Laser melting forming pure copper microstructure parameter optimization method

Abstract

The invention relates to the technical field of pure copper additive manufacturing, in particular to a laser melting forming pure copper microstructure parameter optimization method. According to the method, specially-made sensing powder is used for carrying out a high-throughput experiment, molten pool sensing signals and microstructure data are synchronously collected, and a basic agent model is trained; in the process, the consistency of microstructure quantification is ensured by using a double-sensitization corrosive agent; migration fine tuning can be carried out on the basic model only through verification data of a small amount of commercially available pure copper powder, and a calibration model is obtained; performing high-throughput calculation by using the calibration model, and quickly searching and determining a robust process window meeting a target; the invention provides a data-driven optimization method, which solves the two technical problems of low signal-to-noise ratio of a molten pool signal and difficulty in micro-structure quantification in pure copper laser melting forming through the self-made sensing powder and corrosive agent.

Description

Technical Field

[0001] This invention relates to the field of pure copper additive manufacturing technology, specifically to a method for optimizing the parameters of pure copper microstructure formed by laser melting. Background Technology

[0002] Due to its high laser reflectivity and high thermal conductivity, pure copper has an extremely narrow process window during laser melting and forming, and the molten pool is unstable, with weak and difficult-to-capture thermal signals. This poses a severe challenge to data-driven process optimization. On the one hand, when using conventional copper powder for printing, the signal-to-noise ratio of the molten pool sensing signal is extremely low, making it impossible for machine learning models to establish an accurate physical mapping relationship between transient fluctuations in the molten pool and the final porosity defects. On the other hand, in the microstructure quantification stage, traditional metallographic etchants lack the ability to distinguish between pores and grain boundaries, resulting in low image contrast. This makes it impossible for artificial intelligence image recognition algorithms to accurately quantify density and grain morphology, thus feeding the model contaminated and low-quality data. Furthermore, even when using specific sensing powders to train the model, if only a single fixed ratio is used, the model will be unable to learn the impact of raw material batch differences on the physical process of the molten pool, resulting in poor model generalization ability and decreased robustness when applied to commercially available pure copper powder. In summary, the existing technology lacks a data closed loop that simultaneously solves the problems of low fidelity at the input end and low accuracy at the output end, resulting in the inability to efficiently and robustly search for process windows that meet the requirements of high-performance components. Summary of the Invention

[0003] The purpose of this invention is to provide a method for optimizing parameters of pure copper microstructures formed by laser melting, aiming to solve the problem of data loop gaps caused by low input fidelity and low output accuracy in existing technologies. This method enables efficient and robust searching for process windows that meet the requirements of high-performance components. Specifically, this method includes the following steps: (a) Prepare a series of sensing powders and a dual-sensitizing etchant; (b) High-throughput experiment: Using the series of sensing powders mentioned above, an orthogonal experimental matrix printing was performed, which included laser power, scanning speed, scanning spacing and sensing material ratio, and the molten pool sensing signal was collected simultaneously. (c) Microstructure quantification: The printed test block is cut, inlaid and polished, and metallographic etching is performed using the dual-sensitized etchant. Metallographic images are collected and the density and grain morphology are quantified. (d) Basic model training: Establish a basic proxy model, whose input is the process parameters and the ratio of sensing raw materials, and whose output is the density, grain morphology and molten pool sensing signal characteristics. (e) Model migration and calibration: A small-batch verification experiment was performed using commercially available pure copper powder, and the microstructure of the sample was quantified using the dual-sensitized etchant. The obtained real data was used to fine-tune the basic proxy model to obtain a calibration model. (f) Process window search: Using the calibration model, input process parameters to perform high-throughput calculations and search for process windows that meet the target compactness and grain morphology; (g) Determine the robust process window: Based on the process window search results, determine the final robust process window.

[0004] Preferably, the sensing powder is an indium-cerium oxide@copper composite sensing powder, the preparation of which includes the following steps: (1) Preparation of cerium oxide nanocrystals: Cerium nitrate and citric acid were dissolved in deionized water, stirred and mixed, pH was adjusted, and the mixture was transferred to a reaction vessel for hydrothermal reaction. Then, the mixture was filtered, washed and dried to obtain cerium oxide nanocrystals. (2) Powder composite: The cerium oxide nanocrystals, indium nitrate and PVP dispersant are added to anhydrous ethanol and ultrasonically dispersed; commercially available pure copper powder is added and mechanically stirred at 60°C; the mixed slurry is spray-dried; and finally, heat treatment is carried out in a mixed atmosphere with gradient heating to obtain the sensing powder.

