Metal 3D printed part heat treatment process intelligent optimization system and method
By employing an intelligent optimization method based on defect topology tensor and sequence generation network, the problems of personalized customization and micro-defect repair in the traditional heat treatment process of metal 3D printed parts are solved, realizing a customized and reliable heat treatment solution and improving the performance and production efficiency of metal 3D printed parts.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG TIANXIONG IND TECH CO LTD
- Filing Date
- 2026-05-25
- Publication Date
- 2026-06-19
Smart Images
Figure CN122242312A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of metal additive manufacturing post-processing and intelligent control technology, specifically to an intelligent optimization system and method for the heat treatment process of metal 3D printed parts. Background Technology
[0002] With the increasing popularity of metal 3D printing technology, the intrinsic quality and performance reliability of its parts have become key challenges. During the forming process, due to complex physical and metallurgical behaviors such as melting and solidification, metal 3D printed parts inevitably produce inherent defects such as pores, microcracks and residual stress. These defects will significantly affect the mechanical properties and service life of the printed parts.
[0003] Currently, these defects are mainly repaired and optimized through heat treatment processes. However, traditional heat treatment processes rely on trial and error based on experience. Technicians usually need to manually set parameters such as temperature, time, and cooling rate based on previous experience and general specifications. This method is time-consuming and difficult to personalize for the unique initial defect state of each printed part. In addition, due to the lack of precise quantitative evaluation of the repair effect of micro-defects, the final heat treatment solution may have uncertainty in eliminating potential failure risks. Therefore, how to establish a heat treatment solution that can generate customized, efficient, and reliable performance for the unique defects of each metal 3D printed part has become a technical problem that urgently needs to be solved in this field. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention provides an intelligent optimization system and method for the heat treatment process of metal 3D printed parts. Specifically, the technical solution of this invention is as follows:
[0005] Intelligent optimization methods for heat treatment processes of metal 3D printed parts include:
[0006] Obtain the initial defect field of the printed part;
[0007] Based on the initial defect field, construct the defect topology tensor;
[0008] Based on the engineering design requirements, a target performance vector is set;
[0009] Based on the defect topology tensor, an initial process parameter sequence is generated through a sequence generation network;
[0010] By combining the initial process parameter sequence with the target performance vector, the optimal process parameter sequence is determined through a composite loss function of physical constraints.
[0011] Based on the optimal process parameter sequence, the defect passivation index is calculated using a pre-set defect passivation evaluation model.
[0012] If the defect passivation index is lower than the preset passivation threshold, a penalty term is added to the composite loss function, and the optimal process parameter sequence is re-determined based on the modified composite loss function.
[0013] If the defect passivation index is not lower than the preset passivation threshold, the optimal process parameter sequence is output as the final heat treatment process scheme.
[0014] Optionally, construct the defect topology tensor, including:
[0015] The porosity of the printed parts was obtained using industrial computed tomography (CT) scans.
[0016] The residual stress of the printed part was measured using X-ray diffraction technology;
[0017] Stress field reconstruction based on measured residual stress;
[0018] Porosity and reconstructed residual stress are standardized inputs to generate the defect topology tensor.
[0019] Optionally, the sequence generation network includes:
[0020] It adopts an architecture based on recurrent neural networks;
[0021] The initial hidden state of the network is initialized using the defect topology tensor.
[0022] Optionally, the composite loss function of physical constraints consists of the following parts:
[0023] The dimensionless performance loss is determined based on the difference between the performance prediction error corresponding to the optimal process parameter sequence and the target performance vector.
[0024] The dimensionless physical fit loss is determined based on the difference between the evolution rate of the normalized state vector and the prediction rate of the normalized physical model.
[0025] The dimensionless time cost loss is determined based on the ratio of the total process time to the preset reference time.
[0026] Optionally, the defect passivation index is calculated, including:
[0027] Based on the defect topology tensor, the initial stress concentration factor is calculated through finite element analysis;
[0028] In finite element analysis, thermal stress and microstructure transformation stress generated by the optimal process parameter sequence are applied, and the stress concentration factor after processing is calculated.
[0029] The defect passivation index is determined based on the initial stress concentration factor and the stress concentration factor after treatment.
[0030] Optional, add penalty items, including:
[0031] In the composite loss function, a penalty term is added, which is proportional to the stress concentration factor after treatment.
[0032] A smart optimization system for heat treatment process of metal 3D printed parts, comprising:
[0033] The defect characterization module is used to obtain the initial defect field of the printed part and construct the defect topology tensor based on the initial defect field.
