High-entropy alloy powder and preparation process thereof

Abstract

The invention relates to the technical field of alloy powder preparation, discloses high-entropy alloy powder and a preparation process thereof, and aims at solving the problems that Y2O3 particles are easy to agglomerate, the process depends on experience and the batch stability is poor in a traditional ODS high-entropy alloy process, a preparation method combining multi-scale simulation with digital twinning optimization is constructed, critical ball milling impact energy is simulated through molecular dynamics, and the high-entropy alloy powder is obtained. The method comprises the following steps: judging a ball milling end point in real time by adopting an acoustic emission intelligent monitoring system, and performing internal oxidation in an Ar-(3-10)% O2 atmosphere by utilizing defect engineering to generate Y2O3 nano-phase particles of 5-10nm; monitoring an O + / Ar + spectral line intensity ratio through a fiber optic spectrometer in SPS sintering to intelligently trigger pressure loading; constructing a digital twinborn body to realize virtual iterative optimization; the yield strength of the prepared high-entropy alloy powder reaches 1200-1300 MPa, the ductility is larger than or equal to 12%, and the strength is improved by 20-30% compared with that of a traditional process.

Description

Technical Field

[0001] This invention belongs to the field of alloy powder preparation technology, specifically relating to a high-entropy alloy powder and its preparation process. Background Technology

[0002] High-entropy alloys (HEAs) exhibit excellent mechanical properties, corrosion resistance, and high-temperature stability due to their unique multi-principal element design concept, and have broad application prospects in aerospace, nuclear energy, and chemical industries. To further improve their high-temperature strength and creep resistance, the introduction of oxide nanoparticles for dispersion strengthening (ODS) has become an important research direction.

[0003] However, existing ODS high-entropy alloy preparation technologies still have the following prominent problems: In traditional mechanical alloying (MA) processes, yttrium oxide (Y₂O₃) nanoparticles tend to agglomerate, making it difficult to achieve uniform dispersion in the matrix and resulting in unstable strengthening effects. Key process steps such as ball milling time and sintering parameters rely heavily on empirical settings, lacking real-time process monitoring and endpoint determination methods, leading to poor process repeatability. Furthermore, existing technologies are mostly trial-and-error experiments, lacking pre-design based on multi-scale simulations, making it difficult to accurately predict the formation, distribution, and interfacial behavior of Y₂O₃ with the matrix. Due to the lack of intelligent monitoring and feedback mechanisms, the performance of different batches of products fluctuates significantly, failing to meet the requirements of engineering applications. Traditional processes struggle to establish quantitative relationships between process parameters, microstructure, and macroscopic properties, resulting in long process optimization cycles and high costs.

[0004] Therefore, there is an urgent need to develop a novel ODS high-entropy alloy preparation process that can achieve in-situ controllable generation of Y2O3, intelligent monitoring of the process, and virtual iterative optimization capabilities, in order to solve the above-mentioned technical bottlenecks. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a process for preparing high-entropy alloy powder, comprising the following steps: S1: Based on molecular dynamics model simulation, the critical ball milling impact energy and the target addition amount of yttrium oxide Y2O3 raw material powder required to achieve mechanochemical dissociation of yttrium oxide Y2O3 raw material powder are determined in advance, and the raw material powder is subjected to step vacuum drying treatment. S2: The powder processed in step S1 is placed in a high-energy ball mill for ball milling. Simultaneously, the acoustic emission intelligent monitoring system is activated to collect and analyze the power spectral density of the acoustic emission signal in real time during the ball milling process. When the spectral energy fluctuation is determined to be less than the set threshold for a continuous period of time, a shutdown command is automatically triggered. S3: The powder after ball milling in step S2 is placed in an argon-oxygen mixed atmosphere. The grain boundaries and dislocations introduced by ball milling are used as fast diffusion channels for internal oxidation treatment, so that oxygen atoms preferentially penetrate along the defect channels and react in situ with Y atoms segregated at the grain boundaries to generate Y2O3 nanoparticles. S4: Based on finite element simulation, a preset and optimized multi-parameter process window is used. During the SPS sintering process, the temperature field uniformity is monitored in real time using an infrared thermal imager, and the plasma emission spectrum is monitored using a fiber optic spectrometer. When O is detected... + with Ar + When the spectral line intensity ratio drops to the target ratio, it is determined that the oxide film on the powder surface has been effectively removed and axial pressure is immediately applied for densification sintering. S5: Perform microstructure characterization and mechanical property testing on the green body sintered in step S4, construct a digital twin, and feed back the measured Y2O3 nanoparticle data to the molecular dynamics model in step S1 and the finite element simulation model in step S4. Calibrate the model parameters, predict the multi-parameter process window for the next round of optimization, and perform virtual iteration and optimization of the multi-parameter process.

[0006] Furthermore, in step S1, the molecular dynamics model employs a coarse-grained meta-atomic strategy to construct the CoCrFeNiMn high-entropy alloy matrix, approximating the five elements as a single average atom with equal atomic ratios. The mass of each atom is calculated by weighted average of the atomic fractions of each component, and the lattice constant is determined by linear interpolation based on Vegard's law.

[0007] Furthermore, in step S2, the convolutional neural network of the acoustic emission intelligent monitoring system adopts a lightweight convolutional neural network based on time-spectrum graph recognition. The input layer receives the PSD time-spectrum graph, and then passes through the convolutional layer and the fully connected layer in sequence to output three state classifications: under-polished state, solid solution saturated state, and over-polished state.

[0008] Furthermore, in step S3, the oxygen concentration in the argon-oxygen mixed atmosphere is 4-6%.

[0009] Furthermore, O + with Ar + The spectral line intensity ratio determination uses a dual-condition latching mechanism, O + with Ar + The spectral line intensity is lower than the proportional threshold for a continuous period of time than the R value; and the rate of change of the R value during this continuous period is less than the change threshold.

[0010] Furthermore, the axial pressure in step S4 needs to be triggered at O + with Ar + Once the spectral line intensity ratio and the temperature field uniformity index TUI monitored by the infrared thermal imager both meet the conditions, the PLC will increase the pressure from the pre-pressure to the target pressure within the target time.