[0005] Preferably, the preparation of the dual-sensitized etchant includes the following steps: (1) Preparation of solution A: Dissolve thiourea and potassium dichromate in deionized water and stir magnetically until dissolved; (2) Preparation of solution B: Dissolve benzotriazole in anhydrous ethanol and mix to dissolve; (3) Final mixing: Before use, mix the A solution and B solution with the standard ferric nitrate and copper chloride etching solution, and let it stand to remove bubbles before use.

[0006] Preferably, in the step of preparing the cerium oxide nanocrystals, the mass ratio of cerium nitrate, citric acid, and deionized water is 10:8:100; and the pH is adjusted to 10 using sodium hydroxide solution.

[0007] Preferably, the cerium nitrate and citric acid are stirred and mixed at 60°C for 0.5 hours; the hydrothermal reaction is carried out at 150°C for 12 hours.

[0008] Preferably, the heat treatment is carried out in a hydrogen / argon (5% / 95%) mixed atmosphere, with the temperature gradually increased to 200°C and held for 2 hours.

[0009] Preferably, in solution A, the mass ratio of thiourea, potassium dichromate, and deionized water is 2:1:100; and in solution B, the mass ratio of benzotriazole and anhydrous ethanol is 1:50.

[0010] Preferably, in the final mixture, the volume ratio of liquid A, liquid B, and standard ferric nitrate and copper chloride etching solution is 10:1:100.

[0011] Preferably, after the dual-sensitized etchant etches the printed test block, the edges of the pore defects are deeply etched, while the grain boundaries are light gray, thus presenting a significantly different contrast in the metallographic image.

[0012] Preferably, in the training of the basic proxy model, the features of the molten pool sensing signal are used as auxiliary outputs, forcing the model to understand the intermediate processes of molten pool physics while learning the mapping of process parameters and raw material ratios to microstructure indicators.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention provides a data-driven optimization method that solves two major technical challenges in pure copper laser melting and forming: low signal-to-noise ratio of the molten pool signal and difficulty in microstructure quantification. The indium-cerium oxide@copper composite sensing powder used in this invention leverages the selective evaporation of indium and the high-temperature stability of cerium oxide to generate a strong thermal signal contrast at the molten pool boundary. This significantly improves the signal-to-noise ratio of the molten pool thermal signal, enabling the model to capture high-fidelity intermediate physical processes of the molten pool, overcoming the optimization failure problem in Comparative Example 1 caused by the lack of sensing powder and signal blurring. The dual-sensitized etchant used in this invention, through the synergistic effect of liquid A and liquid B, makes the porosity defects and grain boundaries in the metallographic image exhibit significantly different contrasts. This greatly improves the accuracy of AI image recognition and eliminates the problem of model training inaccuracies caused by low-quality data.

[0014] 2. The method of this invention constructs a high-quality data closed loop covering the entire chain from raw materials to the molten pool to the microstructure. By using a series of sensing powders with different proportions for high-throughput experiments, the basic model is forced to learn the influence of raw material properties on the physical processes of the molten pool, thereby obtaining strong generalization ability and robustness, overcoming the defect of insufficient model generalization ability caused by using only a single powder; when the basic model is transferred to commercially available pure copper powder, it can be quickly calibrated with only a small batch of real data for fine-tuning; this method has a fast convergence speed and can efficiently and robustly lock the optimal process window of commercially available pure copper, stably preparing high-performance pure copper components with a density of up to 99.91% to 99.98% and controllable grain size, meeting the stringent requirements for microstructure control in high-end applications such as integrated circuit micro heat sinks, high-performance motor windings, and regenerative cooling cavities for aerospace engines. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for optimizing the parameters of a pure copper microstructure formed by laser melting according to the present invention. Figure 2Comparative graphs showing the thermal signal and microstructure analysis of Example 1, Comparative Example 1, and Comparative Example 2; Figure 3 A quantitative comparison chart of the performance of key components (sensing powder and etchant); Figure 4 Line graph showing the effect of the number of optimization iterations on the final compaction density (Examples 1-5). Detailed Implementation

