Control method for improving single weight consistency of neodymium iron boron green body

Abstract

The invention discloses a control method for improving single weight consistency of neodymium-iron-boron green bodies, and relates to the technical field of neodymium-iron-boron magnetic material production, which comprises the following steps: before powder filling, according to the particle size distribution and flow characteristics of superfine neodymium-iron-boron powder, the airflow velocity is preset, so that the airflow is kept stable in the filling process, and the powder can be effectively disturbed; therefore, the filling uniformity is improved. Through real-time monitoring, intelligent analysis and dynamic regulation and control, the single weight consistency of the neodymium iron boron green body is improved, and the problem of flow velocity mutation in the powder filling process is solved. Uniform filling is ensured by presetting the airflow velocity, the powder flowing state is intelligently judged in combination with data collection, feature extraction and a deep learning model, and precise control is achieved. When it is detected that the powder flow speed is abnormally increased, the system adjusts the airflow intensity in a self-adaptive mode and temporarily reduces the feeding amount, and powder accumulation is prevented. The powder filling uniformity is optimized, the green body quality, the magnetic performance stability and the production efficiency are improved, and the rejection rate and material waste are reduced.

Description

Technical Field

[0001] The invention relates to the technical field of production of NdFeB magnetic materials, and in particular to a control method for improving the unit weight consistency of NdFeB green compacts. Background Art

[0002] Improving the unit weight consistency of NdFeB green billets means keeping the weight of each green billet (i.e., unsintered NdFeB magnet blank) as consistent as possible during the production process to reduce subsequent processing errors and product performance fluctuations caused by weight deviations. This goal is usually achieved by optimizing the powder metallurgy process, including precisely controlling the particle size distribution of the alloy powder, optimizing the molding process parameters (such as pressing pressure and filling density), improving the automatic feeding and batching system, and reducing human operation errors. Improving unit weight consistency can not only improve the performance stability of the final magnet product, but also reduce material waste, improve production efficiency, and optimize the consistency and yield of subsequent processing (such as sintering and machining).

[0003] The existing technology has the following deficiencies:

[0004] In the prior art, a trace amount of airflow is usually introduced during the powder filling process to cause slight disturbance, and a preset airflow velocity is used to ensure the uniformity of the powder distribution and avoid powder scattering or unstable filling due to excessive airflow. However, during the powder filling process, ultrafine NdFeB powder may form a temporary stagnation area inside the mold due to the powder accumulation effect. When a certain critical point is reached, the powder may suddenly change from a slow flow state to a fast flow state, causing a large amount of powder to flow into certain areas of the mold in a short period of time, resulting in local overfilling, while other areas may still be underfilled. Due to uneven powder filling, the powder density distribution in the mold deviates. In the subsequent pressing process, the powder quality in some areas exceeds the standard, while the powder in other areas is insufficient, which ultimately leads to a significant deviation in the unit weight of the green body. This deviation not only affects the quality consistency of the green body, but may also lead to problems such as uneven pressing density, deformation during sintering, and instability of the final magnetic properties.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a control method for improving the consistency of the single weight of NdFeB green billets, and to improve the consistency of the single weight of NdFeB green billets through real-time monitoring, intelligent analysis and dynamic regulation, so as to solve the problems of sudden change in flow rate, overfilling or underfilling during powder filling. Uniform filling is ensured by presetting the air flow rate, and the powder flow state is intelligently judged by combining data acquisition, feature extraction and deep learning models to achieve precise control. When it is detected that the powder flow rate is abnormally accelerated, the system adaptively adjusts the air flow intensity and temporarily reduces the feed amount to prevent powder accumulation. The present invention optimizes the uniformity of powder filling, improves the quality of green billets, magnetic property stability and production efficiency, reduces the scrap rate and material waste, so as to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a control method for improving the unit weight consistency of NdFeB green billets, comprising the following steps:

[0008] Before powder filling, the air flow rate is pre-set according to the particle size distribution and flow characteristics of the ultrafine NdFeB powder, so that the air flow can remain stable during the filling process and effectively disturb the powder to improve filling uniformity;

[0009] As the powder begins to fill the mold, the filling process is monitored in real time, filling data is acquired in real time, and the dynamic process of powder from entering the mold to distribution and molding is fully recorded;

[0010] After all the collected real-time data are unified and summarized, the acquired data are pre-processed to form a complete and analyzable data set;

[0011] Through feature extraction algorithms, feature vectors reflecting changes in powder flow rate are screened and extracted from the data set, and the extracted feature vectors are comprehensively analyzed to quantify the stability of powder filling.

[0012] The analyzed feature vector is input into a pre-trained deep learning model, which calculates and predicts the powder flow state in real time and outputs a judgment result to determine whether the powder flow rate in the current mold is stable.

[0013] When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, it dynamically reduces the local airflow intensity in the area where the flow rate is accelerated to suppress the entry of excessive powder; at the same time, it temporarily reduces the powder feed amount to reduce the instantaneous powder inflow and prevent local powder accumulation.

[0014] Preferably, during the powder filling process, after the collected real-time data are unified and summarized, the data needs to be pre-processed to ensure its integrity, accuracy and analyzability, and ultimately form a high-quality data set. The specific steps are as follows:

[0015] First, the data acquisition system acquires key data from multiple sensors in real time during the powder filling process and synchronizes the data from different sources to ensure that the data corresponds correctly on the same timeline.

[0016] Next, we enter the data cleaning phase to identify and remove outliers, remove noise signals, and complete missing data to improve data continuity and reliability.

[0017] Then, the data is normalized and standardized, converting data of different units and scales into a unified standard range for subsequent analysis and modeling;

[0018] Subsequently, the data is formatted and structured, and the data from different sensors is stored in a database or data stream according to logical classification to ensure its accessibility and efficient call;

[0019] Ultimately, a complete data set is generated for real-time monitoring of powder filling status, trend analysis, anomaly detection, and intelligent control optimization, ensuring the stability of the filling process, improving the consistency of green body quality, and providing data support for subsequent production optimization.