[0011] Furthermore, in step S5, the atomic-scale information output by the molecular dynamics model simulation is mapped to the material constitutive model of the finite element model through statistical methods, and the macroscopic field information calculated by the finite element model is fed back as the boundary conditions of the molecular dynamics model. The parameters of the two models are simultaneously inverted using the Bayesian optimization framework.

[0012] Furthermore, the potential parameters of the molecular dynamics model were calibrated using the Markov chain Monte Carlo algorithm, and the viscosity and activation energy of the finite element model were inverted using the trust region method. The parameters of both models were then self-updated.

[0013] The present invention also proposes a high-entropy alloy powder prepared according to the above-described preparation process, wherein the high-entropy alloy powder has an average particle size of Y2O3 nanophase of 5-10 nm, a number density of ≥8×10²³ m⁻³, a total oxygen content of 0.3-0.35 wt%, a Hall flow rate of 28 s / 50g, and a tap density of 3.25 g / cm³.

[0014] Furthermore, the high-entropy alloy powder has a yield strength of 1200-1300 MPa and an elongation of ≥12%.

[0015] Compared with the prior art, the present invention has the following significant advantages: By using mechanochemical dissociation and defect-engineered internal oxidation, uniform dispersion of Y2O3 nanoparticles at grain boundaries is achieved. The prepared CoCrFeNiMn-Y2O3ODS high-entropy alloy has a yield strength of 1200-1300 MPa and an elongation of ≥12%, which is 20-30% higher than that of traditional processes.

[0016] By introducing intelligent monitoring and plasma spectroscopy analysis, online determination of ball milling endpoints and sintering states is achieved, eliminating reliance on experience and ensuring good process repeatability. Through multi-batch repeatability verification, the key performance index variation is <2%, meeting the requirements for industrial production. Virtual iterative optimization based on digital twins can reduce trial-and-error experiments by 30-50%, significantly reducing R&D costs and time, and can be extended to other ODS alloy systems and other oxide dispersion phases. Attached Figure Description

[0017] Figure 1 This is a flowchart of the high-entropy alloy powder preparation process in this embodiment; Figure 2 A schematic diagram of a molecular dynamics (MD) simulation. Figure 3 This is a schematic diagram of an SPS device. Figure 4 SEM image of the high-entropy alloy powder prepared in Example 3; Figure 5 The X-ray diffraction pattern of the high-entropy alloy powder prepared in Example 3; Figure 6 The electron backscattering diffraction pattern of the high-entropy alloy powder prepared in Example 3; Among them, 1. Electrode; 2. Gasket; 3. Punch; 4. Mold; 5. Vacuum water cooling cavity; 6. Powder. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in further detail below with reference to the accompanying drawings and specific embodiments, but the implementation and protection scope of this invention are not limited thereto.

[0019] Example 1: As Figure 1 The diagram shown is a flow chart of the high-entropy alloy powder preparation process in this embodiment, which includes the following steps: Step S1: Based on molecular dynamics model simulation, the critical ball milling impact energy required to achieve the mechanochemical dissociation of yttrium oxide (Y2O3) raw material powder and the target addition amount of yttrium oxide (Y2O3) raw material powder are determined in advance, and the raw material powder is subjected to step-by-step vacuum drying treatment.

[0020] Utilize Figure 2 The molecular dynamics (MD) model, as shown, guides the setting of process parameters. First, a coarse-grained MD model of CoCrFeNiMn matrix and Y2O3 is constructed to simulate the dissociation behavior of Y2O3 under different impact energies of 10-25 kJ / g. Simulation results show that when the impact energy is ≥15 kJ / g, the YO bond breaking rate of Y2O3 reaches over 85%, and Y atoms can be efficiently dissolved in the matrix lattice. Based on this, the actual ball milling impact energy is set to 15-20 kJ / g.

[0021] The preferred raw material ratio is equiatomic CoCrFeNiMn, with the addition of 0.2-0.5 at% Y2O3. Step-by-step vacuum drying employs programmed temperature control: heating at 5°C / min to 50°C and holding for 2 hours, then heating at 3°C / min to 80°C and holding for 2 hours, and finally heating at 2°C / min to 120°C and holding for 2 hours, maintaining a vacuum level below 10⁻² Pa throughout the process. The oxygen content of the dried raw material is controlled at 0.05-0.10 wt%.

[0022] Specifically, the molecular dynamics model (MD) employs a coarse-grained atomic strategy to construct the CoCrFeNiMn high-entropy alloy matrix, approximating the five elements as a single average atom with equal atomic ratios. The atomic mass is calculated by weighted average of the atomic fractions of each component, and the lattice constant is determined by linear interpolation based on Vegard's law. In the specific implementation, the atomic mass is calculated to be 55.32 g / mol, and the lattice constant is determined to be 3.585 Å.

[0023] The simulation box size was set based on the ratio of Y₂O₃ addition to the number of matrix atoms, maintaining a ratio of Y₂O₃ molecules to matrix atoms of 1:750~1:800 to ensure a large statistical sample size while controlling computational costs. In practice, the simulation box size was set to 15 nm × 15 nm × 15 nm, containing 150,000 elementary atoms and 200 Y₂O₃ molecules, totaling 158,000 atoms. The impact energy loading employed a non-equilibrium momentum mirror method, applying velocity gradients in the x, y, and z axes to simulate the anisotropic characteristics of ball milling collisions. Specifically, the time step during the impact energy loading phase was 0.1 femtoseconds to ensure high-energy collision accuracy, while the relaxation phase was extended to 1 femtosecond to improve computational efficiency. YO bond breakage rate statistics were based on coordination number changes: dissociation was defined as the number of oxygen atoms within a 3 Å radius around a Y atom decreasing from an initial 3 to less than 1. The average breakage rate of all Y₂O₃ molecules was calculated over a 500 picosecond simulation period, with a sample size covering 200 independent molecules. Error analysis employed a block averaging method, dividing the trajectory into 10 sub-blocks. The standard deviation of the fracture rate was ±3.2%, and the confidence level for an 85% dissociation rate reached 95%. The determination of the actual process window of 15-20 kJ / g was based on triple verification: MD prediction showed that the dissociation rate exceeded the 85% threshold at 15 kJ / g and reached saturation at 18 kJ / g; experimental verification showed that no Y2O3 diffraction peaks were detected by XRD after ball milling at 18 kJ / g; energy efficiency analysis indicated that exceeding 20 kJ / g would lead to accelerated powder cold welding and an exponential increase in equipment wear. Therefore, 15-20 kJ / g was selected as the optimal high-efficiency and energy-saving range.

