of change of internal parameters, which can be determined through model identification. and These are weighting coefficients used to balance the influence of the two trend constraint terms. and These are weighting coefficients used to balance the proportions of static error and dynamic trend error in the total cost; their values ​​are typically determined through offline experiments or historical data analysis. The optimization solver iteratively modifies its internal variable parameters. For example, algorithms such as gradient descent or particle swarm optimization can be used to find the cost function that minimizes the cost function. The parameter set that reaches the minimum value .when The optimization process stops when the parameter set is less than a preset convergence threshold or when the maximum number of iterations is reached. These are the parameters of the calibrated model, and the resulting model is the calibrated model. This calibrated model will be immediately used for the next rolling prediction calculation.

[0037] Optionally, a rolling prediction operation is performed using the calibrated model to generate a magnetic property evolution prediction trajectory, including: obtaining a process parameter plan that defines the planned path of process parameters in the future time period and inputting it into the calibrated model; performing forward simulation based on the calibrated model and the process parameter plan to obtain the predicted values ​​of magnetic property parameters at multiple consecutive time points in the future; and connecting the predicted values ​​of magnetic property parameters at multiple consecutive time points in the future in chronological order to construct a magnetic property evolution prediction trajectory.

[0038] Specifically, a calibrated digital twin model is used for "virtual sintering" to predict the future performance trend of the permanent magnet blank under a given process route. External driving conditions for a future period are provided to the calibrated model as input for its forward simulation. The system first obtains the process parameter plan for the current sintering process. The process parameter plan is a structured time-series data file or database query result, which defines the process parameters from the current moment... Starting with the future prediction time domain Internally, it contains the planned values ​​of all programmable external variables in the sintering process. It includes at least two core data columns: a timestamp sequence. and the process parameter vector corresponding to each timestamp. ,in For furnace temperature setpoint, , , The intensity components of the triaxial magnetic field are preset. The plan is generated by the host computer of the sintering process based on the preset process formula.

[0039] The calibrated model is then driven to simulate and calculate the changes in the magnetic properties of the permanent magnet blank in the future prediction time domain, following a time-stepping approach. The system will then use the previous time step... The system state (i.e., the calibrated model internal parameters) (This is used as the initial condition for simulation, and the process parameters are pre-planned.) and This serves as an external input. Subsequently, the sintering process is simulated by solving a system of ordinary differential equations within the model. The formulas are shown below: , In this formula, yes The vector of variable parameters inside the model at time step. It is the core function of the calibrated dynamic digital twin model, which describes the internal parameters. The rate of change over time is an external process parameter. and and the current internal status The function. It is a sufficiently small simulation time step, typically between 0.1 and 1 second. The system will start from... Initially, the formula is executed iteratively, calculating step by step from... arrive Intrinsic parameters at all discrete time points between .

[0040] The final step is to convert the discrete internal parameter sequence obtained from the simulation into a continuous and intuitive predicted trajectory of magnetic property evolution. At each simulation time point... The system will utilize the internal parameters at that time. By mapping the static output of the model, the predicted values ​​of magnetic properties parameters at that moment are calculated, such as the predicted coercivity. and remanence The formula is shown below. , , In this formula, and These are the static output functions of the model, which map the internal parameter space to the observable magnetic property parameter space. They represent the predicted magnetic property parameters (e.g., ...) at all discrete time points within the future prediction time domain T. , Connecting these points chronologically creates a multi-dimensional magnetic performance prediction trajectory. This trajectory visually demonstrates how key properties of the permanent magnet blank, such as coercivity and remanence, will evolve over a future period if sintering continues according to the current plan. This prediction trajectory, containing information on future trends, serves as a direct basis for forward-looking optimization decisions.

[0041] Optionally, based on the optimization function for optimizing future comprehensive magnetic properties and the predicted trajectory of magnetic property evolution, a comprehensive magnetic field control command is solved and generated, including: generating multiple different candidate magnetic field control schemes and inputting each scheme into a calibrated model to obtain multiple corresponding candidate predicted trajectories; evaluating the multiple candidate predicted trajectories using the optimization function and selecting the target scheme with the optimal future comprehensive magnetic property index; and parsing the target scheme into specific instructions containing the time-varying program of magnetic field strength, direction, and waveform, forming a comprehensive magnetic field control command.