[0034] The target setting module is used to set the target performance vector according to engineering design requirements;
[0035] The process path generation module is used to generate an initial process parameter sequence based on the defect topology tensor through a sequence generation network.
[0036] The process path optimization module is used to combine the initial process parameter sequence with the target performance vector and determine the optimal process parameter sequence through a composite loss function of physical constraints.
[0037] The closed-loop evaluation and correction module is used to calculate the defect passivation index based on the optimal process parameter sequence, and correct the composite loss function or output the final heat treatment process scheme based on the comparison result between the defect passivation index and the preset passivation threshold.
[0038] Optional, closed-loop evaluation and correction module, including:
[0039] The effect evaluation unit is used to calculate the defect passivation index based on the optimal process parameter sequence and through a preset defect passivation evaluation model.
[0040] The feedback control unit is used to add a penalty term to the composite loss function and instruct the process path optimization module to redetermine the optimal process parameter sequence in response to the defect passivation index being lower than the preset passivation threshold.
[0041] The scheme output unit is used to output the optimal process parameter sequence as the final heat treatment process scheme in response to the defect passivation index not being lower than the preset passivation threshold.
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] 1. This invention enables personalized customization of heat treatment processes, solving the problem that traditional methods struggle to account for individual defect differences. By accurately acquiring the unique initial defect information of each printed part and using it as the initial state of a deep learning model, this invention can generate highly targeted heat treatment solutions, achieving intelligent process design tailored to individual needs, thereby maximizing the repair of individual defects and improving the performance of printed parts.
[0044] 2. It ensures the comprehensive balance of the optimization results and overcomes the one-sidedness that may be caused by single-objective optimization. This invention constructs a multi-dimensional composite loss function that includes performance, physical laws and time costs. It can take into account the achievement of macroscopic performance, the authenticity of physical processes and production efficiency and energy consumption in the optimization process, and find a globally optimal solution under multiple engineering constraints, rather than just satisfying a certain performance index.
[0045] 3. A quantitative evaluation of the repair effect of micro-defects has been established, which makes up for the lack of direct evaluation methods in traditional processes. This invention innovatively introduces a defect passivation index based on the principle of fracture mechanics. The change of stress concentration coefficient at the defect tip before and after heat treatment is accurately calculated through finite element simulation. This provides an intuitive and reliable quantitative indicator for the repair effect of micro-defects, thereby enabling the evaluation and improvement of the fatigue resistance and service life of parts from the root.
[0046] 4. A closed-loop feedback adaptive optimization mechanism was constructed to ensure the reliability and effectiveness of the final solution. When the system evaluation finds that the current optimal solution is insufficient to passivate micro-defects, it will automatically correct the optimization target and guide the next iteration by adding a penalty term until both macro performance and micro-defect repair reach the preset standards. This intelligent closed-loop correction capability ensures that the final output process solution is fully verified, which greatly improves the performance reliability and production stability of the final product. Attached Figure Description
[0047] The present invention will be further explained below with reference to the accompanying drawings and embodiments:
[0048] Figure 1 This is a flowchart of the method of the present invention;
[0049] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation
[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0051] Example 1:
[0052] Please see Figure 1 Intelligent optimization methods for heat treatment processes of metal 3D printed parts include:
[0053] Obtain the initial defect field of the printed part;
[0054] Based on the initial defect field, construct the defect topology tensor;
[0055] Based on the engineering design requirements, a target performance vector is set;
[0056] Based on the defect topology tensor, an initial process parameter sequence is generated through a sequence generation network;
[0057] By combining the initial process parameter sequence with the target performance vector, the optimal process parameter sequence is determined through a composite loss function of physical constraints.
[0058] Based on the optimal process parameter sequence, the defect passivation index is calculated using a pre-set defect passivation evaluation model.
[0059] If the defect passivation index is lower than the preset passivation threshold, a penalty term is added to the composite loss function, and the optimal process parameter sequence is re-determined based on the modified composite loss function.
[0060] If the defect passivation index is not lower than the preset passivation threshold, the optimal process parameter sequence is output as the final heat treatment process scheme.
[0061] This embodiment describes the complete process of an intelligent optimization method for the heat treatment process of metal 3D printed parts, aiming to establish a data-driven closed-loop optimization system from initial defect characterization to final process solution output. This method ensures that the generated heat treatment process path not only meets the design requirements in terms of macroscopic performance, but also effectively repairs the inherent defects inside the printed parts in terms of microscopic performance, while taking into account the authenticity of physical laws and production efficiency.