[0024] In a preferred embodiment, to address potential biases caused by matrix simplification in molecular dynamics simulations, the following model optimization strategy can be adopted: While maintaining the coarse-grained meta-atomic strategy to balance computational efficiency, a more refined multi-element embedded atomic potential is used to replace the single average atomic approximation, distinguishing the specific interactions between Co, Cr, Fe, Ni, Mn components and Y and O atoms. Simultaneously, the digital twin feedback mechanism already constructed in the process is fully utilized, and the potential parameters of the molecular dynamics model are systematically calibrated and inverted using experimentally measured data such as Y₂O₃ size and number density through the MCMC algorithm. This combination of model refinement and data-driven calibration can significantly improve the model's prediction accuracy for key behaviors such as the diffusion and binding of Y and O atoms in different elemental environments, while maintaining controllable computational costs.

[0025] Step S2: Place the powder processed in step S1 into a high-energy ball mill for ball milling. Simultaneously, activate the acoustic emission intelligent monitoring system to collect and analyze the power spectral density of the acoustic emission signal in real time during the ball milling process. When the spectral energy fluctuation is determined to be less than the set threshold for a continuous period of time, an automatic shutdown command is triggered.

[0026] The preferred ball milling equipment is a planetary high-energy ball mill. The grinding balls are made of WC-Co cemented carbide, with a preferred diameter ratio of Φ10mm : Φ6mm : Φ3mm of 2:3:5. Oleic acid, a process control agent, is added at the beginning of ball milling to prevent cold welding and to form a thin organic coating on the powder surface, reducing oxygen adsorption.

[0027] The intelligent acoustic emission monitoring system consists of a piezoelectric sensor, a preamplifier, a data acquisition card, and an AI module. The sensor is mounted on the outer wall of the ball mill jar, with a sampling frequency set to 2 MHz. The AI ​​module uses a convolutional neural network (CNN) to perform real-time identification of the PSD spectrum. When the standard deviation of the spectral energy is less than 5% (the upper limit threshold) for 30 consecutive minutes, the ball milling endpoint is determined. This intelligent determination method can shorten the ball milling time by 20-40% compared to the traditional fixed-time method and avoid powder contamination caused by over-milling.

[0028] In a preferred embodiment, the convolutional neural network (CNN) adopts a lightweight convolutional neural network (AE-CNN) based on time-spectrum graph recognition, and its specific structural parameters are as follows: Input layer: Receives a PSD with a size of 256×256 pixels, frequency range 0-500 kHz, time window 30 seconds, overlap rate 50%.

[0029] Convolutional layer C1: 32 3×3 convolutional kernels, ReLU activation function, batch normalization, and 2×2 max pooling.

[0030] Convolutional layer C2: 64 3×3 convolutional kernels, ReLU activation function, batch normalization, and 2×2 max pooling.

[0031] Convolutional layer C3: 128 3×3 convolutional kernels, ReLU activation function, batch normalization, global average pooling.

[0032] Fully connected layer FC1: 128 neurons, Dropout rate 0.3, ReLU activation function.

[0033] Output layer: 3 neurons, Softmax activation function, corresponding to three states: under-grinding state 0, solid solution saturation state 1, and over-grinding state 2.

[0034] The model has approximately 1.2 million parameters and an inference time of <50ms / frame, meeting real-time monitoring requirements. The model weight file is stored in an embedded GPU module for edge inference.

[0035] Acoustic emission signals were collected throughout the entire process of the CoCrFeNiMn-Y2O3 system under five different ball milling impact energies of 10, 15, 18, 20, and 25 kJ / g. Each group was repeated three times to obtain 15 complete ball milling time series data, with a total duration of 180 hours.

[0036] The data was labeled using the offline gold standard method. Ball milling sampling was paused every 10 minutes, and the residual rate of Y2O3 particles was observed by TEM. When the residual rate of Y2O3 was <5%, the corresponding time period was marked as solid solution saturated state 1. When the residual rate was >20%, it was marked as under-milled state 0. When the grain size was <30nm and the oxygen content increased sharply, it was marked as over-milled state 2.

[0037] A total of 54,000 PSD time spectrum images were generated and divided into training, validation and test sets in a 7:2:1 ratio to ensure a balanced distribution of samples in each category.

[0038] In a preferred embodiment, the original acoustic emission signal is bandpass filtered from 50 to 500 kHz to remove mechanical vibration noise; the time-frequency conversion uses the Welch method to calculate the PSD, with a Hamming window function, 4096 FFT points, and a frequency resolution of 122 Hz. The spectral energy fluctuation rate ΔE = (σ / μ) × 100%, where σ is the standard deviation of the total PSD energy within a 30-minute window, and μ is the mean.

[0039] The characteristic frequency band energy ratio R = E(15-25 kHz) / E(50-200 kHz) reflects the relative intensity of particle fracture and plastic deformation.

[0040] When the probability of the CNN outputting state 1 for 5 consecutive frames is greater than 0.95 and ΔE < 5% for 30 minutes, a shutdown command is triggered.

[0041] It should be noted that the 5% threshold corresponds to the critical point where energy input and powder deformation energy storage reach dynamic equilibrium during ball milling. ΔE < 5% indicates that the frequency of particle fracture, cold welding, and re-agglomeration events tends to stabilize. Based on the training dataset, the 95% confidence interval of ΔE in the solid solution saturated state is [2.8%, 5.2%], therefore, 5% is set as the upper limit threshold.