[0020] Preferably, a feature extraction algorithm is used to screen and extract feature vectors reflecting changes in powder flow rate from a data set. The extracted feature vectors include the rebound effect of the powder caused by compression during the filling process and the degree of discreteness of the powder particles during the filling process. The rebound effect of the powder caused by compression during the filling process and the degree of discreteness of the powder particles during the filling process are comprehensively analyzed under a detection window to generate a powder compression rebound reference value and a particle discreteness reference value, respectively. The powder compression rebound reference value and the particle discreteness reference value are used to quantify the stability of the powder filling.

[0021] Preferably, the specific steps of comprehensively analyzing the rebound effect of the powder caused by compression during the filling process in the detection window to generate a powder compression rebound reference value are as follows:

[0022] During the powder filling process, the powder in a local area will be subjected to compression force, resulting in a reduction in the interparticle gap and an increase in local density, forming a compressed state. In order to quantify the degree of powder compression, the powder compression effect factor is defined to measure the compression effect of the powder during the filling process. The calculation expression is as follows:

[0023] , where CEF is the powder compression effect factor, H pre The loose packing height of the powder before filling into the mold, H post is the final height of the powder in the mold after compression, P internal is the local contact pressure inside the powder particles, P ambientis the ambient pressure inside the mold;

[0024] After powder filling, the powder will rebound due to the stress release between particles, resulting in a local decrease in filling density and a sudden change in flow rate. To quantify the rebound phenomenon, the powder compression rebound reference value is calculated to measure the rebound strength of the powder caused by compression. The calculation expression is as follows:

[0025] , where PCR is the powder compression rebound reference value, R exp is the expansion recovery rate of the powder, F shear It is the shear resistance factor of the powder, which describes the degree to which the friction and shearing between powder particles inhibit rebound.

[0026] Preferably, the specific steps of comprehensively analyzing the dispersion degree of powder particles during the filling process in the detection window to generate a particle dispersion reference value are as follows:

[0027] During the powder filling process, the distribution state of the particles is analyzed. First, the spatial distribution characteristics of the powder particles are constructed to identify the arrangement of the particles inside the mold. The detection window is divided into multiple micro-areas. The particle density changes in each micro-area are detected using laser scanning technology. The density ratio of the micro-area to its neighboring micro-areas is calculated to obtain the local discrete offset rate. The calculation expression is as follows:

[0028] , where LDO i is the local discrete offset rate of element i, D i is the particle density within the target element i, D j is the particle density of neighboring element j, N(i) is the set of neighboring elements around target element i, n is the number of neighboring elements, max(D j ) is the maximum value of the density of adjacent infinitesimal elements, and ∈ is a small constant to prevent the denominator from being zero;

[0029] Get the local discrete offset rate LDO of all micro-element areas i Finally, the powder filling state in the entire detection window is integrated to generate a reference value of particle dispersion. The generation formula is as follows:

[0030] , where PDI is the reference value of particle dispersion, m is the total number of microelements in the detection window, and W i is the particle motion intensity weight of the target element I, and λ is a smoothing factor to avoid the denominator being too small.

[0031] Preferably, the analyzed powder compression rebound reference value and particle discreteness reference value are input into a pre-trained deep learning model, and the powder flow rate variation coefficient is generated by the deep learning model. The powder flow rate variation coefficient is used to intelligently evaluate whether the powder flow rate in the current mold is stable and within the normal range.

[0032] Preferably, the powder flow rate variation coefficient generated when the pre-trained deep learning model is used to intelligently predict whether the current powder flow rate in the mold is stable is compared with a pre-set reference threshold value of the powder flow rate variation coefficient to classify whether the current powder flow rate in the mold is stable. The classification steps are as follows:

[0033] If the powder flow rate variation coefficient is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating; if the powder flow rate variation coefficient is less than or equal to the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is normal.

[0034] Preferably, when the deep learning model recognizes that the powder flow rate in the current mold is accelerating, the specific steps of dynamically reducing the local airflow intensity in the area where the flow rate is accelerating and temporarily reducing the powder feed amount are as follows:

[0035] When the deep learning model detects that the powder flow rate variation coefficient PFRV is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating, which will cause local overfilling. The local airflow intensity in the corresponding area is dynamically reduced to suppress the rapid entry of powder. The airflow intensity adjustment formula is as follows:

[0036] , where Q air-new is the new airflow intensity value after dynamic adjustment, Q air-initial is the initial local airflow intensity value before adjustment, PFRV ref is the reference threshold value of the coefficient of variation of powder flow rate, e is the natural base, and α is the adjustment sensitivity factor;

[0037] While dynamically reducing the airflow intensity in a local area, it is also necessary to temporarily adjust the powder feed rate. That is, when the risk of rapid changes in powder flow rate is found, the powder feed rate should be reduced in a timely manner. The feed rate adjustment formula is as follows;

[0038] , where F powder-new It is the dynamically adjusted powder feeding amount, which is used to control the amount of powder delivered into the mold in real time. powder-initial is the powder feeding reference value under normal and stable operation, β is the powder feeding amount adjustment coefficient, and γ is the feeding sensitivity index;

[0039] When the deep learning model recognizes that the powder flow rate has returned to the normal range, it gradually restores the local airflow intensity to the initial set value and synchronously adjusts the powder feed amount to return it to a stable feeding state to ensure the uniformity and continuity of the powder filling process.