[0042] Specifically, the trajectory predicted by the evolution of magnetic properties serves as a "baseline" trajectory, representing the expected outcome without additional intervention. The system generates a series of different candidate magnetic field manipulation schemes. Each scheme is a function of time. Defines the future prediction time domain Inside (from) arrive The system determines how the magnetic field strength, direction, and waveform should change. These candidate schemes can be generated using parameterized templates, such as defining an oscillating field with parameters including amplitude, frequency, and phase superimposed on a reference magnetic field. The system will utilize a calibrated dynamic digital twin model to analyze each candidate scheme. As input, a forward simulation is performed again to obtain the new predicted trajectory of magnetic property evolution under this scheme.

[0043] Then, based on a comprehensive evaluation criterion, the optimal solution is selected from numerous candidate solutions. This evaluation criterion is the pre-defined multi-objective optimization function. The function is designed to achieve optimal global performance, not just improve a single metric. The formula is shown below: , In this formula, It is the value of the multi-objective optimization function to be maximized. At the end of the prediction time domain The maximum magnetic energy product predicted at any given time. and These are the predicted coercivity and remanent trajectory under the candidate schemes, respectively. and It is a preset target value or target trajectory. These are weighting coefficients used to adjust the importance of the final magnetic energy product, process coercivity stability, and process remanence stability in the overall evaluation. These coefficients are set according to the specific permanent magnet grade and performance requirements. The system will calculate the corresponding weighting coefficients for each candidate scheme. value, The candidate solution with the highest value is identified as the target solution under the current circumstances.

[0044] This transforms the abstract target scheme into specific, comprehensive magnetic field control commands that can be precisely executed by the hardware. The selected target scheme itself is a mathematical function describing the change of the magnetic field vector over time. The system needs to discretize it, generating a series of timestamp-magnetic field vector pairs. For example, the prediction time domain T is divided into... Each control step consists of small steps, each lasting approximately 1 to 3 seconds. For each step, the system calculates the average magnetic field strength, spatial direction, and specific waveform parameters (such as DC current and pulse frequency) to be applied within that time period. These parameters are combined to form a series of time-sequential low-level control commands, namely, the integrated magnetic field control instructions. These instructions are then sent to the vector magnetic field actuator of the sintering furnace, guiding it to accurately reproduce the planned magnetic field change process within the subsequent predicted time domain T.

[0045] Optionally, the method also includes an online learning step: after performing the magnetic field intervention operation, new in-situ monitoring data is acquired in the next measurement cycle; the new in-situ monitoring data is compared with the expected data of the magnetic property evolution prediction trajectory at the corresponding time to generate model error data; the learnable parameters of the dynamic digital twin model are fine-tuned using the model error data to continuously improve the model prediction accuracy.

[0046] Specifically, by using real-world results to test and refine the predictive capabilities of the virtual model, the entire control system gains the ability to self-evolve online and continuously improve its accuracy. This is achieved by acquiring real feedback data from the system after predictive intervention. Within a control cycle (e.g., arrive At the end of the measurement cycle, the system will proceed to the next measurement cycle (i.e. At any given moment, the in-situ monitoring module is immediately activated to perform a new magnetic performance measurement, obtaining the true new in-situ monitoring data of the permanent magnet blank at that moment. This data contains the actual response of the physical world to the comprehensive magnetic field control commands applied in the previous cycle, and is the sole factual basis for the model's self-learning.

[0047] The system utilizes the discrepancy between predictions from a quantified virtual model and physical reality. The trajectory is calculated based on the evolution of magnetic properties. The expected data at each time point is compared with the new in-situ monitoring data actually measured in the previous step. This comparison process generates a set of model error data. The formula is shown below: , , In this formula, and These are the two components of the model error data, representing the prediction errors of coercivity and remanence, respectively. and These are the actual magnetic properties extracted from the new in-situ monitoring data. and The trajectory predicted by the evolution of magnetic properties at the beginning of the previous control cycle is given in... The expected data for each moment. This set of model error data accurately reflects the performance of the dynamic digital twin model in this prediction.

