Target density-oriented oscillating pressure sintering process reverse generation method and system

Abstract

The application discloses a target density-oriented oscillating pressure sintering process reverse generation method and system, belonging to the technical field of material preparation and artificial intelligence aided process design. The method acquires material information and target density value or target density range input by a user, inputs the same into a pre-trained reverse generation model, and outputs corresponding oscillating pressure process parameter combination, including temperature curve, pressure amplitude, pressure frequency and pressure waveform. After feasibility verification, the process parameter combination is formatted into a process program file executable by a sintering device. The reverse generation model can be constructed by using a reverse neural network with physical constraint regularization term or a Bayesian optimization with automatic correlation determination and constraint expected improvement. The system comprises a target setting module, a reverse generation module, a verification and formatting module and a feedback updating module. The application realizes intelligent reverse generation from target density to optimal process, effectively solves an underdetermined inverse problem, and is suitable for rapid design of oscillating pressure sintering processes of various materials such as ceramics, metals and cermet.

Description

Technical Field

[0001] This invention belongs to the field of materials preparation and artificial intelligence-aided process design technology, specifically relating to a method and system for automatically generating oscillating pressure sintering process parameters based on target density or target performance indicators using artificial intelligence models such as inverse neural networks or Bayesian optimization. This invention is applicable to the oscillating pressure sintering process design of various materials, including ceramic materials, metallic materials, cermet composites, and cemented carbides. Background Technology

[0002] Oscillatory pressure sintering is an advanced material densification technology that superimposes periodic dynamic pressure oscillations on traditional hot pressing sintering. By adjusting parameters such as pressure amplitude, oscillation frequency, pressure average, and waveform, this technology can significantly promote particle rearrangement, eliminate residual porosity, and inhibit abnormal grain growth, thereby obtaining high-density, high-performance sintered products. This technology has broad application prospects in the preparation of advanced ceramic materials such as silicon carbide (SiC), silicon nitride, and zirconium oxide, as well as metals and their composites such as titanium alloys, nickel-based superalloys, cermets, and cemented carbides.

[0003] However, the process design of oscillating pressure sintering faces the following common technical challenges: (1) The process parameters are highly dimensional and strongly coupled, and the traditional "trial and error method" is extremely inefficient.

[0004] Oscillating pressure sintering involves dozens of control variables, including temperature profiles (heating rate, holding temperature, holding time, cooling rate), oscillation pressure parameters (pressure amplitude, frequency, average value, waveform), atmosphere, and vacuum level. These variables have a highly nonlinear and strongly coupled complex mapping relationship with the final sintered density, a characteristic prevalent in both ceramic and metallic materials. Traditional process design commonly employs a trial-and-error method: setting a set of parameters based on experience, testing the density after sintering, and adjusting the parameters to repeat the experiment if the requirements are not met. A single sintering cycle can last for hours or even days, requiring dozens of rounds of experiments to complete a single round of parameter optimization, taking months and incurring high research and development costs.

[0005] (2) There is a “one-to-many” underdetermined inverse problem between density and process parameters, making reverse design particularly difficult.

[0006] In practical R&D, engineers often aim to deduce the optimal combination of sintering process parameters from a target density (e.g., 3.08 g / cm³ for SiC ceramics or 4.45 g / cm³ for titanium alloys). However, the same density value can be achieved by countless different combinations of process parameters (e.g., high pressure for short-time and low pressure for long-time can both achieve similar densities). This "one-to-many" underdetermined inverse problem makes traditional methods (including simple database matching, empirical formula interpolation, or single-parameter adjustment) inefficient and inaccurate, hindering the achievement of precise "design-as-you-go" manufacturing. This problem is also prominent in the oscillating pressure sintering of metallic materials. For example, the densification behavior of titanium alloys is highly sensitive to both temperature and pressure; different process paths can yield the same density but with significant differences in microstructure.

[0007] (3) Existing AI-assisted sintering methods can only achieve forward prediction or database matching, but cannot achieve reverse generation driven by target density.