[0016] The process of the laser melting and forming pure copper microstructure parameter optimization method of the present invention is as follows: Figure 1 As shown, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0017] This embodiment provides a method for optimizing the parameters of pure copper microstructure formed by laser melting, including the following steps: (a) Prepare a series of sensing powders and a dual-sensitized etchant; wherein the sensing powder is an indium-cerium oxide@copper composite sensing powder; its preparation includes: (1) Preparation of cerium oxide nanocrystals: dissolve cerium nitrate (10 parts by mass) and citric acid (8 parts by mass) in deionized water (100 parts by mass), stir and mix at 60°C for 0.5 hours, adjust the pH to 10 using sodium hydroxide solution; transfer to a reaction vessel for hydrothermal reaction at 150°C for 12 hours, then filter, wash and dry to obtain cerium oxide nanocrystals; (2) Powder composite: add cerium oxide nanocrystals, indium nitrate and PVP dispersant to anhydrous ethanol for ultrasonic dispersion (Note: in this embodiment, the preparation of a series of sensing powders is achieved by systematically adjusting the ratio of indium nitrate to cerium oxide). The ratio of nanocrystals); add commercially available pure copper powder, mechanically stir; spray dry; finally, in a hydrogen / argon (5% / 95%) mixed atmosphere, heat the mixture to 200℃ and keep it for 2 hours to obtain the series of sensing powders; wherein, the preparation of the dual-sensitized etchant includes: (1) Preparation of solution A: dissolve thiourea (2 parts by mass) and potassium dichromate (1 part by mass) in deionized water (100 parts by mass), and stir magnetically until dissolved; (2) Preparation of solution B: dissolve benzotriazole (1 part by mass) in anhydrous ethanol (50 parts by mass), and mix and dissolve; (3) Final mixing: before use, mix solution A, solution B and standard ferric nitrate and copper chloride etching solution in volume parts of 10:1:100, and let stand to remove bubbles before use; (b) High-throughput experiment: Using the series of sensing powders mentioned above, an orthogonal experimental matrix printing was performed, which included laser power, scanning speed, scanning spacing and sensing material ratio, and the molten pool sensing signal was collected simultaneously. (c) Microstructure quantification: The printed test block is cut, inlaid and polished, and metallographic etching is performed using the dual-sensitized etchant; the etchant causes the edges of pore defects to be deeply etched, while the grain boundaries are light gray, showing a significantly different contrast in the metallographic image; metallographic images are acquired and the density and grain morphology are quantified. (d) Basic model training: Establish a basic surrogate model, whose input is the process parameters and the ratio of sensing raw materials, and whose output is the density, grain morphology and molten pool sensing signal characteristics; during training, the molten pool sensing signal characteristics are used as auxiliary outputs, forcing the model to understand the intermediate process of molten pool physics when learning the mapping of process parameters and raw material ratios to microstructure indicators. (e) Model migration and calibration: A small-batch verification experiment was performed using commercially available pure copper powder, and the microstructure of the sample was quantified using the dual-sensitized etchant. The obtained real data was used to fine-tune the basic proxy model to obtain a calibration model. (f) Process window search: Using the calibration model, input process parameters to perform high-throughput calculations and search for process windows that meet the target compactness and grain morphology; (g) Determine the robust process window: Based on the process window search results, determine the final robust process window; The optimization method in this embodiment can determine the process window that meets the target after 50 optimization iterations. Pure copper components manufactured using this window, such as micro heat sinks for integrated circuits, can achieve a density of 99.91% and an average grain size controlled at 68.5 μm, which meets the requirements of high-performance components for microstructure control. Example 2

[0018] This embodiment provides a method for optimizing the parameters of pure copper microstructure formed by laser melting, and its steps (a) to (g) are the same as in Embodiment 1; the difference lies in the preparation parameters of the sensing powder and the dual-sensitized etchant: In step (a), when preparing cerium oxide nanocrystals, cerium nitrate (10 parts by mass) and citric acid (8 parts by mass) were dissolved in deionized water (100 parts by mass), the pH was adjusted to 10, and the hydrothermal reaction was carried out at 155°C for 11 hours; the heat treatment of the powder composite was maintained at 210°C for 2 hours; when preparing the dual-sensitized etchant, solution A was prepared with thiourea (2.2 parts by mass), potassium dichromate (1.1 parts by mass), and deionized water (100 parts by mass); solution B was prepared with benzotriazole (1 part by mass) and anhydrous ethanol (50 parts by mass); when finally mixed, the volume ratio of solution A, solution B, and standard etchant was 10:1.1:100; The optimization method in this embodiment can further stabilize the process window after 100 optimization iterations. Pure copper components manufactured using this window, such as irregular armature windings for high-performance motors, can achieve a density of 99.95% and an average grain size controlled at 72.3 μm, exhibiting good microstructure consistency. Example 3