[0040] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0041] The present invention effectively improves the consistency of the unit weight of NdFeB green billets through real-time monitoring, intelligent analysis, and dynamic control, and solves problems such as sudden changes in flow rate and local overfilling or underfilling caused by the powder accumulation effect during the powder filling process. By presetting an appropriate airflow rate, powder scattering and filling instability are reduced while ensuring uniform powder filling. Real-time data acquisition and preprocessing technology are used to comprehensively record the powder filling dynamics and construct a high-quality data set for subsequent intelligent analysis. A feature extraction algorithm is used to screen and quantify changes in powder flow rate, and a deep learning model is combined to intelligently judge the powder filling status to ensure the stability of the powder flow rate. When an abnormal acceleration of the powder flow rate is detected, the system adaptively adjusts the airflow disturbance intensity and temporarily reduces the powder feed rate to accurately control the powder filling process and avoid local powder accumulation and underfilling. Through intelligent, dynamic, and precise control, the present invention significantly improves the consistency of green billet quality and reduces density deviations during subsequent pressing and sintering processes, thereby improving the dimensional accuracy and magnetic property stability of NdFeB magnets. It also reduces the scrap rate and material waste in the production process, improving production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0043] Figure 1 The present invention is a flow chart of a control method for improving the unit weight consistency of NdFeB green bodies. DETAILED DESCRIPTION

[0044] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0045] The present invention provides Figure 1The control method for improving the unit weight consistency of NdFeB green compacts shown in the figure comprises the following steps:

[0046] Before powder filling, the air flow rate is pre-set according to the particle size distribution and flow characteristics of the ultrafine NdFeB powder, so that the air flow can remain stable during the filling process and effectively disturb the powder to improve filling uniformity;

[0047] Before powder filling, the airflow rate is pre-set according to the particle size distribution and flow characteristics of the ultrafine NdFeB powder. This is done by analyzing the powder's particle size, particle morphology, surface characteristics, fluidity, and electrostatic effects before filling begins, and selecting appropriate airflow parameters (such as speed, direction, and pressure) to optimize the filling process. Its purpose is to ensure that the powder maintains a stable flow during the filling process while being sufficiently disturbed to improve filling uniformity. A stable airflow can prevent powder scattering or unstable filling, while appropriate disturbance can destroy the electrostatic adsorption and cohesion between powder particles, prevent powder accumulation or bridging, and improve the uniformity of powder density in the filling area. In addition, a reasonable airflow rate setting can also reduce local powder flow rate mutations and reduce local over- or underfilling caused by powder accumulation and collapse during the filling process, thereby ensuring the consistency of the green unit weight and improving the dimensional accuracy and magnetic stability of the final product.

[0048] A micro-flow of air is used to disperse or vibrate the powder, breaking the cohesive forces and electrostatic effects between the powders and helping to evenly distribute the powder within the mold. This preset airflow is typically moderate, as this prevents powder scattering and airflow disturbances within the mold, but it also cannot be too low, as this can make it difficult to overcome adhesion or agglomeration of ultrafine powders.

[0049] As the powder begins to fill the mold, the filling process is monitored in real time, filling data is acquired in real time, and the dynamic process of powder from entering the mold to distribution and molding is fully recorded;

[0050] During the powder filling process, the filling process is monitored in real time, that is, high-precision sensors (such as optical monitoring, laser ranging, electrostatic sensors, ultrasonic density scanning, image acquisition devices, etc.) are used to continuously collect key parameters such as the flow state, filling rate, and distribution uniformity of the powder, and obtain dynamic data after the powder enters the mold in real time. The purpose of this process is to comprehensively record the entire process from the powder entering the mold to the final filling completion, ensure the traceability of the production process, and provide accurate data support for subsequent intelligent regulation. Through this monitoring mechanism, it is possible to detect in real time whether the powder filling is uniform, whether there is local accumulation or insufficient filling problems, and when the powder filling state is abnormal (such as sudden change in flow rate, accumulation collapse, etc.), the filling parameters can be quickly adjusted, such as dynamically adjusting the airflow disturbance, feeding rate, etc., to improve the consistency and stability of the filling, and ultimately ensure the quality uniformity and magnetic property stability of the green body.

[0051] After all the collected real-time data are unified and summarized, the acquired data are pre-processed to form a complete and analyzable data set;

[0052] During the powder filling process, after the collected real-time data is unified and summarized, it is necessary to pre-process the data to ensure its completeness, accuracy, and analyzability, and ultimately form a high-quality data set. The specific steps are as follows:

[0053] First, the data acquisition system acquires key data from the powder filling process, such as powder flow rate, filling density, and local distribution, in real time from multiple sensors (such as optical cameras, laser rangefinders, electrostatic sensors, and ultrasonic density scanners). The system then synchronizes the data from these different sources to ensure that they are aligned correctly on the same timeline. Next, the data cleaning phase identifies and removes outliers, removes noise, and fills any missing data (e.g., using interpolation or mean filling) to improve data continuity and reliability. The data is then normalized and standardized, converting data of varying units and scales to a unified standard range for subsequent analysis and modeling. Subsequently, the data is formatted and structured for storage, with data from different sensors logically categorized (e.g., by chronological order, spatial distribution, or process batch) and stored in a database or data stream processor to ensure accessibility and efficient retrieval. Ultimately, the system generates a comprehensive data set for real-time monitoring of the powder filling state, trend analysis, anomaly detection, and intelligent control optimization. This ensures the stability of the filling process, improves green compact quality consistency, and provides data support for subsequent production optimization.

[0054] Through feature extraction algorithms, feature vectors reflecting changes in powder flow rate are screened and extracted from the data set, and the extracted feature vectors are comprehensively analyzed to quantify the stability of powder filling.

[0055] Through the feature extraction algorithm, the characteristic vector reflecting the change of powder flow rate is screened and extracted from the data set. The extracted characteristic vector includes the rebound effect caused by compression of the powder during the filling process and the degree of discreteness of the powder particles during the filling process. The rebound effect caused by compression of the powder during the filling process and the degree of discreteness of the powder particles during the filling process are comprehensively analyzed under the detection window to generate the powder compression rebound reference value and the particle discreteness reference value respectively. The powder filling stability is quantified by the powder compression rebound reference value and the particle discreteness reference value.