[0048] The dynamic digital twin model is fine-tuned using model error data to improve its accuracy in future predictions. This process is an online learning process. The system uses the model error data... and As input, a subset of learnable parameters in the dynamic digital twin model are adjusted using an update algorithm. These parameters may include empirical coefficients in the model's internal physical equations or network weights in the data-driven components. The formula is shown below. , In this formula, These are the updated learnable parameters of the model. These are the parameters before the update. It is a learning rate that controls the step size of each adjustment. Its value is usually within a small range, such as 0.001 to 0.01, to ensure the stability of the learning process. It is the gradient of the error cost function with respect to the learnable parameters, indicating the direction of parameter adjustment. Error cost function Typically, this is the square of the model error data norm. By repeatedly performing this learning and optimization step, the dynamic digital twin model can automatically adapt to slow changes caused by factors such as equipment aging and batch differences in materials, continuously maintaining its high predictive accuracy. This allows the performance of the entire control system to be continuously optimized as operating time increases.

[0049] Optionally, the optimization function is configured to take maximizing the predicted maximum magnetic energy product at the end of the prediction time domain as the main objective, while simultaneously using the stability of the changes in coercivity and remanence in the prediction time domain as constraints for joint optimization.

[0050] Specifically, a complex optimization problem involving trade-offs among multiple performance indicators is transformed into a single-objective mathematical problem solvable by a computer. A quantitative scoring system is established to evaluate the overall merits of any candidate magnetic field control scheme. This system not only rewards schemes that improve the performance of the final product but also penalizes schemes that lead to instability in the sintering process. The pre-defined multi-objective optimization function consists of a main objective term and two penalty terms. The main objective term directly corresponds to the core performance indicator of the permanent magnet, namely the maximum magnetic energy product. The system extracts the predicted trajectory of magnetic property evolution at the end of the prediction time domain. Predictive coercivity And predicting remanence Based on this, the predicted value of the maximum magnetic energy product is estimated. The formula is as follows: , In this formula, It is the main target item, and its value represents the final performance level that the control scheme can achieve. It is the predicted maximum magnetic energy product, which is usually obtained by an approximation formula. and It can be calculated or obtained by searching a pre-established materials database.

[0051] The fluctuations in coercivity and remanence, i.e., the stability of these changes, are then quantified throughout the entire prediction time domain. An ideal sintering process requires not only a good final result but also a smooth and controllable process. To this end, the system calculates the predicted coercivity trajectory. and residual magnetic trajectory The degree of deviation from its own mean or a preset target trajectory. The cumulative amount of this deviation is the penalty term, as shown in the formula below: , In this formula, It is the total penalty item. and These are time-varying predicted values ​​obtained from the predicted trajectory of magnetic property evolution. and They are in the entire prediction time domain The average value within the range. Integration operations from... arrive The cumulative fluctuation energy of the entire process was calculated. and These are preset weighting coefficients, whose physical meaning lies in adjusting the system's sensitivity to coercivity fluctuations and remanence fluctuations. For example, for high-temperature grades that are more sensitive to coercivity... The values ​​will be relatively large. These coefficients also include a dimension conversion function to ensure that the penalty term and the main objective term can be mathematically calculated meaningfully. (Weight coefficients) and The setting involves considering the fluctuation amplitude of coercivity and remanence (dimension 1). and ) Dimensionless, and with the main objective term (dimensions are) Normalization is performed before dimensional operations. Specifically, .in and These are the typical expected values ​​of coercivity and remanence for that grade of permanent magnet (e.g., N52 grade). ), To predict the length of the time domain, and It is a dimensionless penalty intensity coefficient, typically ranging from 0.01 to 0.1, used to adjust the system's tolerance to process fluctuations. Its specific value can be determined through offline process simulation optimization or by correlation analysis of performance fluctuations and the final magnetic energy product in historical production data.

[0052] Finally, the main objective term and the penalty term are combined to form the final, single-objective function to be optimized. The goal of the optimization decision module is to find a candidate magnetic field control scheme that maximizes the value of this function. The formula is shown below: , In this formula, This is the final comprehensive evaluation value. By subtracting the penalty term from the main objective term, the formula clearly reflects the optimization objective: to maximize the final magnetic energy level while suppressing performance fluctuations during the process. Through iteration, the optimization algorithm will naturally favor those methods that achieve both high performance and low efficiency. And can make The smallest possible solution was adopted. Co-optimization of final performance and process stability was achieved at the virtual simulation level.