[0019] This embodiment provides a method for optimizing the parameters of pure copper microstructure formed by laser melting, and its steps (a) to (g) are the same as in Embodiment 1; the difference lies in the preparation parameters of the sensing powder and the dual-sensitized etchant: In step (a), when preparing cerium oxide nanocrystals, cerium nitrate (10 parts by mass) and citric acid (8 parts by mass) were dissolved in deionized water (100 parts by mass), the pH was adjusted to 10, and the hydrothermal reaction was carried out at 145°C for 13 hours; the heat treatment of the powder composite was maintained at 190°C for 2.5 hours; when preparing the dual-sensitized etchant, solution A was prepared with thiourea (1.8 parts by mass), potassium dichromate (0.9 parts by mass), and deionized water (100 parts by mass); solution B was prepared with benzotriazole (1 part by mass) and anhydrous ethanol (45 parts by mass); when finally mixed, the volume ratio of solution A, solution B, and standard etchant was 9:1:100. The optimization method in this embodiment can obtain a highly robust process window after 150 optimization iterations; the pure copper parts manufactured using this window can achieve a density of 99.96% and an average grain size controlled at 70.5μm, making them suitable for regenerative cooling cavities of aerospace engines with low defect tolerance. Example 4

[0020] This embodiment provides a method for optimizing the parameters of pure copper microstructure formed by laser melting, and its steps (a) to (g) are the same as in Embodiment 1; the difference lies in the preparation parameters of the sensing powder and the dual-sensitized etchant: In step (a), when preparing cerium oxide nanocrystals, cerium nitrate (12 parts by mass) and citric acid (9 parts by mass) are dissolved in deionized water (100 parts by mass), the pH is adjusted to 10, and the hydrothermal reaction is carried out at 150°C for 10 hours; the heat treatment of the powder composite is maintained at 200°C for 2 hours; when preparing the dual-sensitized etchant, solution A is prepared with thiourea (2 parts by mass), potassium dichromate (1 part by mass) and deionized water (100 parts by mass); solution B is prepared with benzotriazole (1.2 parts by mass) and anhydrous ethanol (50 parts by mass); when finally mixed, the volume ratio of solution A, solution B and standard etchant is 10:1:100; The optimization method in this embodiment can finely control the grain morphology after 180 optimization iterations; the pure copper components manufactured using this window can achieve a density of 99.97% and an average grain size controlled at 75.1 μm, meeting the requirements of specific electrical components for conductivity and microstructure. Example 5

[0021] This embodiment provides a method for optimizing the parameters of pure copper microstructure formed by laser melting, and its steps (a) to (g) are the same as in Embodiment 1; the difference lies in the preparation parameters of the sensing powder and the dual-sensitized etchant: In step (a), when preparing cerium oxide nanocrystals, cerium nitrate (9 parts by mass) and citric acid (7 parts by mass) were dissolved in deionized water (100 parts by mass), the pH was adjusted to 10, and the hydrothermal reaction was carried out at 160°C for 12 hours; the heat treatment of the powder composite was maintained at 210°C for 1.5 hours; when preparing the dual-sensitized etchant, solution A was prepared with thiourea (2.5 parts by mass), potassium dichromate (1 part by mass) and deionized water (100 parts by mass); solution B was prepared with benzotriazole (1 part by mass) and anhydrous ethanol (55 parts by mass); when finally mixed, the volume ratio of solution A, solution B and standard etchant was 11:1:100; The optimization method of this embodiment can obtain a highly stable and robust process window after 200 optimization iterations; the pure copper parts manufactured using this window can stably reach a density of 99.98%, and the average grain size is precisely controlled at 71.8μm; compared with the traditional trial and error method, this method has significant improvements in both optimization efficiency and forming quality.