[0062] The purpose of obtaining the initial defect field of the printed part is to comprehensively and accurately quantify the microscopic defect information that exists inside the metal 3D printed part in its initial state and has a decisive influence on the final mechanical properties. In this embodiment, this step is achieved through non-destructive testing technology. Industrial computed tomography (CT) technology is used to perform a global scan of the printed part, thereby obtaining the three-dimensional spatial distribution and porosity of its internal pore defects. Scalar field data were obtained; simultaneously, X-ray diffraction (XRD) technology was used to measure key areas of the printed part to acquire residual stress. Tensor data.
[0063] Based on the initial defect field, a defect topology tensor is constructed. The purpose is to mathematically unify and structure the multi-source, heterogeneous defect data obtained in the preceding steps, forming a standardized input format that can be directly processed by a machine learning model. This representation method establishes a direct mathematical bridge between defect information and process parameter optimization.
[0064] In this embodiment, the defect topology tensor Defined as a spatial location Combinatorial tensor at the location: The defective topological tensor here In spatial location The mathematical description of the defect state serves as the initial state input for the subsequent sequence generation network, ensuring that the generation of the process path considers the individualized defect characteristics of the printed part from the source; its components Derived from CT data, representing the porosity at that point; Derived from XRD measurements and subsequent stress field reconstruction algorithms, it represents the residual stress tensor at that point;
[0065] According to engineering design requirements, a target performance vector is defined to provide a clear and quantifiable final convergence target for the entire optimization process. This vector serves as the final benchmark for evaluating the merits of all candidate process paths. In this embodiment, the target performance vector... Based on product design specifications or relevant materials engineering standards, its composition is as follows: This target performance vector This refers to the set of macroscopic mechanical performance indicators that the heat treatment process needs to achieve. Its role in optimization algorithms is to act as the convergence target of the loss function, guiding the optimization direction. The target Vickers hardness value, Both parameters, representing the target fracture toughness value, are derived from specific engineering design requirements.
[0066] Based on the defect topology tensor, an initial process parameter sequence is generated through a sequence generation network. The core objective of this step is to utilize a deep learning model to establish a complex nonlinear mapping relationship between the initial defect state of the printed part and a set of potentially effective heat treatment process parameter sequences. In this embodiment, the sequence generation network adopts an architecture based on a recurrent neural network (RNN). The technical concept is that the initial hidden state of the network is generated from the defect topology tensor in the previous step. Initialization is performed; this design ensures that the process path generated by the network is tailored to the defects of the specific printed part; the network outputs an initial sequence of process parameters. ,in Represents the temperature corresponding to a specific metallurgical sub-stage. and heat preservation time .
[0067] By combining the initial process parameter sequence with the target performance vector, the optimal process parameter sequence is determined through a composite loss function based on physical constraints. The purpose of this step is to evaluate and iteratively optimize the initially generated process parameter sequence, ensuring that the final sequence not only approximates the target performance but is also physically feasible and takes time costs into account. To this end, this embodiment constructs a dimensionless composite loss function based on physical constraints that has undergone rigorous normalization. By minimizing this loss function using optimization algorithms such as gradient descent, the optimal sequence of process parameters can be found in a vast parameter space. ;
[0068] Based on the optimal process parameter sequence, the defect passivation index is calculated using a pre-defined defect passivation evaluation model. This step is designed to add an explicit evaluation step of the microscopic defect repair effect on top of macroscopic performance optimization, thus forming a more comprehensive evaluation system. In this embodiment, a defect passivation index based on fracture mechanics principles is introduced. As an evaluation metric, this index is calculated using finite element analysis software. It quantifies the effectiveness of the process path in reducing the risk of microcrack initiation by comparing the stress concentration factor at the defect tip before and after applying the optimal process path.
[0069] In response to the defect passivation index falling below the preset passivation threshold, a penalty term is added to the composite loss function, and the optimal process parameter sequence is re-determined based on the corrected composite loss function. This is a closed-loop feedback correction mechanism, the purpose of which is to automatically adjust the optimization target and guide the next round of optimization toward a direction more conducive to defect passivation when it is found that the current optimal path is insufficient to repair micro-defects.