[0042] It should be noted that the convolutional neural network model of the acoustic emission intelligent monitoring system was trained and validated using a pre-established gold standard dataset through offline sampling and TEM observation during the training phase. Acoustic emission signals were collected throughout the entire process of three typical ODS alloy systems—CoCrFeNiMn-Y2O3, FeCrAl-Y2O3, and Ni-based superalloy-Y2O3—on a planetary high-energy ball mill, covering three stages: under-grinding, solution saturation, and over-grinding. Each system was repeated five times independently. In each experiment, three piezoelectric sensors were evenly distributed on the outer wall of the mill jar to continuously collect time-domain signals and simultaneously record milling time, powder particle size, XRD phase composition, and TEM microstructure as gold standard labels, obtaining a total of no less than 500 hours of raw acoustic emission time-series data. The training set used CoCrFeNiMn-Y2O3 data, accounting for 70%, while the validation and test sets each independently extracted 15% of FeCrAl-Y2O3 and Ni-based alloy data. The model achieved a macroscopic accuracy of >94% on the test set, a macro-average F1 score of >0.92, and a prediction error of <8% for unseen alloy systems.

[0043] In actual production applications, the trained acoustic emission intelligent monitoring system can directly collect and analyze acoustic emission signals during the ball milling process in real time. The trained AI model can infer the ball milling status online and automatically trigger a shutdown command. The entire practical application process no longer requires and does not rely on real-time offline sampling and analysis.

[0044] Step S3: The powder after ball milling in step S2 is placed in an argon-oxygen mixed atmosphere. The grain boundaries and dislocations introduced by ball milling are used as fast diffusion channels for internal oxidation treatment, so that oxygen atoms preferentially penetrate along the defect channels and react in situ with Y atoms segregated at the grain boundaries to generate Y2O3 nanoparticles.

[0045] The internal oxidation furnace uses a tubular furnace for precise atmosphere control. Controlling the oxygen partial pressure is crucial; too low a pressure results in insufficient oxidation, while too high a pressure leads to over-oxidation of the matrix. Ar-5% O2 is the preferred atmosphere, with the gas flow rate controlled at 50-100 mL / min. The internal oxidation temperature of 450°C is chosen because this temperature is higher than the diffusion activation energy threshold of Y, but lower than the matrix recrystallization temperature, thus preserving the defect structure introduced by ball milling and promoting in-situ nucleation of Y2O3 along grain boundaries.

[0046] After processing, the powder is sieved with electrostatic field assistance. The fine powder is effectively separated from the coarse powder by a high voltage electrostatic field of 5-10 kV. The sieve efficiency is 15-20% higher than that of traditional mechanical sieve, and the yield can reach more than 98%.

[0047] Controlling the oxygen partial pressure in the internal oxidation atmosphere requires a precise balance between the oxidation of Y and the risk of excessive matrix oxidation. Experimental data shows that at an oxygen partial pressure of 2%, the oxide layer growth rate is only 0.02 μm / h, the average Y₂O₃ particle size is 4.3 nm, but the number density is low at 2.1 × 10²³ m⁻³; at 5% oxygen partial pressure, the oxidation rate is 0.08 μm / h, the particle size increases to 7.8 nm, and the number density reaches a peak of 8.5 × 10²³ m⁻³; at an oxygen partial pressure of 8%, the matrix oxidation rate jumps to 0.25 μm / h, the Y₂O₃ size abnormally coarsens to over 15 nm, and the number density drops back to 5.2 × 10²³ m⁻³. The 3-10% process window is defined based on the particle size distribution pattern: in the 4-6% oxygen partial pressure range, a narrow distribution of 5-10 nm particles can be obtained, with a span < 1.5 and a uniformity index U > 0.90; above 7%, the proportion of particle agglomerates exceeds 15%, and the distribution becomes severely broadened. The critical oxygen partial pressure is determined using a thermodynamic-kinetic coupling standard, where the oxygen partial pressure leads to an oxidation free energy ΔG of the matrix Cr / Mn elements. ox An oxidation level <-120 kJ / mol and an oxide layer thickness >50 nm is considered excessive. Calculations show that ΔG under an Ar-6.5%O2 atmosphere... Cr-ox =-118 kJ / mol, oxide layer thickness 45 nm, which is the upper limit of safety. Therefore, an O2 concentration of 4-6% is preferred, which ensures the efficiency of in-situ Y2O3 generation while avoiding matrix oxidation and degradation, and the relative standard deviation of process repeatability is <3%.

[0048] Step S4: Based on the finite element simulation model, a preset optimized multi-parameter process window is used. During the SPS sintering process, the temperature field uniformity is monitored in real time using an infrared thermal imager, and the plasma emission spectrum is monitored using a fiber optic spectrometer. When O is detected... + with Ar + When the spectral line intensity ratio drops to the target ratio, it is determined that the oxide film on the powder surface has been effectively removed and axial pressure is immediately applied for densification sintering.

[0049] SPS device schematic diagram as follows Figure 3 As shown, the SPS equipment is equipped with a multi-parameter collaborative monitoring system. It utilizes an infrared thermal imager to monitor the temperature distribution on the mold surface in real time, ensuring a temperature difference of <30°C and preventing uneven density caused by temperature gradients.

[0050] In a specific embodiment, the spectrometer probe is installed through a 45° angled insertion hole on the side wall of the SPS mold. The radial distance between the probe tip and the powder compact is 15±2 mm, ensuring that the acquisition area covers more than 70% of the sample diameter. The axial position is located at the geometric mid-section of the graphite mold height, 20 mm from both the upper and lower indenters, to avoid interference from the axial temperature gradient on the spectral signal. The probe axis forms a 45° angle with the radial direction of the mold to prevent indenter obstruction and plasma jet interference during vertical observation. Preferably, an M12×1.5 threaded sealing interface is used, and the probe is double-sealed with the mold wall by a PTFE sealing ring and a graphite gasket to ensure a vacuum degree <10 Pa.