[0056] Increased rebound caused by compression during powder filling can indicate a sudden shift from slow to fast flow. This is primarily due to stress release between powder particles and the transient adjustment of their packing structure. During the filling process, powders are typically slightly compressed by the mold walls or external airflow, resulting in a decrease in interparticle spacing and an increase in local density. However, due to the high cohesion and elasticity of ultrafine powders, when the external force decreases or the powder reaches a critical packing state, the elastic potential energy within the powder is suddenly released, disrupting the tight packing between particles and causing rapid rebound of the powder as a whole. This in turn triggers a transient loosening of the powder and accelerated flow in localized areas. This phenomenon is similar to the rebound effect of elastic materials. When the powder packing density is too high and there is a lack of adequate flow buffering, localized areas of powder may experience a sudden release of stress, causing accelerated particle diffusion toward lower-density areas, resulting in a short-term high flow state. Furthermore, if the powder is subject to minimal airflow disturbances or external forces during the rebound process, the inertial effects between particles can further amplify the sudden change in flow rate, leading to sudden flow instability during the filling process. Therefore, the increased rebound of powder after compression is often a precursor signal for the powder to develop from slow flow to fast flow. In production control, it is necessary to monitor and dynamically adjust the filling parameters in real time to prevent the occurrence of local uneven filling or excessive accumulation problems.

[0057] The specific steps for comprehensively analyzing the rebound effect of powder caused by compression during the filling process under the detection window to generate the powder compression rebound reference value are as follows:

[0058] During the powder filling process, the powder in a local area will be subjected to compression force, resulting in a reduction in the interparticle gap and an increase in local density, forming a compressed state. To quantify the degree of powder compression, the powder compression effect factor is defined to measure the compression effect of the powder during the filling process. The powder compression effect factor is based on the ratio of the initial filling height of the powder to the height after compression, as well as the microscopic pressure changes between the particles within the powder. The calculation expression is as follows:

[0059] , where CEF is the powder compression effect factor, Hpre It is the loose pile height of powder before filling into the mold, reflecting the natural state of the powder before filling. post It is the final height of the powder in the mold after being compressed, indicating the degree of compaction of the powder. internal It is the local contact pressure inside the powder particles, which is usually determined by the particle distribution, particle shape and external compression force of the powder. ambient is the ambient pressure inside the mold, that is, the external pressure state of the powder during the filling process;

[0060] This step measures the overall compression of the powder. Large variations in the powder height indicate significant compression. Interparticle stress effects are also corrected to account for microscopic stress variations within the powder. A large powder compression factor indicates significant compression during filling, potentially leading to significant rebound, which can affect flow rate stability.

[0061] After powder filling, the powder will rebound due to the stress release between particles, resulting in a local decrease in filling density and a sudden change in flow rate. To quantify the rebound phenomenon, the powder compression rebound reference value is calculated to measure the rebound strength of the powder caused by compression. The calculation expression is as follows:

[0062] , where PCR is the powder compression rebound reference value, R exp is the expansion recovery rate of the powder, which indicates the degree of rebound of the powder after filling. shear It is the shear resistance factor of the powder, which describes the degree to which the friction and shearing between powder particles inhibit rebound.

[0063] A high PCR indicates that the powder rebounds strongly after compression, causing the powder to loosen and flow faster in local areas, potentially causing sudden changes in filling rate or localized filling unevenness. Conversely, a low PCR indicates a weaker powder rebound effect, a stable filling state, and no sudden changes in flow rate. Therefore, PCR can be used as a key parameter to predict sudden changes in powder flow rate. In intelligent filling control systems, it can be used to adjust airflow disturbances or feed rate in advance to ensure the stability and consistency of the filling process.

[0064] The powder compression-rebound reference value, generated by comprehensively analyzing the rebound effect caused by compression during the filling process within the detection window, indicates that the powder may suddenly shift from a slow flow to a fast flow trend. Conversely, a smaller value indicates a relatively stable flow trend, with no sudden acceleration. The fundamental mechanism of this phenomenon lies in the change in flow state caused by the compression, deformation, and rebound release of powder particles during the filling process. When powder is filled into the mold, local areas of powder are compressed to a certain extent by gravity, mold wall forces, or airflow disturbances, resulting in a decrease in interparticle spacing and an increase in density. However, during the filling process, as the external force decreases or particles accumulate to a certain level, the powder may experience a rebound effect. This is the sudden release of stored stress under compression, causing a sudden change in the internal particle arrangement of the powder, leading to rapid disaggregation and accelerated flow. When the powder compression rebound reference value is high, it means that the powder is greatly affected by compression and the released rebound energy is strong. At this time, the flow state may change from slow filling to fast flow, and may even cause instantaneous accumulation collapse or uneven filling of the powder. On the contrary, when the powder compression rebound reference value is low, it indicates that the potential energy stored in the powder compression process is small, the structural adjustment between particles is relatively smooth, the filling state is stable, and the powder will not undergo sudden acceleration.