[0070] The preset passivation threshold here This is an engineering parameter set according to the fatigue life safety level requirements of the part or relevant industry standards, such as the design specifications of an aero-engine turbine disk. Its value can be calibrated using a standard material fatigue performance database or SN curve model to ensure that the treated part meets the expected service life requirements; if Below The system will use the composite loss function Add a factor related to the stress concentration factor after treatment. Proportional penalty , here This is a positive weighting coefficient used to adjust the intensity of the penalty, thereby penalizing process paths that lead to high stress concentration in subsequent iterations.
[0071] If the defect passivation index is not lower than the preset passivation threshold, the optimal process parameter sequence is output as the final heat treatment process scheme; when all evaluation indicators meet the preset engineering requirements, the iterative optimization process terminates. The final, fully validated and optimized heat treatment process solution is then output.
[0072] This invention constructs an end-to-end intelligent optimization method from defect input to process output. Compared with traditional methods that rely on trial and error based on experience, this invention can generate customized, optimal, and physically reliable heat treatment schemes for the unique initial defect state of each printed part. By introducing a defect topology tensor, multi-source defect information is directly integrated into the initial state of the deep learning model, realizing personalized instruction in process generation. The innovative composite loss function and defect passivation closed-loop correction mechanism ensure that the optimization results achieve a high balance between macroscopic performance, microscopic repair, physical realism, and economic cost, greatly improving the final performance reliability and production efficiency of metal 3D printed parts.
[0073] Example 2:
[0074] Constructing the defect topology tensor includes:
[0075] The porosity of the printed parts was obtained using industrial computed tomography (CT) scans.
[0076] The residual stress of the printed part was measured using X-ray diffraction technology;
[0077] Stress field reconstruction based on measured residual stress;
[0078] The porosity and reconstructed residual stress are unified as standardized inputs to generate the defect topology tensor.
[0079] This embodiment is a detailed explanation of the technical feature of constructing the defect topology tensor in the method of embodiment 1, and aims to illustrate how to transform the raw defect data measured in the physical world into mathematical language that can be understood by machines.
[0080] Porosity of printed parts obtained through industrial computed tomography (CT): By using industrial CT to perform 3D scanning of the printed parts and analyzing the reconstructed volume data, the location, size, and shape of internal pores can be accurately identified, and the porosity of each spatial location can be calculated. porosity at This forms a scalar field covering the entire printed part;
[0081] Residual stress in printed parts was measured using X-ray diffraction (XRD): XRD equipment was used to perform non-destructive stress measurements on key areas of the printed part's surface and near-surface, obtaining the residual stress tensor at these discrete locations. .
[0082] Stress field reconstruction based on measured residual stress: To obtain the stress distribution across the entire domain, this embodiment employs a stress field reconstruction algorithm. This algorithm, based on a physical model and the finite element method, uses discrete measured stress data as boundary conditions or calibration points, and through interpolation and calculation, deduces all spatial locations within the printed part. Residual stress tensor at the location This creates a continuous stress field;
[0083] Unifying porosity and reconstructed residual stress as standardized inputs to generate the defect topology tensor: this step is central to data fusion; through each spatial coordinate point Above, the porosity at this point and residual stress tensor Combined into a higher-dimensional mathematical object, that is This generates the defect topology tensor.
[0084] This invention not only achieves a comprehensive characterization of two key defects: porosity and residual stress, but more importantly, through stress field reconstruction and tensor construction, it successfully integrates defect information with two different physical properties and data structures into a unified and standardized mathematical framework. This greatly improves the completeness and usability of defect information, laying a solid data foundation for the accurate modeling and effective training of subsequent deep learning models.
[0085] Example 3:
[0086] Sequence generation networks, including:
[0087] It adopts an architecture based on recurrent neural networks;
[0088] The initial hidden state of the network is initialized using the defect topology tensor.
[0089] This embodiment describes the specific technical implementation of the sequence generation network, the core computing tool, in the method described in Embodiment 1;
[0090] An architecture based on recurrent neural networks (RNN) is adopted: The core reason for choosing RNN or its variants as the basic architecture is that the heat treatment process itself is a time-dependent process; the recurrent structure of RNN can effectively capture and learn this time-series dependency, that is, generate the optimal parameters for the next stage based on the state and output of the previous stage, which is highly consistent with the physical nature of the heat treatment process.