[0053] Optionally, the vector magnetic field actuator is a three-axis magnetic field generator. The integrated magnetic field control command drives the three-axis magnetic field generator to generate an intervention magnetic field whose intensity and direction can be independently and programmatically controlled in three-dimensional space.

[0054] Specifically, the ability to rapidly, precisely, and multidimensionally control magnetic field vectors provides the physical basis for executing complex predictive intervention commands. This is achieved by constructing a physical device capable of generating magnetic fields of arbitrary direction and intensity in three-dimensional space. The triaxial magnetic field generator is a special electromagnetic system integrated within the sintering furnace. It consists of three sets of mutually orthogonal Helmholtz coil pairs, or a hybrid structure of permanent magnets and electromagnetic coils. Taking the Helmholtz coil pairs as an example, the three sets of coils are respectively along the spatial rectangular coordinate system... The axial arrangement allows each coil to independently generate a uniform magnetic field component along its respective axis. By controlling the magnitude and direction of the current flowing through these three sets of coils, a magnetic field vector of arbitrary direction and intensity can be synthesized in the central region of the coil pair, i.e., at the location of the permanent magnet blank. The formula is shown below. , In this formula, It is the final synthesized magnetic field vector. It represents its components on the three coordinate axes. It is the unit vector of the coordinate axis. The magnitude of each component is proportional to the current of the corresponding coil group, that is... ,in Let be the coil constant. The current is used. The magnetic field strength of this device is typically adjustable between 0 and 1.5 Tesla, with an adjustment accuracy of less than 0.01 Tesla, and a spatial direction pointing accuracy better than ±2°.

[0055] The digitized integrated magnetic field control command output from the upper-level decision-making module is converted into an analog electrical signal driving the coil. When the vector magnetic field actuator receives the integrated magnetic field control command, its internal controller first parses the command. This command is a time series, containing the target magnetic field vector at control steps (e.g., 3 seconds) over a future period (e.g., 180 seconds). The formula is shown below: , In this formula, Is The target magnetic field vector to be achieved at any given time is explicitly specified by the command. Based on these target components, the controller calculates in real time the target current values ​​to be applied to the three sets of coils. Subsequently, using a high-precision programmable DC power supply, the controller adjusts the output with an extremely fast response speed (typically less than 3 seconds) to drive the three sets of coils to generate the required current.

[0056] By precisely replicating the time-varying magnetic field program planned by the integrated magnetic field control command, the controller continuously updates the output current according to the timestamps and corresponding magnetic field vectors in the command. Therefore, the magnetic field throughout the sintering furnace is no longer a fixed value, but dynamically changes according to a pre-optimized program. For example, the command might require applying a magnetic field along the timeline from the 1st to the 30th second. A 1.2 Tesla stabilizing magnetic field is applied along the axis to reinforce orientation; from 31 to 35 seconds, a magnetic field is applied along the axis. A 0.5 Tesla reverse pulse is applied to the axis to break certain unfavorable magnetic domain structures; between seconds 36 and 60, in... A tiny alternating ripple field with a frequency of 50 Hz and an amplitude of 0.05 Tesla is superimposed on the main magnetic field of the triaxial axis to promote the uniform distribution of grain boundary phases. In this way, the triaxial magnetic field generator can accurately execute complex intervention operations involving changes in intensity, direction, and waveform, physically realizing the optimal strategy derived by the digital twin, thereby completing the execution link in the entire predictive control closed loop.