[0051] During the SPS sintering process, a fiber optic spectrometer monitors the plasma through a quartz window, detecting O at 441.5 nm. + Spectral lines and Ar at 442.6 nm + The line intensity ratio is a direct indicator of the surface oxide film removal process. Preferably, the line intensity ratio determination employs a dual-condition latching mechanism. + with Ar + The spectral intensity ratio R is continuously below the proportional threshold for a given time period; and the rate of change of R during this continuous time period is less than the change threshold. Specifically, Voigt fitting is performed every second in the 440-444nm band to separate the net peak areas of O⁺ and Ar⁺, the intensity ratio R is calculated, and an exponentially weighted moving average smoothing is applied with a weighting factor of 0.1. When the R value is below 0.015 for 30 consecutive seconds and the absolute value of its time derivative is less than 0.001, the system determines that the surface oxygen atom coverage is <5%, and the chemical potential has dropped below the decomposition threshold. This decomposition threshold corresponds to the inflection point of Y₂O₃ reduction kinetics. Applying axial pressure at this point can prevent secondary oxidation during the densification window.

[0052] An infrared thermal imager monitors the temperature field uniformity index (TUI) of the mold surface at 30Hz. During the holding period at 1050°C, TUI < 5% and R ≤ 0.015 must be simultaneously satisfied. The decision signal is sent from the edge computing node to the PLC within a 4ms cycle via the Profinet IRT protocol. After receiving the dual-condition ready flag, the PLC increases the pressure from the pre-compression 10MPa to 30MPa within 500ms using an S-curve at 40MPa / s, with a total response delay of <100ms. This moment precisely corresponds to the clean state of the powder surface and the peak of sintering activity, and the pressure is instantly transferred to the interparticle space, achieving rapid densification. After holding for 8 minutes, the system cools to 600°C at 50°C / min to depressurize. Plasma monitoring continues throughout the process until the temperature is below 800°C to ensure no risk of oxygen reversion. A hard time limit protection mechanism forces sintering to be completed within 12 minutes to prevent overheating.

[0053] Step S5: Characterize the microstructure and test the mechanical properties of the green body after sintering in Step S4, construct a digital twin, and feed back the measured Y2O3 nanoparticle data to the molecular dynamics model in Step S1 and the finite element simulation model in Step S4. Calibrate the model parameters, predict the multi-parameter process window for the next round of optimization, and perform virtual iteration and optimization of the multi-parameter process.

[0054] Microstructure characterization was performed using high-resolution transmission electron microscopy (HRTEM) combined with energy-dispersive X-ray spectroscopy (EDS) to quantitatively analyze the size distribution and number density of Y₂O₃ particles. Mechanical property testing was conducted according to ASTM E8 standard, including room temperature tensile tests.

[0055] A digital twin is constructed, comprising, from bottom to top, a data perception layer, a model computation layer, a collaborative optimization layer, and a decision application layer. The data perception layer communicates with the experimental equipment PLC via a RESTful API interface to acquire real-time measured Y2O3 data, with a data sampling interval of ≤5 seconds, and transmits the data in JSON format. The model computation layer integrates the molecular dynamics model (MD) and the finite element simulation model (FEM). The two models achieve memory-level data exchange through Python middleware, avoiding file I / O latency. The collaborative optimization layer deploys a Bayesian calibration framework and a multi-objective genetic algorithm, responsible for model parameter inversion and process window prediction. The decision application layer generates visualized process parameter cloud maps and provides a virtual experimental console, supporting online adjustment of 12 key process variables such as ball milling impact energy, oxygen partial pressure, and sintering temperature, with a response latency of <100 milliseconds.

[0056] The workflow of a digital twin is as follows: The experimentally measured Y2O3 size d, number density ρ, and grain size D are input into the molecular dynamics model MD and the finite element simulation model FEM.

[0057] The molecular dynamics model (MD) employs a coarse-grained modeling approach based on elementary atoms, treating the CoCrFeNiMn matrix as a single average atom. Matrix-matrix interactions utilize Finnis-Sinclair type embedded atom potentials, whose parameters are fitted using first-principles calculations of single-vacancy formation energy, stacking fault energy, and elastic constants. ReaxFF reaction fields are used within YO bonds and YO clusters to capture bond breaking and recombination. Y-matrix interatomic interactions are represented by a hybrid form of Ziegler-Biersack-Littmark (ZBL) shielded Coulomb and Lennard-Jones potentials. Short-range repulsion terms are described by ZBL, while long-range attraction terms are characterized by LJ potentials. LJ parameters... and The simulation was corrected by calculating using the Lorentz-Berthelot mixing rule and combining it with experimentally measured solid solubility data of Y in the matrix. The simulation employed a non-equilibrium momentum mirror method to apply impact energy, combined with a giant canonical ensemble to simulate the internal diffusion of oxygen along grain boundaries, outputting dislocation density and Y₂O₃ cluster distribution every 10 picoseconds.

[0058] The finite element model (FEM) is based on DEM-FEM coupling, discretizing the powder into rigid clusters and embedding SPH nodes. The Drucker-Prager Cap constitutive model is used to describe the compressive behavior, and the internal variables of back stress within grain boundary elements characterize the dispersion strengthening effect of Y₂O₃. Density evolution during sintering is calculated using the Skorhod-Olevsky viscous flow equation, and the mold thermal boundary employs a combination of convective heat transfer and Joule heating.

[0059] Model calibration employed a Bayesian optimization algorithm to adjust the interaction potential function parameters of YO and Y-Fe, ensuring the simulation prediction error was less than 5%. During the virtual experiment, the combined effects of different Y2O3 addition amounts, ball milling energy, and sintering temperatures were simulated in a digital twin to predict the optimal process window. Based on the virtual optimization results, a new round of experiments was conducted to verify the prediction accuracy and establish feedback optimization.