[0091] Initializing the network's initial hidden state using a defective topology tensor is key to the sequence generation network design of this invention. Traditional RNNs typically use zero vectors or random values as their initial hidden state; however, in this embodiment, the defective topology tensor constructed in the preceding steps is used... Alternatively, their reduced feature representations can be directly used as the initial hidden states of an RNN. The initial hidden state here. It is the internal state of the RNN before processing the first time step of the sequence, and it carries the context information before the start of the sequence;
[0092] By using the defect topology tensor As the initial condition of the network, the entire process path generation process is aware of all the initial defect information of the workpiece to be processed from the very beginning. This design goes beyond simply using defect information as a common input feature of the network, but elevates it to a context or boundary condition that defines the starting point of the entire generation process. This enables the network to generate highly personalized and targeted process parameter sequences, improving the optimization scheme's ability to repair defects in specific workpieces and the accuracy of achieving the final performance.
[0093] Example 4:
[0094] The composite loss function of physical constraints consists of the following parts:
[0095] The dimensionless performance loss is determined based on the difference between the performance prediction error corresponding to the optimal process parameter sequence and the target performance vector.
[0096] The dimensionless physical fit loss is determined based on the difference between the evolution rate of the normalized state vector and the prediction rate of the normalized physical model.
[0097] The dimensionless time cost loss is determined based on the ratio of the total process time to the preset reference time.
[0098] This embodiment is a composite loss function for the physical constraints of Embodiment 1. A detailed explanation of its specific composition and design principles; the design of this loss function is crucial to ensuring the physical meaning and engineering value of the optimization results; ;
[0099] in, The weighting coefficients for dimensionless performance loss. For the dimensionless physical conformity loss, The weighting coefficients for the dimensionless time cost loss; this composite loss function is composed of the following three designed dimensionless components weighted together: dimensionless performance loss The calculation formula is as follows: ;
[0100] The purpose of this is to drive the final performance achieved by the optimized process path. Infinitely approximating the target performance vector of engineering design ;in, It is a performance prediction model, which can specifically employ a neural network model, such as a multilayer perceptron (MLP) network with two hidden layers, based on the final microstructure state at the end of heat treatment. Predict macroscopic performance; It is based on the physical model and the process parameter sequence The results were obtained from evolutionary calculations, among which... This indicates the calculation of the L2 norm of a vector.
[0101] By calculating and normalizing the Euclidean distance between the predicted and target performance, the loss term is made a dimensionless scalar; this ensures the performance prediction model... For accuracy, its internal parameters need to be fitted using a set of independent calibration experimental data; for example, preparing a series of samples with different microstructures. The sample was tested, and its corresponding macroscopic properties, such as hardness, were measured. and resilience This forms a dataset; based on this dataset, the model is calibrated using methods such as regression analysis. Parameters; dimensionless physical conformity loss The calculation formula is as follows: ;
[0102] In the formula Indicates time Temperature parameters; the purpose of this is to ensure that the evolution path of process parameters strictly follows known materials science and physics laws; it is achieved by comparing the normalized state vector. actual evolution rate Compared with physical dynamics models This is achieved through the difference between predicted evolution rates; the normalized state vector here... A dimensionless vector is formed by dividing each of the microscopic structural states containing different dimensions by a reference value.
[0103] To ensure that the loss term is dimensionless, a characteristic time constant is introduced into the formula. Used to convert the evolution rate term Dimensionless, and then in the total process time Integrate and average over, thus ensuring The entire loss term is a dimensionless scalar, and this constant originates from the material's physical properties or characteristic processing time, used to make the entire loss term dimensionless; physical dynamics model. Similarly, these are pre-calibrated using independent experimental data. Specifically, phase-field models or cellular automata models can be used to describe the evolution of the microstructure. For example, a kinetic model based on JMAK theory can be used to describe the isothermal phase transition process, minimizing dimensionless time cost losses. The calculation formula is as follows: ;
[0104] The purpose of this project is to find a shorter process time while meeting performance and physical constraints; it achieves this by reducing the total process time. With a preset reference duration Divide to get;
[0105] By transforming the three core optimization objectives—performance, physical laws, and process time—into dimensionless loss terms, this invention solves the common problems of dimension inconsistency and difficulty in setting weights in multi-objective optimization; this makes the weight coefficients... The physical meaning of these becomes clear: they purely represent the designer's preference for the trade-offs between performance achievement, process authenticity, and production efficiency.
[0106] Example 5:
[0107] The defect passivation index is calculated, including:
[0108] Based on the defect topology tensor, the initial stress concentration factor is calculated through finite element analysis;
[0109] In finite element analysis, thermal stress and microstructure transformation stress generated by the optimal process parameter sequence are applied, and the stress concentration factor after processing is calculated.