[0057] Based on the same inventive concept, the present invention also provides an in-situ monitoring and predictive control system for the magnetic properties of permanent magnets during the sintering process, for implementing the above-mentioned method, the system comprising: The in-situ monitoring module is configured to acquire in-situ monitoring data of the permanent magnet billet in the sintering furnace and extract dynamic trend parameters from it. The digital twin and prediction module has a built-in dynamic digital twin model, which is configured to perform model calibration and rolling prediction based on in-situ monitoring data and dynamic trend parameters, and output the predicted trajectory of magnetic property evolution. The intelligent decision-making module is configured to solve for the predicted trajectory and optimization function based on the evolution of magnetic properties, and generate comprehensive magnetic field control commands. The vector magnetic field execution module is configured to receive and execute comprehensive magnetic field control commands. The in-situ monitoring module, digital twin and prediction module, intelligent decision-making module and vector magnetic field execution module are connected in sequence to form a closed-loop control loop.

[0058] In this embodiment, to verify the feasibility and effectiveness of the present invention in actual sintering processes, it is applied to the sintering process of N52 grade neodymium iron boron (NdFeB) permanent magnets. The goal of this process is to densify the permanent magnet blank at high temperatures while obtaining an intrinsic coercivity of not less than 1200 kA / m. ) and the highest possible magnetic energy product ( This ensures the consistency of product performance within a batch. This embodiment will describe in detail the complete operation process within a control cycle and provide specific engineering parameters and test data to demonstrate the derivation process between various technical features.

[0059] In this embodiment, the monitoring and control system is activated during a critical heat preservation sintering stage (furnace temperature 850°C). The following will be described in terms of time points. Starting with this, we will explain a complete predictive control cycle.

[0060] First of all, At any given moment, the in-situ monitoring module is activated to perform a magnetic property measurement in order to obtain in-situ monitoring data and extract dynamic trend parameters.

[0061] The system applies a composite pulsed magnetic field sequence to the permanent magnet billet located in the high-temperature zone of the sintering furnace. This sequence includes a high-intensity main pulse with an amplitude of 1.8 Tesla (T) to drive the magnetization of the billet; after the main pulse ends, a reverse probe sub-pulse of -0.5T is applied after a 10-millisecond interval. The entire sequence lasts for 45 ms. A high-speed data acquisition system acquires the compensated induction coil voltage signal at a sampling rate of 5 MS / s. .

[0062] The system then processes the signal using the formula: The system constant The calibration is 2.5e-7 Wb / V, and the magnetic flux is calculated. Then, using the formula: Geometric factors based on billet size The average magnetization M(t) of the material is calculated from the given value of 1.8e⁴ T / Wb. Combined with the synchronously recorded excitation magnetic field H(t), a dynamic hysteresis loop segment is constructed, and the magnetization is extracted from it. Instantaneous intrinsic coercivity at any moment The value is 785 kA / m. This is the in-situ monitoring data for this study.

[0063] Simultaneously, the system extracts dynamic trend parameters from the magnetic response waveform M(t). This is achieved through the formula: The change in magnetization was measured within a 2ms time window following the rising edge of the main pulse. The first dynamic trend parameter is calculated to be 1.3T. =1.3T / 2ms = 650T / s. (From the formula:) The demagnetization curve segment of the current cycle Compared with the previous cycle By comparing the two curves, the area enclosed by them is calculated, and the second dynamic trend parameter is obtained. It is 0.035.

[0064] Table 1 Data table of magnetic property monitoring and trend parameter extraction ( time)

[0065] In Table 1, the system constants =2.5e-7Wb / V is calculated based on actual measurements using standard samples using an offline calibration method. Geometric factor =1.8e4T / Wb is a volume calculated based on the actual dimensions of the permanent magnet blank in the embodiment (e.g., diameter 10mm, height 12mm). Then, substitute into the formula Calculated (where Second dynamic trend parameter =0.035 is calculated based on the area method, comparing the demagnetization curves of the current and previous cycles within the magnetic field range. It is calculated by inner integral.

[0066] Next, the system performs real-time calibration of the dynamic digital twin model based on the above data, and measures the instantaneous coercivity. Input model, model in current internal parameters The first predicted value of the output is kA / m. Simultaneously, dynamic trend parameters... and As a constraint, substitute it into the cost function Optimize accordingly. Assign weights. The optimization solver iteratively adjusts the variable parameters within the model. (e.g., simulating grain size, orientation, etc.) to make Value less than After optimization, a calibrated model is obtained.