[0008] In recent years, some studies have attempted to incorporate artificial intelligence (AI) technology into sintering process design. For example, Chinese patent document CN121494584A discloses an intelligent post-processing method for integrated debinding and sintering of photocurable ceramics. This method automatically generates an integrated debinding and sintering process program based on the input ceramic powder type and resin type using a built-in database. However, this approach has the following inherent shortcomings: The direction is positive, not negative: its input is "type of ceramic powder and type of resin", and the output is a pre-stored debinding and sintering curve. Essentially, it is a positive database matching ("material → process"). This solution cannot generate a customized process based on the user-specified target density, nor can it extrapolate to density values ​​not included in the database.

[0009] Incomplete output parameters: The output process program is mainly based on temperature curves and does not involve key parameters unique to oscillating pressure sintering, such as pressure amplitude, oscillation frequency, and pressure waveform. Therefore, it cannot be directly applied to the oscillating pressure sintering process.

[0010] Lacks target performance-driven capability: This scheme cannot solve the inverse optimization problem of "finding the optimal process for a given density", and it cannot perform the selection when the same density corresponds to multiple process schemes.

[0011] Another type of existing technology, such as Chinese patent document CN119335853A, uses models such as random forest and convolutional neural network to predict sintering process parameters, but its direction is a forward prediction of "process parameters → material level / performance", which is opposite to the reverse design direction required by this invention.

[0012] Furthermore, in the field of metallic material sintering, existing process design methods also rely on experience databases and manual adjustments, lacking systematic reverse engineering tools. For example, the development of oscillating pressure sintering processes for titanium alloys still mainly relies on trial and error, lacking intelligent mapping methods from target density to process parameters. Summary of the Invention

[0013] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a reverse generation method and system for oscillating pressure sintering process oriented towards target density that has high process design efficiency and can handle underdetermined inverse problems.

[0014] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A reverse generation method for oscillating pressure sintering process oriented towards target density includes the following steps: S1: Obtain the material information and target density value or target density range input by the user.

[0015] S2: Input the material information and target density value or target density range into a pre-trained reverse generation model. The reverse generation model is a machine learning model obtained by reverse training with historical sintering process parameters as input and the corresponding actual sintering density as label, or a proxy model constructed by Bayesian optimization.

[0016] S3: The reverse generation model outputs a combination of oscillating pressure process parameters corresponding to the material information and the target density value or target density range. The combination of process parameters includes at least a temperature curve, pressure amplitude, pressure frequency, and pressure waveform. The pressure waveform is selected from one of a sine wave, a triangular wave, or a square wave.

[0017] S4: Perform feasibility verification on the output process parameter combination. The feasibility verification includes: determining whether each parameter is within the equipment's capability range. If it exceeds the range, automatically adjust it to the boundary value and generate a prompt message. After the verification is passed, format the process parameter combination into a process program file that can be executed by the sintering equipment.

[0018] As a further improvement to the above technical solution: The inverse generation model is an inverse neural network, and its training process uses the following loss function: Where N is the batch size. For the predicted process parameter vector, For the actual process parameter vector, For physical constraint regularization terms, is the regularization coefficient.

[0019] The reverse generation model is constructed through Bayesian optimization, specifically including: using a Gaussian process as a surrogate model, wherein the Gaussian process uses automatic correlation to determine the kernel function and learns the feature length scale of each dimension of the process parameters to reduce the redundant dimensions of the parameter space.

[0020] The automatic correlation determination kernel function is the ARD Matérn kernel function, and its expression is: in, , Let be the length scale parameter of the d-th dimension feature. For signal variance, For smoothing parameters, For gamma function, This is a modified Bessel function of the second type.

[0021] The Bayesian optimization uses constrained expectation improvement as the acquisition function, and its expression is: In the formula, To improve the standard, Let x be the feasible probability that the parameter combination satisfies all constraints, and its expression is: In the formula, The standard normal cumulative distribution function is... and These are the j-th constraint functions. The Gaussian process predicts the mean and standard deviation, where J is the total number of constraints.

[0022] The constraints include at least one of the following types: Pressure physical feasibility constraints: ,and ,in, The average pressure. This refers to the pressure amplitude. The maximum pressure allowed by the equipment. The minimum pressure allowed by the equipment; Equipment limit constraints: , , ,in, The highest sintering temperature, Where f is the maximum allowable temperature of the equipment, and f is the oscillation frequency. The highest frequency allowed by the device. This represents the average maximum allowable pressure of the equipment. Cracking risk constraints: ,in The cracking probability predicted by a classifier trained on historical data. This is a preset threshold.