[0065] Excessive dispersion of powder particles during the filling process typically indicates that the powder may suddenly shift from a slow to a fast flow pattern. The root cause is a decrease in the cohesive force between powder particles and a sudden change in the particle distribution. Under normal filling conditions, powder particles rely on interparticle van der Waals forces, electrostatic forces, or mechanical interlocking to maintain a relatively stable flow state. However, when these interactions are weakened or external disturbances (such as enhanced airflow, mold wall slip, vibration impact, etc.) act on the powder, the particles may enter a discrete state, resulting in a sudden increase in localized fluidity and uncontrolled accelerated powder flow into the mold. Especially in high filling density environments, the arrangement of powder particles is restricted. When local particles exceed the critical density or are driven by external forces and suddenly disperse, the powder loses its stable filling pattern and instantly transitions from a slow accumulation state to a high-speed filling state. This may even cause local powder collapse, resulting in uneven filling. In addition, overly discrete powder particles are more likely to form fast flow channels under the action of aerodynamics, accelerating the movement of other particles and causing the overall powder filling rate to exceed expectations. Therefore, excessive dispersion of powder particles is a key signal of a sudden change in powder flow from slow to fast flow, which requires real-time monitoring and dynamic control to prevent uneven filling or local overfilling.

[0066] The specific steps for comprehensively analyzing the dispersion degree of powder particles during the filling process under the detection window to generate a reference value for particle dispersion are as follows:

[0067] During the powder filling process, the distribution state of the particles is analyzed. First, the spatial distribution characteristics of the powder particles are constructed to identify the arrangement of the particles inside the mold. The detection window is divided into multiple micro-areas. The particle density changes in each micro-area are detected using laser scanning technology. The density ratio of the micro-area to its neighboring micro-areas is calculated to obtain the local discrete offset rate. The calculation expression is as follows:

[0068] , where LDO i is the local discrete deviation rate of the infinitesimal element i, which measures the fluctuation degree of the particle density. i is the particle density within the target element i, which refers to the number of particles per unit volume, D j is the particle density of neighboring element j, N(i) is the set of neighboring elements around target element i, n is the number of neighboring elements, max(D j ) is the maximum value of the density of adjacent infinitesimal elements, and ∈ is a small constant to prevent the denominator from being zero;

[0069] This step is used to quantify the local density variation of powder particles in space. i A higher value indicates that the particles in this area are more discrete and the gaps between particles are larger, which means that the powder filling may enter a fast flow trend. i It is low, indicating that the powder particles are evenly distributed in the micro-element area and the flow trend is relatively stable.

[0070] Get the local discrete offset rate LDO of all micro-element areas i Finally, the powder filling state in the entire detection window is integrated to generate a reference value of particle dispersion. The generation formula is as follows:

[0071] , where PDI is the reference value of particle dispersion, m is the total number of microelements in the detection window, and W i is the particle motion intensity weight of the target element i, which indicates the degree of dynamic change of particles in the element area, and λ is a smoothing factor to avoid the denominator being too small.

[0072] This step is used to calculate the overall particle dispersion during the entire filling process, where W i Representing the motion state of particles, such as the velocity gradient or shear stress of a powder, this parameter ensures that areas of high fluidity are given a higher weight, allowing PDI to reflect not only the degree of spatial distribution dispersion but also the flow state of the powder. A high PDI indicates that the arrangement of powder particles tends to be loose, the fluidity is enhanced, and the filling rate may experience sudden changes. A low PDI indicates that the powder is evenly packed, the flow rate is stable, and it is not prone to sudden flow changes.

[0073] The larger the particle discreteness reference value generated after a comprehensive analysis of the degree of discreteness of the powder particles during the filling process within the detection window, the weaker the interaction forces between the powder particles (such as van der Waals forces, electrostatic forces, or mechanical locking), and the particles tend to move more freely, making them more susceptible to external factors (such as airflow disturbances, gravity, vibration, or shear forces) and causing sudden increases in flow rate. When the particle discreteness reference value reaches a certain threshold, the powder may rapidly transition from a slow accumulation or restricted flow state to a high-speed flow state, resulting in a sudden increase in local filling rate, and may even cause uneven filling or local powder collapse. On the contrary, if the particle discreteness reference value is small, it means that the powder particles still maintain a high degree of cohesion and stability, there is still a strong binding force between the particles, the flow rate is relatively stable, and it will not easily enter a rapid flow trend.

[0074] The analyzed feature vector is input into a pre-trained deep learning model, which calculates and predicts the powder flow state in real time and outputs a judgment result to determine whether the powder flow rate in the current mold is stable.

[0075] The analyzed powder compression rebound reference value and particle discreteness reference value are input into a pre-trained deep learning model, and the powder flow rate variation coefficient is generated by the deep learning model. The powder flow rate variation coefficient is used to intelligently evaluate whether the current powder flow rate in the mold is stable and within the normal range.

[0076] A pre-trained deep learning model refers to a deep learning model that has been trained based on a large amount of historical data before the powder filling system is deployed and has generalization capabilities. During the offline training phase, the model uses a large amount of data from the powder filling process, such as powder compression rebound reference value, particle discreteness reference value, flow rate change trend, stacking damage index, flow stability index and other features, combined with supervised learning or unsupervised learning methods to train a model that can predict changes in powder flow rate. During the training process, long short-term memory networks (LSTM), convolutional neural networks (CNN), autoencoders (Autoencoder) or hybrid models (such as CNN-LSTM) are usually used to learn the temporal change pattern of powder flow rate and construct a mathematical relationship that can accurately predict changes in powder flow rate. After multiple rounds of iterative optimization, the model can identify powder flow patterns under different filling conditions and be used for real-time reasoning after deployment to intelligently determine whether the powder flow rate is stable.

[0077] During the online inference phase, a pre-trained deep learning model receives real-time input data, such as reference values for powder compression rebound and particle dispersion, and automatically calculates the powder flow rate variation coefficient. This coefficient reflects the stability of the powder flow rate within the mold and whether it is within the normal range. When the powder flow rate variation coefficient exceeds the set threshold, the system triggers intelligent control mechanisms, such as reducing local airflow intensity and instantaneous feed rate to ensure uniform powder filling. Because the model has learned a large amount of historical filling data during the training phase, it can adapt to the powder flow characteristics of different batches, different particle sizes, and different environmental conditions, achieving accurate predictions and intelligent adjustments. This pre-trained deep learning model can significantly improve the stability of the powder filling process, reduce local accumulation or underfilling, thereby ensuring the consistency of the green body weight and improving the final performance stability of the NdFeB magnet.