[0110] The defect passivation index is determined based on the initial stress concentration factor and the stress concentration factor after treatment.
[0111] This embodiment describes the specific implementation of the key evaluation step of calculating the defect passivation index in the method of embodiment 1;
[0112] Based on the defect topology tensor, the initial stress concentration factor is calculated through finite element analysis: using the defect topology tensor Using the geometric and residual stress field information, a high-fidelity digital twin model of the printed part in its initial state is established in finite element analysis (FEA) software. Preset load conditions are applied to this model, and stress analysis is performed to calculate the initial stress concentration factor at the tip of the critical defect. The initial stress concentration factor here. It is a dimensionless parameter used to characterize the factor by which the presence of a defect causes the local stress around it to increase relative to the nominal stress.
[0113] In the finite element analysis, thermal stress and microstructure transformation stress generated by the optimal process parameter sequence are applied, and the stress concentration factor after processing is calculated. Based on the above finite element model, further stress is applied by the optimal process parameter sequence. The heat treatment process is described; this includes simulating the thermal stress generated by heating and cooling processes, as well as the microstructure transformation stress caused by solid-state phase transformation; under the combined action of the internal stress field introduced by these processes and external loads, the stress distribution at the same defect location is recalculated to obtain the stress concentration factor after treatment. .
[0114] Based on the initial stress concentration factor and the stress concentration factor after treatment, the defect passivation index is determined using the following formula: The calculation formula is as follows: ;
[0115] in, The initial stress concentration factor is . This refers to the stress concentration factor after treatment; the defect passivation index is used here. The physical significance lies in quantifying the effectiveness of heat treatment processes in reducing stress concentration at defect tips; The closer the value is to 1, the more significantly the stress concentration is reduced, the defect is effectively passivated, and the better the repair effect.
[0116] This embodiment precisely quantifies the correlation between macroscopic heat treatment process parameters and microscopic fracture mechanics behavior through finite element simulation; it provides an intuitive and reliable indicator. This allows for a direct evaluation of the actual effectiveness of the process solution in repairing inherent printing defects. This enables the optimization process to go beyond simply pursuing abstract macroscopic performance indicators and delve into the root causes of part failure, thereby significantly improving the reliability of the optimization solution and the fatigue life of the parts.
[0117] Example 6:
[0118] Add penalties, including:
[0119] In the composite loss function, a penalty term is added, which is proportional to the stress concentration factor after treatment.
[0120] This embodiment is a detailed explanation of the key control action of adding a penalty term in the closed-loop correction mechanism of Embodiment 1;
[0121] In the composite loss function, a penalty term is added, which is proportional to the stress concentration factor after processing: when the feedback control logic determines the defect passivation index of the current optimal path. Below the preset threshold At that time, the system will automatically adjust the composite loss function. The correction is made by adding a new penalty term to the original function; this penalty term is designed to be related to the stress concentration factor after processing. Proportional to the original value, the corrected loss function takes the following form: ;
[0122] in, It is the original composite loss function. It is a positive weighting coefficient used to adjust the intensity of the penalty;
[0123] This correction method directly incorporates the microscopic mechanical evaluation results. This was incorporated into the macroscopic process parameter optimization objective; in the next round of gradient descent optimization, due to... It was added to the loss function, and the optimizer minimizes it. Therefore, we must find ways to lead to lower... Value process parameter sequence ;
[0124] This is equivalent to applying a clear guide in the direction of optimization, forcing the algorithm to explore additional process regions that can more effectively reduce the stress at the defect tip and achieve better passivation effects while satisfying the original goal. This dynamic and adaptive penalty mechanism ensures that the final convergence point of the optimization process is a solution that meets the requirements in both macroscopic performance and microscopic reliability.
[0125] This invention also provides an intelligent optimization system for the heat treatment process of metal 3D printed parts, which is the hardware implementation carrier of the method in any of the above embodiments.
[0126] Example 7:
[0127] Please see Figure 2 A smart optimization system for heat treatment process of metal 3D printed parts, comprising:
[0128] The defect characterization module is used to obtain the initial defect field of the printed part and construct the defect topology tensor based on the initial defect field.
[0129] The target setting module is used to set the target performance vector according to engineering design requirements;
[0130] The process path generation module is used to generate an initial process parameter sequence based on the defect topology tensor through a sequence generation network.
[0131] The process path optimization module is used to combine the initial process parameter sequence with the target performance vector and determine the optimal process parameter sequence through a composite loss function of physical constraints.