[0023] The inverse generation model outputs multiple Pareto optimal process schemes, each with a predicted density value, confidence interval, and energy consumption estimate, which users can choose as needed.

[0024] The equipment capability range includes: maximum pressure, minimum pressure, oscillation frequency range, maximum temperature, and maximum heating rate; the automatic adjustment to boundary values ​​includes: setting parameters exceeding the upper limit as the upper limit value, and setting parameters below the lower limit as the lower limit value.

[0025] The temperature curve includes the heating rate, holding temperature, holding time, and cooling rate.

[0026] It also includes a model update step: taking the process parameter combination and corresponding actual density that have been successfully verified by actual sintering as new samples, adding them to the training set, and periodically retraining or fine-tuning the reverse generation model.

[0027] The material information includes one or more of the following: material type, material composition, or material grade; the method is applicable to the reverse generation of ceramic materials, metallic materials, or metal-ceramic composite materials by oscillating pressure sintering process.

[0028] This invention also provides a reverse generation system for an oscillating pressure sintering process oriented towards a target density, comprising: The target setting module is used to receive material information and target density values ​​or target density ranges input by the user. The reverse generation module, deployed on a server or edge computing device, embeds the reverse generation model as described in claim 1, and is used to output the corresponding combination of oscillating pressure process parameters based on the material information and the target density value or target density range; The verification and formatting module is used to perform the feasibility verification described in step S4 of claim 1, and output the process program file that the sintering equipment can execute; The feedback update module is used to collect actual sintering results and trigger the update of the reverse generation model.

[0029] As a further improvement to the above system technical solution: The inverse generation model is either an inverse neural network or a Bayesian optimization model; when it is an inverse neural network, its training loss function adopts the aforementioned loss function; when it is a Bayesian optimization model, it adopts the aforementioned Bayesian optimization method.

[0030] The verification and formatting module is also used to: generate warning information and automatically output the adjusted boundary value scheme when the output process parameter combination exceeds the equipment's capability range.

[0031] The feedback update module is configured to add the process parameter combination and actual sintering density that the user has confirmed and successfully executed to the training dataset, and automatically retrain the reverse generation model when the number of new samples reaches a preset threshold.

[0032] Compared with the prior art, the advantages of the present invention are as follows: This invention presents a reverse generation method for oscillating pressure sintering processes oriented towards target density. By acquiring material information and target density values ​​or ranges, the target density is used as the core input for reverse design, changing the traditional forward matching mode guided by material type. This allows process design to directly address performance requirements. Material information and target density are input into a pre-trained reverse generation model. This model, using historical process parameters as input and actual density as labels for reverse training, learns the nonlinear mapping relationship between process parameters and density, thereby extrapolating and predicting arbitrary target density values. Simultaneously, by introducing physical constraint regularization terms or constraint expectations to improve the acquisition function, it effectively solves the underdetermined inverse problem of "one-to-many" relationships between density and process parameters, selecting process parameter combinations that conform to physical laws from numerous possible solutions. The process parameters output by the reverse generation model include at least temperature curves, pressure amplitude, pressure frequency, and pressure waveform. The pressure waveform is explicitly selected from sine waves, triangular waves, or square waves, ensuring that the output parameters fully cover all control variables required for oscillating pressure sintering and can be directly used for equipment execution. Finally, a feasibility check is performed to determine whether each parameter is within the equipment's capabilities. If it exceeds the limits, it is automatically adjusted to the boundary values ​​and a prompt message is generated. After the check passes, it is formatted into a process program file that can be executed by the sintering equipment, thereby ensuring that the output process scheme is not only theoretically optimal, but can also be run directly on the actual equipment, avoiding invalid output and experimental failures, and realizing end-to-end automatic mapping from target density to executable process.