[0060] Specifically, the Bayesian optimization framework uses a multi-objective weighted function as its core, and the Gaussian process surrogate model employs a Matern 5 / 2 covariance kernel. Three sets of candidate parameters are recommended in parallel using a q-EI acquisition function. Molecular dynamics (MD) model pre-simulation rapidly screens for Y₂O₃ dissociation rates >80%, and FEM further predicts combinations with relative densities >98% for inclusion in physical experiments. After each experiment, the Y₂O₃ distribution and density curves are fed back to the cloud via a RESTful API. The MCMC algorithm is used to calibrate the MD potential parameters, and the trust region method is used to invert the FEM viscosity and activation energy, enabling model self-updating.

[0061] Cross-scale mapping averages the dislocation density output of the molecular dynamics model (MD) to the FEM model, while the high-strain region serves as the initial defect seed for the MD model, forming a feedback loop. Uncertainties are quantified using a heteroscedastic Gaussian process, and the robust optimal solution is determined by a 95% confidence lower bound. Computational acceleration strategies include GPU parallelization and POD order reduction, reducing a single iteration from 8 hours to 20 minutes.

[0062] It should be noted that in the iterative optimization process of the digital twin, a clear cross-scale data interface is established. Atomic-scale information such as the average dislocation density and Y atom segregation concentration output from the molecular dynamics model simulation is mapped to the material constitutive equations of the finite element model as initial conditions or internal variable parameters through volume averaging or statistical averaging methods. Conversely, information on local high-strain regions or temperature gradients calculated by the finite element model is fed back to the molecular dynamics model as construction conditions or boundary conditions for its simulation box, simulating the microscopic evolution of key regions. Parameter calibration is performed uniformly through a Bayesian optimization framework, using experimental data to simultaneously invert the parameters of the two models, ensuring consistency in multi-scale predictions.

[0063] Example 2: Preparation of CoCrFeNiMn-0.3 at% Y2O3 ODS high-entropy alloy powder and bulk.

[0064] Based on molecular dynamics (MD) model simulations, a key process threshold was set for the in-situ formation of Y₂O₃ nanoparticles in a CoCrFeNiMn matrix: the ball milling impact energy needs to reach 18 kJ / g to ensure effective mechanochemical dissociation of Y₂O₃. Accordingly, equiatomic proportions of high-purity (≥99.9%) Co, Cr, Fe, Ni, and Mn metal powders were accurately weighed, with an initial particle size d₀. 50 =45 μm. Add 0.3 at% nano Y2O3 powder with a particle size of 20 nm. After premixing the above powder in a V-type mixer for 2 hours, place it in a vacuum rotary dryer for step-by-step heating and drying, preferably at 50°C, 80°C, and 120°C, with each stage held for 2 hours, until the total oxygen content of the raw material is reduced to 0.08 wt%.

[0065] The premixed powder was loaded into a 500 mL cemented carbide ball mill jar, using WC-Co grinding balls with a ball-to-powder ratio of 12:1, totaling 720 g, and 1.5 wt% oleic acid was added as a process control agent. After sealing the ball mill jar, a vacuum of 5 Pa was applied, followed by the introduction of high-purity Ar gas with an oxygen partial pressure <30 ppm. The planetary ball mill was started, with the speed set to 350 rpm, and the acoustic emission intelligent monitoring system was activated simultaneously. After 12 hours of ball milling, the AI ​​system analyzed the power spectral density (PSD) of the acoustic emission signal, determining that the energy fluctuation in the 15-25 kHz frequency band had been less than 5% for 30 consecutive minutes, reaching a solid solution saturation state, and automatically triggered a shutdown command. Sampling and testing were performed to determine the powder particle size d. 50 =3.5 μm, X-ray diffraction (XRD) showed that the characteristic peaks of Y2O3 disappeared, indicating that Y had been dissolved into the matrix.

[0066] The ball-milled activated powder was evenly spread in an alumina crucible, placed in a tube furnace, and subjected to an Ar-5% O2 mixed atmosphere at a flow rate of 80 mL / min for internal oxidation treatment at 450°C for 2 hours. During this process, the ball milling introduced high-density grain boundaries, with a dislocation density of approximately 10¹. 4 The m⁻² region acts as a rapid diffusion channel, increasing the oxygen atom diffusion coefficient by 2-3 orders of magnitude. Oxygen preferentially penetrates to the grain boundaries and reacts in situ with the Y atoms segregated there, generating Y₂O₃ nanoparticles. After treatment, the powder is sieved using an electrostatic field-assisted method at a voltage of 5 kV, passing through a 300-mesh sieve with a pore size of 48 μm. The oversize content is 1.5%, and the yield reaches 98.5%.

[0067] The internally oxidized powder was loaded into a Φ20 mm graphite mold and pre-pressed to 10 MPa. Based on the optimized process window preset by the finite element method (FEM), the SPS sintering program was started. The AI ​​system monitored the temperature field on the mold surface using an infrared thermal imager and monitored the plasma emission spectrum in real time using a fiber optic spectrometer. When O was detected... + (441.5 nm) and Ar + When the intensity ratio of the spectral line at (442.6 nm) dropped to 0.015, the system determined that the oxide film on the powder surface had been effectively removed and immediately triggered the application of an axial pressure of 30 MPa. The sintering temperature was maintained at 1050°C for 8 minutes, followed by cooling to 600°C at a rate of 50°C / min to release the pressure, obtaining a sintered green body. The relative density of the green body was 99.2% as determined by the Archimedes method. Metallographic observation showed an average grain size of approximately 8.5 μm and a uniform, crack-free microstructure.

[0068] Systematic characterization of the sintered green body: Transmission electron microscopy (TEM) showed that the Y2O3 nanoparticles were 5-10 nm in size, uniformly distributed in the grain boundaries and dislocation network, with a number density of about 2.5 × 10²³ m⁻³, and no agglomeration.

[0069] Tensile specimens with a gauge length of Φ3 mm × 15 mm were processed according to ASTM E8 standard. The room temperature tensile test showed that the yield strength σ0.2 was 1250 MPa, the tensile strength was 1380 MPa, and the elongation was 15%, with excellent strength-plasticity matching.