[0132] The closed-loop evaluation and correction module is used to calculate the defect passivation index based on the optimal process parameter sequence, and correct the composite loss function or output the final heat treatment process scheme based on the comparison result between the defect passivation index and the preset passivation threshold.
[0133] This system implements all the functions of the method through modular design; the system includes:
[0134] The defect characterization module physically integrates industrial CT scanning equipment and XRD measurement equipment, and is equipped with corresponding computing units. Its function is to acquire the initial defect field of the printed part as described in the aforementioned method, and then, through a built-in algorithm model, construct a defect topology tensor based on the initial defect field. The output of this module is a structured defect topology tensor. .
[0135] The target setting module is typically a software interface or human-computer interaction interface that allows engineers or operators to input and set target performance vectors according to specific engineering design requirements. ;
[0136] The process path generation module, at its core, is a processor or computing server deployed with a pre-trained sequence generation network; this module receives data from the defect characterization module. As its initial network state, and based on the defect topology tensor, an initial process parameter sequence is generated through a sequence generation network. Functions;
[0137] The process path optimization module is the computational core of the system, typically composed of high-performance computing units; it receives initial data from the generation module. and from the settings module It internally incorporates the calculation logic of the composite loss function based on physical constraints, and uses an iterative optimization algorithm to determine the optimal sequence of process parameters.
[0138] The closed-loop evaluation and correction module integrates finite element analysis software and control logic unit. It receives the current optimal process parameter sequence from the optimization module, performs the function of calculating the defect passivation index based on the optimal process parameter sequence, and corrects the composite loss function or outputs the final heat treatment process scheme based on the comparison result of the defect passivation index and the preset passivation threshold.
[0139] This system integrates complex data acquisition, model calculation, optimization iteration, and closed-loop evaluation processes into an automated modular system. Each module has a clear responsibility and works collaboratively, realizing full-process intelligentization from raw material printing to the optimal heat treatment solution. This reduces reliance on human experience and improves the automation level and repeatability of the process determination process.
[0140] Example 8:
[0141] The closed-loop evaluation and correction module includes:
[0142] The effect evaluation unit is used to calculate the defect passivation index based on the optimal process parameter sequence and through a preset defect passivation evaluation model.
[0143] The feedback control unit is used to add a penalty term to the composite loss function and instruct the process path optimization module to redetermine the optimal process parameter sequence in response to the defect passivation index being lower than the preset passivation threshold.
[0144] The scheme output unit is used to output the optimal process parameter sequence as the final heat treatment process scheme in response to the defect passivation index not being lower than the preset passivation threshold.
[0145] Before implementing this invention, the robustness of the constructed mathematical model needs to be checked, which includes:
[0146] Boundary condition testing: Testing the defect topology tensor The components are set to extreme values, such as porosity close to 0 or the theoretical maximum value, to verify whether the model’s behavior under extreme conditions conforms to physical common sense.
[0147] Noise robustness test: Random noise is added to the initial defect field data to simulate measurement error and to evaluate whether the process scheme output by the model remains stable.
[0148] Sensitivity analysis: Analyze the model's sensitivity to different input parameters to identify the key factors that have the greatest impact on the final optimization result;
[0149] This embodiment is a further detailed explanation of the internal structure and collaborative working method of the closed-loop evaluation and correction module in the system of Embodiment 7; this module consists of the following three interrelated units:
[0150] Effect Evaluation Unit: This unit is the executor of the evaluation function and integrates a preset defect passivation evaluation model; it receives the optimal process parameter sequence from the process path optimization module. Based on this, the defect passivation index is calculated. ;
[0151] Feedback Control Unit: This unit is the core of decision-making and control; it receives feedback from the effect evaluation unit. and compare it with the preset passivation threshold in internal storage. The comparison is performed; in response to a defect passivation index falling below a preset passivation threshold, the unit is activated and performs a correction action: a penalty term is added to the composite loss function, and the process path optimization module is instructed to use this corrected new loss function. To redetermine the optimal sequence of process parameters.
[0152] Solution Output Unit: This unit is the output of the final result; it also receives judgment signals from the feedback control unit; in response to the defect passivation index not being lower than the preset passivation threshold, this unit is activated, and the optimal process parameter sequence determined by the current process path optimization module is applied. It is output as the final heat treatment process solution.