[0033] The reverse generation system for oscillating pressure sintering process oriented towards target density of this invention receives user-input material information and target density values ​​or ranges through a target setting module, providing a standardized user interface that encapsulates complex process design knowledge into intuitive input parameters, lowering the barrier to entry. The reverse generation module is deployed on a server or edge computing device, embedding the reverse generation model described in claim 1, allowing users to flexibly choose large-scale cloud training or rapid edge inference according to actual needs, balancing computational accuracy and deployment costs. The verification and formatting module performs feasibility verification and outputs a process program file executable by the sintering equipment, converting the abstract parameters generated by the algorithm into formatted instructions recognizable by the specific equipment, realizing a complete automated link from user requirements to equipment execution, eliminating the error risk of manual transcription. The feedback update module collects actual sintering results and triggers updates to the reverse generation model, feeding back successful experiences or lessons learned from each production run to the model training process. With data accumulation, the model is periodically retrained or fine-tuned, continuously improving system performance and forming a closed-loop self-optimization mechanism, achieving an evolutionary capability of "becoming more accurate with use," and solving the technical defect of static systems being unable to absorb new knowledge. Detailed Implementation

[0034] The present invention will be described in further detail below, but the embodiments of the present invention are not limited thereto.

[0035] Example 1: This embodiment provides a reverse generation method for oscillating pressure sintering process oriented towards target density. It adopts a Bayesian optimization algorithm with automatic correlation determination and constraint expectation improvement, and supports a variety of materials such as ceramics, metals and cermets.

[0036] The steps are as follows: Step S1: Obtain input information.

[0037] Users input material information and target density values ​​through a human-machine interface. For example, the material information is "SiC ceramic" (material type), and the target density is 3.08 g / cm³. The material information can also be "Ti-6Al-4V titanium alloy" or "WC-Co cemented carbide," and the system will automatically identify the material category and retrieve the corresponding historical database.

[0038] Step S2: Input the reverse generation model.

[0039] The system inputs material information and target density into a pre-built Bayesian optimization surrogate model. This model takes historical sintering process parameters (maximum temperature, holding time, average pressure, pressure amplitude, oscillation frequency, and pressure waveform) as input and the corresponding actual sintering density as labels, and is trained by maximizing marginal likelihood.

[0040] The Gaussian process uses the ARD Matérn 5 / 2 kernel function, and its expression is as follows: , , in, Let be the length scale parameter of the d-th dimension feature. For signal variance, For smoothing parameters, For gamma function, This is a modified Bessel function of the second type. After training, the system automatically identifies key parameters that significantly affect density (such as mean pressure and maximum temperature), achieving dimensionality reduction in the parameter space.

[0041] Step S3: Output process parameter combination.

[0042] The model employs a constraint expectation-based improved acquisition function: , , In the formula, To improve the standard, Let x be the feasible probability that the parameter combination x satisfies all constraints. The standard normal cumulative distribution function is... and These are the j-th constraint functions. The Gaussian process predicts the mean and standard deviation, where J is the total number of constraints.

[0043] The constraints include: Pressure physical feasibility constraints: ,and ,in, The average pressure. This refers to the pressure amplitude. The maximum pressure allowed by the equipment. The minimum pressure allowed by the equipment; Equipment limit constraints: , , ,in, The highest sintering temperature, Where f is the maximum allowable temperature of the equipment, and f is the oscillation frequency. The highest frequency allowed by the device. This represents the average maximum allowable pressure of the equipment. Cracking risk constraints: ,in The cracking probability predicted by a classifier trained on historical data. This is a preset threshold.

[0044] After 15 iterations of optimization, the system outputs a set of Pareto optimal process schemes (with predicted density, confidence interval, and energy consumption estimates). The user selects one of these schemes, for example: Temperature profile: Heating rate 20℃ / min to 1748℃, holding temperature for 288min, then cooling with the furnace; Pressure amplitude: 1.2 t; Pressure frequency: 0.9 Hz; Pressure waveform: sine wave.

[0045] Step S4: Feasibility verification and formatted output.

[0046] System verification: Maximum pressure 7.5t ≤ 20t, minimum pressure 5.1t ≥ 0.5t, temperature 1748℃ ≤ 2250℃, frequency 0.9Hz ≤ 2Hz, cracking probability 0.08 ≤ 0.2. After verification, the parameter combination is formatted into an XML process program file, which can be directly imported into the sintering equipment PLC for execution. If the parameters exceed the boundaries, the system automatically adjusts to the boundary values ​​and generates a prompt message.

[0047] Model update: New samples with successful actual sintering (process parameters and measured density) are added to the training set. The Gaussian process model is automatically retrained every 50 new samples, and the kernel function length scale parameter is updated to achieve continuous optimization.