[0078] The deep learning model is not limited here. Any machine learning model that can perform a comprehensive analysis of the powder compression rebound reference value PCR and the particle dispersion reference value PDI to generate the powder flow rate variation coefficient PFRV can be used. In order to realize the technical solution of the present invention, the present invention provides a specific implementation method.

[0079] The formula for generating the powder flow rate variation coefficient PFRV is as follows: PFRV = p1·PCR + p2·PDI, where p1 and p2 are the preset proportional coefficients of the powder compression rebound reference value PCR and the particle dispersion reference value PDI, respectively, and both p1 and p2 are greater than 0.

[0080] Preset proportional coefficients p1 and p2 serve as weighting parameters to measure the influence of the powder compression rebound reference value (PCR) and the particle dispersion reference value (PDI) on the powder flow rate variation coefficient (PFRV). These coefficients are typically pre-set or optimized through experiments and data training to ensure that the calculated PFRV accurately reflects the changing trend of the powder flow rate.

[0081] In practical applications, the purpose of pre-setting proportionality coefficients is to adjust the contribution of different parameters to match the characteristics of different powder types. For example, if the compression-rebound effect of certain powders has a greater impact on flow rate changes, the p1 value may be larger. Meanwhile, if particle dispersion has a more significant impact on powder filling stability, the p2 value may account for a larger proportion. Furthermore, these coefficients can be optimized through deep learning models or regression analysis, allowing them to be dynamically adjusted under different operating conditions to improve the accuracy of PFRV calculations. Ultimately, this can be used to intelligently control the powder filling process and ensure the stability of green body quality.

[0082] It can be seen from the powder flow rate variation coefficient that the larger the powder compression rebound reference value generated after comprehensive analysis of the rebound effect caused by compression of the powder during the filling process under the detection window, and the larger the particle discreteness reference value generated after comprehensive analysis of the discreteness of the powder particles during the filling process under the detection window, the larger the powder flow rate variation coefficient generated when the pre-trained deep learning model is used to intelligently predict whether the powder flow rate in the current mold is stable, indicating that the probability of accelerating the powder flow rate of the current mold is greater, and vice versa, the probability of accelerating the powder flow rate of the current mold is smaller.

[0083] The powder flow rate variation coefficient generated by the pre-trained deep learning model when intelligently predicting whether the current powder flow rate in the mold is stable is compared with the pre-set reference threshold of the powder flow rate variation coefficient to classify whether the current powder flow rate in the mold is stable. The classification steps are as follows:

[0084] If the powder flow rate variation coefficient is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating; if the powder flow rate variation coefficient is less than or equal to the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is normal.

[0085] When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, it dynamically reduces the local airflow intensity in the area where the flow rate is accelerating to suppress the entry of excessive powder. At the same time, it temporarily reduces the powder feed rate to reduce the instantaneous powder inflow and prevent local powder accumulation.

[0086] When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, the system takes two measures: dynamically reducing the local airflow intensity in the area with accelerated flow rate and temporarily reducing the powder feed rate. Its core function is to accurately control the powder filling rate and avoid local overfilling, thereby ensuring filling uniformity and improving the quality consistency of the green body.

[0087] During the powder filling process, factors such as powder accumulation, static discharge, and airflow disturbances can cause the powder flow rate to suddenly accelerate, causing powder to rush into certain areas within a short period of time, resulting in localized overfilling. Without timely intervention, this can lead to uneven powder density within the mold, varying levels of localized powder compaction, and even affect subsequent pressing and sintering processes.

[0088] First, dynamically reducing local airflow intensity reduces the powder's flow dynamics in that area, allowing the powder to settle slowly rather than suddenly filling the mold bottom due to excessive airflow. This approach effectively balances the powder filling rate, preventing excessive powder accumulation in some areas and underfilling in others. Simultaneously, the localized airflow adjustment ensures that the remaining areas of the mold maintain appropriate powder flow, preventing impact on overall filling efficiency.

[0089] Secondly, briefly reducing the powder feed rate reduces the amount of powder entering the mold at the source, further suppressing sudden increases in powder flow. Maintaining the original feed rate when the powder flow rate accelerates can lead to overfilling or even overflow of the mold, rendering filling control ineffective. A brief reduction in the feed rate provides a window for the powder flow to self-regulate, allowing it to return to a stable state within a relatively short period of time. Afterward, the feed rate can be gradually restored to avoid affecting overall filling efficiency.

[0090] Furthermore, intelligent control in this step enhances the adaptive nature of powder filling, avoiding issues like uneven filling and powder waste caused by delayed manual intervention. By combining real-time monitoring and prediction with a deep learning model, this method can adjust the powder filling rate at an early stage of sudden changes, reducing the occurrence of anomalies and ensuring more consistent green weight. This ensures the stability of subsequent processing and improves the quality and yield of the final product.