[0067] Then, the system uses the calibrated model to perform rolling prediction and optimization decisions, setting the prediction time domain. Seconds. The proposed process parameters are as follows: within the next 180 seconds, the temperature will linearly increase from 850℃ to 860℃, and the applied magnetic field will be along... A constant magnetic field of 1.2T is applied to the shaft (this is the baseline scheme). Simulations using a calibrated model are performed to obtain the predicted trajectory of the magnetic property evolution of the baseline scheme. The prediction is made after 180 seconds. It will reach 1150 kA / m. The value was 49.5 MGOe, which is below the target.

[0068] The intelligent decision-making module then generates two candidate magnetic field control schemes and utilizes a multi-objective optimization function. (in for , An evaluation is conducted for the process fluctuation penalty item.

[0069] Candidate Solution 1: In On a 1.2T main magnetic field, a magnetic field along the axis is superimposed. An alternating magnetic field with an axis, amplitude of 0.1T, and frequency of 20Hz.

[0070] Candidate Solution 2: In At 1 second, a -0.4T Z-axis reverse pulse is applied for 5 seconds, while the main magnetic field of 1.2T is maintained for the rest of the time.

[0071] The system performs simulation evaluations on each scheme, and the results are shown in Table 2.

[0072] Table 2 Predictive Control and Optimization Decision Data Table ( to (s-time domain)

[0073] In Table 2, "Process Fluctuation Penalty Item" The calculation basis for " is the aforementioned optimization function" Item. Taking the baseline scheme as an example, its calculation process is as follows: First, extract the coercivity within the next 180 seconds from the predicted trajectory of the scheme. and remanence The predicted value sequence; calculate their variances relative to their respective mean values ​​over the entire time domain; then substitute them into the formula. Perform discrete integral summation, where and According to the method settings, This is the simulation step size. Candidate schemes 1 and 2... The value varies accordingly due to the different fluctuations in the predicted trajectory caused by the interference of its magnetic field.

[0074] According to Table 2 and Figure 3 As shown, the comprehensive evaluation value of candidate scheme 2 The highest-ranking option was selected as the target solution. (Refer to...) Figure 3 The system then parses this into specific integrated magnetic field control commands and sends them to the vector magnetic field actuator. This command requires the triaxial magnetic field generator to... to Generated during seconds The magnetic field of T; in to Generated during seconds The magnetic field of T; in to Recover within seconds The magnetic field of T.

[0075] Finally, At the second mark, the system performs another in-situ measurement to obtain the true coercivity of the physical billet. The measured value is 1208 kA / m. This measured value is compared with the predicted value of the target scheme. By comparing kA / m, the model error can be obtained. kA / m. The system uses this error data, through the formula... (Set learning rate) The empirical coefficients in the dynamic digital twin model are fine-tuned. This learning process continues across multiple control cycles, continuously improving the model's prediction accuracy.

[0076] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for in-situ monitoring and control of magnetic properties during the sintering process of permanent magnets, characterized in that, The method includes: Acquire the magnetic response signal of the permanent magnet billet inside the sintering furnace to generate in-situ monitoring data; Dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank are extracted from in-situ monitoring data. Based on in-situ monitoring data and dynamic trend parameters, the dynamic digital twin model simulating the dynamic relationship between magnetic properties and process parameters is calibrated in real time to obtain the calibrated model. By performing rolling prediction calculations using a calibrated model, a magnetic property evolution prediction trajectory characterizing the direction of magnetic property changes in future time periods is generated. Based on the optimization function for optimizing future comprehensive magnetic properties and the predicted trajectory of magnetic property evolution, the comprehensive magnetic field control command is solved and generated. The integrated magnetic field control command is sent to the vector magnetic field actuator to drive it to perform the corresponding magnetic field intervention operation.

2. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The process of acquiring the magnetic response signal of the permanent magnet blank in the sintering furnace and generating in-situ monitoring data includes: applying a short-time pulsed magnetic field sequence containing at least two sub-pulses with different magnetic field characteristics to the permanent magnet blank; acquiring the complete magnetic response waveform generated by the permanent magnet blank in response to the short-time pulsed magnetic field sequence in real time; performing integration and conversion processing on the complete magnetic response waveform to parse out the in-situ monitoring data containing instantaneous magnetic performance parameters.

3. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The step of extracting dynamic trend parameters reflecting the rate of change of the internal state of the permanent magnet blank from the in-situ monitoring data includes: calculating the rate of change of magnetization intensity within a specific time window from the complete magnetic response waveform to obtain a first dynamic trend parameter; quantifying the shape difference of the complete magnetic response waveform in a specific interval between the current measurement cycle and the previous measurement cycle to obtain a second dynamic trend parameter; and combining the first dynamic trend parameter and the second dynamic trend parameter to form the dynamic trend parameter.

4. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The real-time calibration of the dynamic digital twin model based on in-situ monitoring data and dynamic trend parameters to obtain a calibrated model includes: inputting the instantaneous magnetic property parameters from the in-situ monitoring data into the dynamic digital twin model to calculate the predicted value output by the model; applying constraints on the evolution gradient of the internal state variables of the dynamic digital twin model through the dynamic trend parameters; and adjusting the internal variable parameters in the dynamic digital twin model in reverse using an optimization algorithm that minimizes the difference between the predicted value and the instantaneous magnetic property parameters to complete the real-time calibration and obtain the calibrated model.

5. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The step of generating a magnetic property evolution prediction trajectory by performing rolling prediction calculations through a calibrated model includes: obtaining a process parameter plan that defines the planned path of process parameters in the future time period and inputting it into the calibrated model; performing forward simulation based on the calibrated model and the process parameter plan to obtain the predicted values ​​of magnetic property parameters at multiple consecutive time points in the future; and connecting the predicted values ​​of magnetic property parameters at multiple consecutive time points in the future in chronological order to construct the magnetic property evolution prediction trajectory.

6. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The process involves optimizing a function based on future comprehensive magnetic properties and, based on the predicted trajectory of magnetic property evolution, solving for and generating comprehensive magnetic field control instructions. This includes: generating multiple different candidate magnetic field control schemes and inputting each scheme into the calibrated model to obtain multiple corresponding candidate predicted trajectories; evaluating the multiple candidate predicted trajectories using the optimization function to select the target scheme with the optimal future comprehensive magnetic property index; and parsing the target scheme into specific instructions containing the time-varying program of magnetic field strength, direction, and waveform, thus forming the comprehensive magnetic field control instructions.

7. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The method further includes an online learning step: after performing the magnetic field intervention operation, new in-situ monitoring data is acquired in the next measurement cycle; the new in-situ monitoring data is compared with the expected data of the magnetic property evolution prediction trajectory at the corresponding time to generate model error data; The learnable parameters of the dynamic digital twin model are fine-tuned using the model error data to continuously improve the model's prediction accuracy.

8. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 6, characterized in that, The optimization function is configured to maximize the predicted maximum magnetic energy product at the end of the prediction time domain as the main objective, while simultaneously using the stability of changes in coercivity and remanence in the prediction time domain as constraints for collaborative optimization.

9. The method for in-situ monitoring and control of magnetic properties during the sintering process of a permanent magnet according to claim 1, characterized in that, The vector magnetic field actuator is a three-axis magnetic field generator. The integrated magnetic field control command drives the three-axis magnetic field generator to generate an intervention magnetic field whose intensity and direction can be independently and programmatically controlled in three-dimensional space.

10. A system for in-situ monitoring and predictive control of magnetic properties during the sintering process of permanent magnets, characterized in that, The system for implementing the method as described in any one of claims 1 to 9 includes: The in-situ monitoring module is configured to acquire in-situ monitoring data of the permanent magnet billet in the sintering furnace and extract dynamic trend parameters from it. The digital twin and prediction module has a built-in dynamic digital twin model, configured to perform model calibration and rolling prediction based on the in-situ monitoring data and the dynamic trend parameters, and output the predicted trajectory of magnetic property evolution. The intelligent decision-making module is configured to solve the predicted trajectory and optimization function based on the magnetic property evolution to generate comprehensive magnetic field control commands; The vector magnetic field execution module is configured to receive and execute the comprehensive magnetic field control command; wherein the in-situ monitoring module, digital twin and prediction module, intelligent decision-making module and vector magnetic field execution module are connected in sequence to form a closed-loop control loop.

Images