[0048] Example 2: This embodiment provides a reverse generation system for an oscillating pressure sintering process oriented towards a target density. The system corresponding to the above method embodiment includes the following modules: Target setting module: Touch screen interface, receives material information input by the user (select material type from drop-down menu or manually input composition) and target density value.

[0049] Reverse engineering module: Deployed on an edge computing industrial control computer (Intel i7, 32GB RAM), embedding the above-mentioned Bayesian optimization model, and outputting a recommended solution in about 0.5 seconds.

[0050] Verification and formatting module: Performs feasibility verification, outputs XML format process files, and generates warning messages (if parameters are out of bounds).

[0051] Feedback update module: The actual sintering density is automatically collected through the laboratory information system or manually entered by the operator; when the number of new samples reaches 50, the background model retraining is automatically triggered.

[0052] The system has a built-in materials database covering ceramics (SiC, ... , The system supports various materials including ceramics (SiC, metals (Ti-6Al-4V, Inconel 718, 304 stainless steel), and cermets (WC-Co, Ti(C,N)-based cermet). Users input the material type, composition (e.g., "SiC-5% carbon nanotubes"), or grade (e.g., "YG8"), and the system automatically loads the corresponding historical process database. For new grades, the system initializes model parameters from a database of similar materials through transfer learning. This system is applicable to ceramic materials (SiC, metals (Ti-6Al-4V, Inconel 718, 304 stainless steel), and cermets (WC-Co, Ti(C,N)-based cermets). , The reverse generation of materials such as metals (Ti-6Al-4V, Inconel 718) and metal-ceramic composites (WC-Co) through oscillating pressure sintering only requires switching the corresponding historical database.

[0053] Example 3: This embodiment provides an alternative implementation method to Embodiment 1, which is more suitable for scenarios with large-scale historical data and millisecond-level response.

[0054] Inverse Neural Network Model: A 4-layer fully connected neural network is constructed: an input layer with 1 node (target density), hidden layers with 128, 256, and 128 nodes respectively (ReLU activation, Dropout 0.2), and an output layer with 6 nodes (process parameters). The loss function is: Where N is the batch size. For the predicted process parameter vector, For the actual process parameter vector, For physical constraint regularization terms, This is the regularization coefficient. Physical constraint regularization term. During training, the correlation coefficient between the output pressure amplitude and the input density is calculated for each batch. If it is negative, a penalty is added, λ=0.01.

[0055] Training and Reasoning: The network was trained for 300 rounds using 500 sets of historical data (covering materials such as SiC, titanium alloys, and WC-Co). After the user inputs a target density of 3.08 g / cm³, the network outputs the following process parameters during forward propagation: 1746℃, 285 min, 6.25 t, 1.18 t, 0.88 Hz, and a sine wave. Inference time was <0.1 seconds.

[0056] Example 4: This embodiment, based on Embodiment 1, adds an energy consumption prediction model E(x) (linear regression, inputting holding time, maximum temperature, and average pressure). It employs the ParEGO multi-objective acquisition function, outputting a Pareto frontier set after 15 iterations. Each scheme includes prediction density, confidence interval (given by the posterior variance of the Gaussian process), and energy consumption prediction. After the user selects a scheme through the objective setting module, the system outputs the corresponding executable process file.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A reverse generation method for oscillating pressure sintering process oriented towards target density, characterized in that, Includes the following steps: S1: Obtain the material information and target density value or target density range input by the user; S2: Input the material information and target density value or target density range into a pre-trained reverse generation model. The reverse generation model is a machine learning model obtained by reverse training with historical sintering process parameters as input and the corresponding actual sintering density as label, or a proxy model constructed by Bayesian optimization. S3: The reverse generation model outputs a combination of oscillating pressure process parameters corresponding to the material information and the target density value or target density range. The combination of process parameters includes at least a temperature curve, pressure amplitude, pressure frequency, and pressure waveform. The pressure waveform is selected from one of a sine wave, a triangular wave, or a square wave. S4: Perform feasibility verification on the output process parameter combination. The feasibility verification includes: determining whether each parameter is within the equipment's capability range. If it exceeds the range, automatically adjust it to the boundary value and generate a prompt message. After the verification is passed, format the process parameter combination into a process program file that can be executed by the sintering equipment.

2. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The inverse generation model is an inverse neural network, and its training process uses the following loss function: Where N is the batch size. For the predicted process parameter vector, For the actual process parameter vector, For physical constraint regularization terms, is the regularization coefficient.

3. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The reverse generation model is constructed through Bayesian optimization, specifically including: using a Gaussian process as a surrogate model, wherein the Gaussian process uses automatic correlation to determine the kernel function and learns the feature length scale of each dimension of the process parameters to reduce the redundant dimensions of the parameter space.

4. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 3, characterized in that, The automatic correlation determination kernel function is the ARD Matérn kernel function, and its expression is: in, , Let be the length scale parameter of the d-th dimension feature. For signal variance, For smoothing parameters, For gamma function, This is a modified Bessel function of the second type.

5. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 3, characterized in that, The Bayesian optimization uses constrained expectation improvement as the acquisition function, and its expression is: middle, To improve the standard, Let x be the feasible probability that the parameter combination satisfies all constraints, and its expression is: In the formula, The standard normal cumulative distribution function is... and These are the j-th constraint functions. The Gaussian process predicts the mean and standard deviation, where J is the total number of constraints.

6. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 5, characterized in that, The constraints include at least one of the following types: Pressure physical feasibility constraints: ,and ,in, The average pressure. This refers to the pressure amplitude. The maximum pressure allowed by the equipment. The minimum pressure allowed by the equipment; Equipment limit constraints: , , ,in, The highest sintering temperature, Where f is the maximum allowable temperature of the equipment, and f is the oscillation frequency. The highest frequency allowed by the device. This represents the average maximum allowable pressure of the equipment. Cracking risk constraints: ,in The cracking probability predicted by a classifier trained on historical data. This is a preset threshold.

7. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The inverse generation model outputs multiple Pareto optimal process schemes, each with a predicted density value, confidence interval, and energy consumption estimate, which users can choose as needed.

8. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The equipment capability range includes: maximum pressure, minimum pressure, oscillation frequency range, maximum temperature, and maximum heating rate; the automatic adjustment to boundary values ​​includes: setting parameters exceeding the upper limit as the upper limit value, and setting parameters below the lower limit as the lower limit value.

9. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The temperature curve includes the heating rate, holding temperature, holding time, and cooling rate.

10. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, It also includes a model update step: taking the process parameter combination and corresponding actual density that have been successfully verified by actual sintering as new samples, adding them to the training set, and periodically retraining or fine-tuning the reverse generation model.

11. The reverse generation method of the oscillating pressure sintering process oriented towards target density according to claim 1, characterized in that, The material information includes one or more of the following: material type, material composition, or material grade; the method is applicable to the reverse generation of ceramic materials, metallic materials, or metal-ceramic composite materials by oscillating pressure sintering process.

12. A reverse generation system for an oscillating pressure sintering process oriented towards a target density, characterized in that, include: The target setting module is used to receive material information and target density values ​​or target density ranges input by the user. The reverse generation module, deployed on a server or edge computing device, embeds the reverse generation model as described in claim 1, and is used to output the corresponding combination of oscillating pressure process parameters based on the material information and the target density value or target density range; The verification and formatting module is used to perform the feasibility verification described in step S4 of claim 1, and output the process program file that the sintering equipment can execute; The feedback update module is used to collect actual sintering results and trigger the update of the reverse generation model.

13. The reverse generation system for oscillating pressure sintering process oriented towards target density according to claim 12, characterized in that, The inverse generation model is an inverse neural network or a Bayesian optimization model; when it is an inverse neural network, its training loss function adopts the loss function described in claim 2; when it is a Bayesian optimization model, it adopts the Bayesian optimization method described in any one of claims 3 to 6.

14. The reverse generation system for oscillating pressure sintering process oriented towards target density according to claim 12, characterized in that, The verification and formatting module is also used to: generate warning information and automatically output the adjusted boundary value scheme when the output process parameter combination exceeds the equipment's capability range.

15. The reverse generation system for oscillating pressure sintering process oriented towards target density according to claim 12, characterized in that, The feedback update module is configured to add the process parameter combination and actual sintering density that the user has confirmed and successfully executed to the training dataset, and automatically retrain the reverse generation model when the number of new samples reaches a preset threshold.

Images