[0091] When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, it dynamically reduces the local airflow intensity in the area where the flow rate is accelerating. At the same time, the specific steps for temporarily reducing the powder feed rate are as follows:

[0092] When the deep learning model detects that the powder flow rate variation coefficient PFRV is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating, which will cause local overfilling. The local airflow intensity in the corresponding area is dynamically reduced to suppress the rapid entry of powder. The airflow intensity adjustment formula is as follows:

[0093] , where Q air-new It is the new airflow intensity value after dynamic adjustment, which is used to adjust the local airflow intensity in real time to accurately control the powder flow rate. air-initial It is the initial local airflow intensity value before adjustment, that is, the preset stable working airflow intensity under normal circumstances, which is used to ensure the basic stability of powder flow and prevent initial accumulation. ref is the reference threshold of the powder flow rate variation coefficient, E is the natural base, and α is the adjustment sensitivity factor used to adjust the airflow adjustment amplitude;

[0094] By dynamically reducing the local airflow intensity in areas where the flow rate accelerates, the risk of a sudden influx of powder into the mold due to a sudden change in flow rate is reduced, thereby suppressing overfilling, ensuring that powder enters the mold evenly, and preventing localized accumulation or uneven density. At the same time, this step uses a nonlinear airflow control mechanism to adaptively adjust the airflow intensity based on the degree of change in powder flow rate. This allows for gentle airflow adjustments when the powder flow rate changes slightly, while rapidly reducing airflow disturbances when the powder flow rate changes dramatically, achieving efficient and precise intelligent control and maintaining the stability and consistency of the filling process.

[0095] While dynamically reducing the airflow intensity in a local area, it is also necessary to temporarily adjust the powder feed rate. That is, when the risk of rapid changes in powder flow rate is found, the powder feed rate should be reduced in a timely manner. The feed rate adjustment formula is as follows;

[0096] , where F powder-new It is the dynamically adjusted powder feeding amount, which is used to control the amount of powder delivered into the mold in real time. powder-initial is the powder feeding reference value under normal and stable operation, β is the powder feeding adjustment coefficient, which is used to determine the magnitude of the reduction in feeding amount, and γ is the feeding sensitivity index, which is used to adjust the nonlinearity of feeding control so that the magnitude of feeding reduction responds exponentially with the change in powder flow rate, thereby enhancing the control accuracy.

[0097] By temporarily reducing the powder feed rate and the instantaneous rate at which powder enters the mold, the dynamic changes in powder filling are controlled at the source, preventing localized powder accumulation or overfilling. This strategy not only responds quickly to sudden changes in powder flow rate and reduces the impact of instantaneous feed rate, but also, combined with airflow control, achieves precise dual control of powder flow rate and feed rate, improving powder filling uniformity and green body weight consistency.

[0098] When the deep learning model recognizes that the powder flow rate returns to the normal range (i.e., the powder flow rate variation coefficient PFRV ≤ PFRV ref ), gradually restore the local airflow intensity to the initial set value, and synchronously adjust the powder feeding amount to return it to a stable feeding state to ensure the uniformity and continuity of the powder filling process.

[0099] The present invention effectively improves the consistency of the unit weight of NdFeB green billets through real-time monitoring, intelligent analysis, and dynamic control, and solves problems such as sudden changes in flow rate and local overfilling or underfilling caused by the powder accumulation effect during the powder filling process. By presetting an appropriate airflow rate, powder scattering and filling instability are reduced while ensuring uniform powder filling. Real-time data acquisition and preprocessing technology are used to comprehensively record the powder filling dynamics and construct a high-quality data set for subsequent intelligent analysis. A feature extraction algorithm is used to screen and quantify changes in powder flow rate, and a deep learning model is combined to intelligently judge the powder filling status to ensure the stability of the powder flow rate. When an abnormal acceleration of the powder flow rate is detected, the system adaptively adjusts the airflow disturbance intensity and temporarily reduces the powder feed rate to accurately control the powder filling process and avoid local powder accumulation and underfilling. Through intelligent, dynamic, and precise control, the present invention significantly improves the consistency of green billet quality and reduces density deviations during subsequent pressing and sintering processes, thereby improving the dimensional accuracy and magnetic property stability of NdFeB magnets. It also reduces the scrap rate and material waste in the production process, improving production efficiency and product quality.

[0100] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0101] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0102] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0103] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0105] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0106] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0107] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0108] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0109] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A control method for improving the unit weight consistency of NdFeB green billets, characterized in that: The following steps are involved: Before powder filling, the air flow rate is pre-set according to the particle size distribution and flow characteristics of the ultrafine NdFeB powder, so that the air flow can remain stable during the filling process and effectively disturb the powder to improve filling uniformity; As the powder begins to fill the mold, the filling process is monitored in real time, filling data is acquired in real time, and the dynamic process of powder from entering the mold to distribution and molding is fully recorded; After all the collected real-time data are aggregated, the acquired data are pre-processed to form a complete and analyzable data set. Through feature extraction algorithms, feature vectors reflecting changes in powder flow rate are screened and extracted from the data set, and the extracted feature vectors are comprehensively analyzed to quantify the stability of powder filling. The analyzed feature vector is input into a pre-trained deep learning model, which calculates and predicts the powder flow state in real time and outputs a judgment result to determine whether the powder flow rate in the current mold is stable. When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, it dynamically reduces the local airflow intensity in the area where the flow rate is accelerated to suppress the entry of excessive powder; at the same time, it temporarily reduces the powder feed amount to reduce the instantaneous powder inflow and prevent local powder accumulation.

2. The control method for improving the unit weight consistency of NdFeB green billets according to claim 1, characterized in that: During the powder filling process, after the collected real-time data is consolidated, it is necessary to pre-process the data to ensure its integrity, accuracy, and analyzability, ultimately forming a high-quality data set. The specific steps are as follows: First, the data acquisition system acquires key data from multiple sensors in real time during the powder filling process and synchronizes the data from different sources to ensure that the data corresponds correctly on the same timeline. Next, we enter the data cleaning phase to identify and remove outliers, remove noise signals, and fill in missing data to improve data continuity and reliability. Then, the data is normalized and standardized, converting data of different units and scales into a unified standard range for subsequent analysis and modeling; Subsequently, the data is formatted and structured, and the data from different sensors is stored in a database or data stream according to logical classification to ensure its accessibility and efficient call; Ultimately, a complete data set is generated for real-time monitoring of powder filling status, trend analysis, anomaly detection, and intelligent control optimization, ensuring the stability of the filling process, improving the consistency of green body quality, and providing data support for subsequent production optimization.