[0070] The measured Y₂O₃ distribution density and size data were fed back into the MD model, and the YO interaction potential function parameters [E₀, σ, ε] were calibrated using a Bayesian optimization method, reducing the simulation prediction error from the initial 15% to 3%. Virtual experiments were conducted using the updated digital twin, and predictions showed that fine-tuning the Y₂O₃ addition to 0.28 at%, while simultaneously increasing the ball milling impact energy to 18 kJ / g, could further improve elongation while maintaining strength. A new round of experiments was conducted based on this prediction, and the prepared alloy achieved a yield strength of 1220 MPa and an elongation of 16.5%, with an error of approximately 2.4% compared to the digital twin prediction, verifying the predictive accuracy and virtual process optimization capabilities of the digital twin model.

[0071] Comparative Example 1: Using raw materials with the same composition as the present invention (0.3 at% Y2O3), but without MD simulation pre-design, AI acoustic emission monitoring, and internal oxidation process, a traditional process was adopted. Y2O3 nanoparticles were directly mixed and ball-milled for 24 hours, followed by sintering under the same SPS conditions of 1050°C, 30 MPa, and 8 min. Microstructure analysis of the resulting green body showed significant agglomeration of Y2O3, with particle sizes ranging from 50-200 nm and extremely uneven distribution. Mechanical property tests showed a yield strength of only 950 MPa and an elongation of 8%, both significantly lower than the present invention. This comparison highlights the significant advantages of the present invention in microstructure control and performance.

[0072] The complete process of this invention was repeated three times, and the key performance results are shown in Table 1: Table 1

[0073] Data shows that the batch-to-batch critical performance error is less than 1.5%, and the relative standard deviation is less than 1%, which fully demonstrates that the process of this invention has extremely high repeatability and stability, meeting the stringent requirements of industrial production and engineering applications.

[0074] This embodiment provides a high-entropy alloy powder preparation process for ODS based on multi-scale simulation pre-design, intelligent process monitoring, and digital twin iterative optimization. Verification through this embodiment shows that this process successfully solves the core problems of inhomogeneous microstructure and unstable performance in traditional processes, achieving in-situ controllable generation and uniform distribution of the Y2O3 nanophase, and obtaining high-entropy bulk alloy materials with excellent strength-toughness matching and outstanding batch stability.

[0075] Example 3: This example describes in detail the specific characteristics, technical indicators, and batch consistency verification data of the CoCrFeNiMn-0.3at% Y2O3 ODS high entropy alloy powder product prepared using the preparation process of the present invention.

[0076] The high-entropy alloy powder in this embodiment was prepared strictly following the process route in Table 2 below: Table 2

[0077] The powder prepared by this process was systematically characterized, and its physicochemical properties are as follows: (1) Oxygen content and oxygen distribution The total oxygen content was 0.32 wt% as determined by inert gas melting-infrared spectroscopy, and the dissolved oxygen was 0.09 wt% as determined by pulse-heated inert gas melting-mass spectrometry; the Y2O3 bound oxygen was 0.21 wt%; and the surface adsorbed oxygen was 0.02 wt%.

[0078] Surface scanning was performed using secondary ion mass spectrometry (SIMS). The relative standard deviation (RSD) of the O⁻ signal intensity distribution was 6.2%, indicating that oxygen was highly uniformly distributed both inside and between powder particles.

[0079] (2) In-situ formation state of Y2O3 nanophase One hundred fields of view were randomly observed in the ball-milled powder using high-resolution transmission electron microscopy (HRTEM), and the statistical results are as follows: Y2O3 particle size: average diameter d p =7.2±0.8 nm, with a minimum size of 5 nm.

[0080] Particle morphology: Nearly spherical or polyhedral, with a clear coherent / semi-coherent interface with the matrix.

[0081] The number density ρ = 8.5 × 10²³ m⁻³, which is significantly higher than that of conventionally mechanically mixed ODS powder, whose number density is usually less than 10²² m⁻³.

[0082] Spatial distribution: Y2O3 particles preferentially distribute along the grain boundaries and dislocation lines inside the powder particles, with a distribution uniformity index U=0.92, confirming the high efficiency of oxidation within defect engineering.

[0083] (3) Characterization of defect structure The average dislocation density was determined to be 1.8 × 10¹ using the line intercept method of transmission electron microscopy. 4 m⁻², Figure 4 SEM morphology of high-entropy alloy powder; Figure 5 X-ray diffraction patterns of high-entropy alloy powders, such as Figure 6 The electron backscatter diffraction (EBSD) pattern shown indicates that the powder particles are dominated by small-angle grain boundaries, accounting for 67%, which is beneficial for grain boundary migration and densification during the sintering process.

[0084] (4) Liquidity and filling properties According to GB / T 1482 standard, 50 g of powder flowed through a funnel with a standard aperture of 2.5 mm, and the measured flow rate was 28 s / 50 g, indicating that the powder has good flowability and meets the requirements for compression molding.

[0085] The loose density is 1.85 g / cm³, which is 23.2% of the theoretical density; the tapped density is 3.25 g / cm³, which is 40.8% of the theoretical density, indicating excellent filling performance.

[0086] (5) Thermal behavior and sintering activity Under Ar atmosphere, at a heating rate of 10°C / min, the powder exhibits a significant recrystallization exothermic peak at 625°C, with a peak temperature of 685°C and an exothermic heat ΔH = 42 J / g, confirming that the high-energy ball-milled powder has extremely high sintering activity.

[0087] At 1050°C, the powder weight loss was only 0.08 wt%, mainly due to the decomposition of residual oleic acid, and the oxidative weight gain was <0.05 wt%, indicating that the powder has good thermal stability and is suitable for rapid SPS sintering.

[0088] (6) Process window verification The tolerance of the powder sintering process window was verified by designing orthogonal experiments: Sintering in the range of 1020-1080°C can achieve a relative density of >98.5%, with 1050°C being the optimal equilibrium point.

[0089] Within the range of 25-35 MPa, the density difference is <0.5%, confirming that the powder is not sensitive to pressure parameters and has strong process stability.