[0153] By subdividing the closed-loop evaluation and correction module into three functional units—evaluation, control, and output—this system decouples the evaluation and control logic, making the entire closed-loop feedback mechanism clearer, more reliable, and easier to maintain. The effect evaluation unit focuses on accurate calculation, the feedback control unit focuses on correct decision-making, and the scheme output unit focuses on stable output. This refined functional division ensures that the system can perform adaptive optimization efficiently and accurately, thereby guaranteeing the quality of the final output scheme and realizing intelligent closed-loop control.
[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for intelligent optimization of heat treatment process for metal 3D printed parts, characterized in that, include: Obtain the initial defect field of the printed part; Based on the initial defect field, construct the defect topology tensor; Based on the engineering design requirements, a target performance vector is set; Based on the defect topology tensor, an initial process parameter sequence is generated through a sequence generation network; By combining the initial process parameter sequence with the target performance vector, the optimal process parameter sequence is determined through a composite loss function of physical constraints. Based on the optimal process parameter sequence, the defect passivation index is calculated using a pre-set defect passivation evaluation model. If the defect passivation index is lower than the preset passivation threshold, a penalty term is added to the composite loss function, and the optimal process parameter sequence is re-determined based on the modified composite loss function. If the defect passivation index is not lower than the preset passivation threshold, the optimal process parameter sequence is output as the final heat treatment process scheme.
2. The intelligent optimization method for heat treatment process of metal 3D printed parts according to claim 1, characterized in that, Constructing the defect topology tensor includes: The porosity of the printed parts was obtained using industrial computed tomography (CT) scans. The residual stress of the printed part was measured using X-ray diffraction technology; Stress field reconstruction based on measured residual stress; Porosity and reconstructed residual stress are standardized inputs to generate the defect topology tensor.
3. The intelligent optimization method for heat treatment process of metal 3D printed parts according to claim 1, characterized in that, Sequence generation networks, including: It adopts an architecture based on recurrent neural networks; The initial hidden state of the network is initialized using the defect topology tensor.
4. The intelligent optimization method for heat treatment process of metal 3D printed parts according to claim 1, characterized in that, The composite loss function of physical constraints consists of the following parts: The dimensionless performance loss is determined based on the difference between the performance prediction error corresponding to the optimal process parameter sequence and the target performance vector. The dimensionless physical fit loss is determined based on the difference between the evolution rate of the normalized state vector and the prediction rate of the normalized physical model. The dimensionless time cost loss is determined based on the ratio of the total process time to the preset reference time.
5. The intelligent optimization method for heat treatment process of metal 3D printed parts according to claim 1, characterized in that, The defect passivation index is calculated, including: Based on the defect topology tensor, the initial stress concentration factor is calculated through finite element analysis; In finite element analysis, thermal stress and microstructure transformation stress generated by the optimal process parameter sequence are applied, and the stress concentration factor after processing is calculated. The defect passivation index is determined based on the initial stress concentration factor and the stress concentration factor after treatment.
6. The intelligent optimization method for heat treatment process of metal 3D printed parts according to claim 1, characterized in that, Add penalties, including: In the composite loss function, a penalty term is added, which is proportional to the stress concentration factor after treatment.
7. A smart optimization system for the heat treatment process of metal 3D printed parts, based on the smart optimization method for the heat treatment process of metal 3D printed parts according to any one of claims 1-6, characterized in that, include: The defect characterization module is used to obtain the initial defect field of the printed part and construct the defect topology tensor based on the initial defect field. The target setting module is used to set the target performance vector according to engineering design requirements; The process path generation module is used to generate an initial process parameter sequence based on the defect topology tensor through a sequence generation network. The process path optimization module is used to combine the initial process parameter sequence with the target performance vector and determine the optimal process parameter sequence through a composite loss function of physical constraints. The closed-loop evaluation and correction module is used to calculate the defect passivation index based on the optimal process parameter sequence, and correct the composite loss function or output the final heat treatment process scheme based on the comparison result between the defect passivation index and the preset passivation threshold.
8. The intelligent optimization system for heat treatment process of metal 3D printed parts according to claim 7, characterized in that, The closed-loop evaluation and correction module includes: The effect evaluation unit is used to calculate the defect passivation index based on the optimal process parameter sequence and through a preset defect passivation evaluation model. The feedback control unit is used to add a penalty term to the composite loss function and instruct the process path optimization module to redetermine the optimal process parameter sequence in response to the defect passivation index being lower than the preset passivation threshold. The scheme output unit is used to output the optimal process parameter sequence as the final heat treatment process scheme in response to the defect passivation index not being lower than the preset passivation threshold.