3. The control method for improving the unit weight consistency of NdFeB green billets according to claim 1, characterized in that: Through the feature extraction algorithm, the characteristic vector reflecting the change of powder flow rate is screened and extracted from the data set. The extracted characteristic vector includes the rebound effect caused by compression of the powder during the filling process and the degree of discreteness of the powder particles during the filling process. The rebound effect caused by compression of the powder during the filling process and the degree of discreteness of the powder particles during the filling process are comprehensively analyzed under the detection window to generate the powder compression rebound reference value and the particle discreteness reference value respectively. The powder filling stability is quantified by the powder compression rebound reference value and the particle discreteness reference value.

4. The control method for improving the unit weight consistency of NdFeB green billets according to claim 3, characterized in that: The specific steps for comprehensively analyzing the rebound effect of powder caused by compression during the filling process under the detection window to generate the powder compression rebound reference value are as follows: During the powder filling process, the powder in a local area will be subjected to compression force, resulting in a reduction in the interparticle gap and an increase in local density, forming a compressed state. In order to quantify the degree of powder compression, the powder compression effect factor is defined to measure the compression effect of the powder during the filling process. The calculation expression is as follows: , where CEF is the powder compression effect factor, H pre The loose packing height of the powder before filling into the mold, H post is the final height of the powder in the mold after compression, P internal is the local contact pressure inside the powder particles, P ambient is the ambient pressure inside the mold; After powder filling, the powder will rebound due to the stress release between particles, resulting in a local decrease in filling density and a sudden change in flow rate. To quantify the rebound phenomenon, the powder compression rebound reference value is calculated to measure the rebound strength of the powder caused by compression. The calculation expression is as follows: , where PCR is the powder compression rebound reference value, R exp is the expansion recovery rate of the powder, F shear It is the shear resistance factor of the powder, which describes the degree to which the friction and shearing between powder particles inhibit rebound.

5. The control method for improving the unit weight consistency of NdFeB green bodies according to claim 3, characterized in that: The specific steps for comprehensively analyzing the dispersion degree of powder particles during the filling process under the detection window to generate a reference value for particle dispersion are as follows: During the powder filling process, the distribution state of the particles is analyzed. First, the spatial distribution characteristics of the powder particles are constructed to identify the arrangement of the particles inside the mold. The detection window is divided into multiple micro-areas. The particle density changes in each micro-area are detected using laser scanning technology. The density ratio of the micro-area to its neighboring micro-areas is calculated to obtain the local discrete offset rate. The calculation expression is as follows: , where LDO i is the local discrete deviation rate of element i, D i is the particle density within the target element i, D j is the particle density of neighboring element j, N(i) is the set of neighboring elements around target element i, n is the number of neighboring elements, max(D j ) is the maximum value of the density of adjacent infinitesimal elements, and ∈ is a small constant to prevent the denominator from being zero; Get the local discrete offset rate LDO of all micro-element areas i Finally, the powder filling state in the entire detection window is integrated to generate a reference value of particle dispersion. The generation formula is as follows: , where PDI is the reference value of particle dispersion, m is the total number of microelements in the detection window, and W i is the particle motion intensity weight of target element i, and λ is a smoothing factor to avoid the denominator being too small.

6. The control method for improving the unit weight consistency of NdFeB green bodies according to claim 3, characterized in that: The analyzed powder compression rebound reference value and particle discreteness reference value are input into a pre-trained deep learning model, and the powder flow rate variation coefficient is generated by the deep learning model. The powder flow rate variation coefficient is used to intelligently evaluate whether the current powder flow rate in the mold is stable and within the normal range.

7. The control method for improving the unit weight consistency of NdFeB green bodies according to claim 6, characterized in that: The powder flow rate variation coefficient generated by the pre-trained deep learning model when intelligently predicting whether the current powder flow rate in the mold is stable is compared with the pre-set reference threshold of the powder flow rate variation coefficient to classify whether the current powder flow rate in the mold is stable. The classification steps are as follows: If the powder flow rate variation coefficient is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating; if the powder flow rate variation coefficient is less than or equal to the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is normal.

8. The control method for improving the unit weight consistency of NdFeB green bodies according to claim 1, characterized in that: When the deep learning model recognizes that the powder flow rate in the current mold is accelerating, it dynamically reduces the local airflow intensity in the area where the flow rate is accelerating. At the same time, the specific steps for temporarily reducing the powder feed rate are as follows: When the deep learning model detects that the powder flow rate variation coefficient PFRV is greater than the preset powder flow rate variation coefficient reference threshold, it indicates that the powder flow rate in the current mold is accelerating, which will cause local overfilling. The local airflow intensity in the corresponding area is dynamically reduced to suppress the rapid entry of powder. The airflow intensity adjustment formula is as follows: , where Q air-new is the new airflow intensity value after dynamic adjustment, Q air-initial is the initial local airflow intensity value before adjustment, PFRV ref is the reference threshold value of the coefficient of variation of powder flow rate, e is the natural base, and α is the adjustment sensitivity factor; While dynamically reducing the airflow intensity in a local area, it is also necessary to temporarily adjust the powder feed rate. That is, when the risk of rapid changes in powder flow rate is found, the powder feed rate should be reduced in a timely manner. The feed rate adjustment formula is as follows; , where F powder-new It is the dynamically adjusted powder feeding amount, which is used to control the amount of powder delivered into the mold in real time. powder-initial is the powder feeding reference value under normal and stable operation, β is the powder feeding amount adjustment coefficient, and γ is the feeding sensitivity index; When the deep learning model recognizes that the powder flow rate has returned to the normal range, it gradually restores the local airflow intensity to the initial set value and synchronously adjusts the powder feed amount to return it to a stable feeding state to ensure the uniformity and continuity of the powder filling process.

Images