[0090] Comparative Example 2: To highlight the technical advantages of the powder product of the present invention, the following comparative data is set up in Table 3: Table 3

[0091] 5. Powder batch stability verification Under the same process conditions, three batches of powder (batch numbers: ODS-HEA-P240115 / #1, #2, #3) were prepared consecutively on different dates. The statistical results of key quality attributes are shown in Table 4. Table 4

[0092] The relative standard deviation (RSD) of all key indicators of the three batches of powder was <4%, which is far below the <5% threshold usually required for industrial production; the size and number density of the Y2O3 nanophase fluctuated very little between batches, confirming the precise controllability of the internal oxidation process; the dynamic indicators were stable, indicating that the powder surface state was consistent, which is conducive to the standardized operation of subsequent forming processes.

[0093] The powder of this invention was sealed in an argon-protected aluminum foil bag and stored at room temperature for 6 months. After the storage, the oxygen content changed from 0.32 wt% to 0.34 wt%, with an increase of <0.02 wt%; the flowability changed from 28 s / 50g to 30 s / 50g; and there was no significant change in the particle size and distribution of Y2O3, indicating that the powder has excellent antioxidant and storage stability.

[0094] The powder was used in a selective laser melting (SLM) experiment with a layer thickness of 30 μm, a laser power of 280 W, and a scanning speed of 1200 mm / s. The relative density of the formed part reached 99.5%, and no spheroidization or crack defects were observed. The Y2O3 nanophase remained stable during the rapid laser solidification process without coarsening or dissolution.

[0095] The mechanical properties are comparable to those of SPS sintered parts, with a yield strength of 1220 MPa and an elongation of 14%, proving that the powder is suitable for multi-mode advanced manufacturing.

[0096] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application.

Claims

1. A process for preparing high-entropy alloy powder, characterized in that, Includes the following steps: S1: Based on molecular dynamics model simulation, the critical ball milling impact energy and the target addition amount of yttrium oxide Y2O3 raw material powder required to achieve mechanochemical dissociation of yttrium oxide Y2O3 raw material powder are determined in advance, and the raw material powder is subjected to step vacuum drying treatment. S2: The powder processed in step S1 is placed in a high-energy ball mill for ball milling. Simultaneously, the acoustic emission intelligent monitoring system is activated to collect and analyze the power spectral density of the acoustic emission signal in real time during the ball milling process. When the spectral energy fluctuation is determined to be less than the set threshold for a continuous period of time, a shutdown command is automatically triggered. S3: The powder after ball milling in step S2 is placed in an argon-oxygen mixed atmosphere. The grain boundaries and dislocations introduced by ball milling are used as fast diffusion channels for internal oxidation treatment, so that oxygen atoms preferentially penetrate along the defect channels and react in situ with Y atoms segregated at the grain boundaries to generate Y2O3 nanoparticles. S4: Based on finite element simulation, a preset and optimized multi-parameter process window is used. During the SPS sintering process, the temperature field uniformity is monitored in real time using an infrared thermal imager, and the plasma emission spectrum is monitored using a fiber optic spectrometer. When O is detected... + with Ar + When the spectral line intensity ratio drops to the target ratio, it is determined that the oxide film on the powder surface has been effectively removed and axial pressure is immediately applied for densification sintering. S5: Perform microstructure characterization and mechanical property testing on the green body sintered in step S4, construct a digital twin, and feed back the measured Y2O3 nanoparticle data to the molecular dynamics model in step S1 and the finite element simulation model in step S4. Calibrate the model parameters, predict the multi-parameter process window for the next round of optimization, and perform virtual iteration and optimization of the multi-parameter process.

2. The preparation process according to claim 1, characterized in that, In step S1, the molecular dynamics model uses a coarse-grained atomic strategy to construct the CoCrFeNiMn high-entropy alloy matrix. The five elements are approximated as a single average atom with equal atomic ratios. The mass of each atom is calculated by weighted average of the atomic fractions of each component, and the lattice constant is determined by linear interpolation based on Vegard's law.

3. The preparation process according to claim 1, characterized in that, In step S2, the convolutional neural network of the acoustic emission intelligent monitoring system adopts a lightweight convolutional neural network based on time spectrum recognition. The input layer receives the PSD time spectrum and passes through the convolutional layer and the fully connected layer in sequence to output three state classifications: under-polished state, solid solution saturated state, and over-polished state.

4. The preparation process according to claim 1, characterized in that, In step S3, the oxygen concentration in the argon-oxygen mixed atmosphere is 4-6%.

5. The preparation process according to claim 1, characterized in that, O + with Ar + The spectral line intensity ratio determination uses a dual-condition latching mechanism, O + with Ar + The spectral line intensity is lower than the proportional threshold for a continuous period of time than the R value; and the rate of change of the R value during this continuous period is less than the change threshold.

6. The preparation process according to claim 5, characterized in that, The axial pressure in step S4 needs to be triggered at O + with Ar + Once the spectral line intensity ratio and the temperature field uniformity index TUI monitored by the infrared thermal imager both meet the conditions, the PLC will increase the pressure from the pre-pressure to the target pressure within the target time.

7. The preparation process according to claim 1, characterized in that, In step S5, the atomic-scale information output by the molecular dynamics model simulation is mapped to the material constitutive model of the finite element model using statistical methods, and the macroscopic field information calculated by the finite element model is fed back as the boundary conditions of the molecular dynamics model. The parameters of the two models are simultaneously inverted using a Bayesian optimization framework.

8. The preparation process according to claim 7, characterized in that, The potential parameters of the molecular dynamics model were calibrated using the Markov chain Monte Carlo algorithm, and the viscosity and activation energy of the finite element model were inverted using the trust region method. The parameters of both models were self-updated.

9. A high-entropy alloy powder prepared by the preparation process according to any one of claims 1-8, characterized in that, The high-entropy alloy powder has an average particle size of 5-10 nm for the Y2O3 nanophase, a number density ≥8×10²³ m⁻³, a total oxygen content of 0.3-0.35 wt%, a Hall flow rate of 28 s / 50g, and a tap density of 3.25 g / cm³.

10. The high-entropy alloy powder according to claim 9, characterized in that, The high-entropy alloy powder has a yield strength of 1200-1300 MPa and an elongation of ≥12%.

Images