[0022] In Examples 1-5 and Comparative Examples 1-3, cerium nitrate (Ce(NO3)3·6H2O) was from Shanghai Aladdin Biochemical Technology Co., Ltd., CAS No.: 10294-41-4, purity ≥99.5%; citric acid was from Sinopharm Chemical Reagent Co., Ltd., AR grade, CAS No.: 77-92-9; indium nitrate (In(NO3)3·xH2O) was from Shanghai Maclean Biochemical Technology Co., Ltd., purity ≥99.9%, CAS No.: 13770-61-1; PVP dispersant (polyvinylpyrrolidone) was from BASF (China) Co., Ltd., model K30, CAS No.: 9003-39-8; commercially available pure copper powder (used in the experiment) Examples and Comparative Examples 1) were from Jiangsu Tianyi Ultrafine Metal Powder Co., Ltd., spherical, D50=25μm, purity>99.9%; thiourea was from Shanghai Aladdin Biochemical Technology Co., Ltd., AR grade, CAS No.: 62-56-6; potassium dichromate was from Sinopharm Chemical Reagent Co., Ltd., GR grade, CAS No.: 7778-50-9; benzotriazole (BTA) was from Shanghai Maclean Biochemical Technology Co., Ltd., purity≥99.0%, CAS No.: 95-14-7; standard ferric nitrate and copper chloride etching solution (used in Comparative Example 2) was purchased from Suzhou Ruisibo Chemical Co., Ltd.; sodium hydroxide, anhydrous ethanol, hydrogen, and argon were all commercially available analytical grade or high-purity products.

[0023] Comparative Example 1: This comparative example provides a parameter optimization method, which differs from Example 1 in that: in the high-throughput experiment in step (b), the indium-cerium oxide@copper composite sensing powder is not used, but ordinary commercially available pure copper powder is used throughout the experiment; Due to the lack of sensing powder, the signal-to-noise ratio of the molten pool sensing signal is low during data acquisition. This results in the basic surrogate model being unable to accurately learn the physical relationship between molten pool fluctuations and porosity defects. Therefore, even based on 50 iterations of Example 1, the density finally obtained by the method in this comparative example is only 99.65%, which is lower than 99.91% of Example 1 and cannot meet the requirements of high-performance components.

[0024] Comparative Example 2: This comparative example provides a parameter optimization method, which differs from Example 1 in that: in step (c) microstructure quantification, the grain boundary / defect dual-sensitizing etchant is not used, but a standard ferric nitrate and copper chloride etching solution is used for metallographic etching. Due to the lack of a dual-sensitized etchant, the metallographic images obtained using ordinary etchants have low recognition accuracy, and the AI ​​image recognition algorithm cannot accurately distinguish between pore defects and grain boundaries. This leads to the provision of incorrect microstructure data to the basic proxy model, resulting in inaccurate model training. Therefore, even based on 50 iterations of Example 1, the final density obtained by the method in this comparative example is only 99.71%, which is lower than 99.91% of Example 1, demonstrating the importance of the accuracy of microstructure data to this method.

[0025] Comparative Example 3: This comparative example provides a parameter optimization method, which differs from Example 1 in that: in steps (a) and (b), instead of using a series of sensing powders, only a single fixed ratio of sensing powder (i.e., the baseline example ratio of indium-cerium oxide@copper composite sensing powder) is used for high-throughput experiments; in step (d) basic model training, the input does not include the variable of sensing material ratio. Because only a single fixed ratio of sensing powder is used, the basic surrogate model fails to learn the corresponding changes in the physical processes of the molten pool when the material properties (such as batch differences) change during training, resulting in insufficient model generalization ability; when this model is transferred to commercially available pure copper material in step (e), its calibration accuracy and robustness decrease; therefore, even based on 50 iterations of Example 1, the final density obtained by the method of this comparative example is 99.80%, which is lower than 99.91% of Example 1, demonstrating the necessity of using a series of sensing powders to improve the robustness of the model.

[0026] The optimized methods used in Examples 1-5 and Comparative Examples 1-3, and the pure copper components finally prepared, were subjected to the following performance tests, and the test results are shown below: (1) Convergence of optimization method and final forming quality test This test aims to compare the differences in optimization efficiency and final forming quality between the method of the present invention and a control method with missing key components; Examples 1-5 and Comparative Examples 1-3 were used as test groups. Each group employed its corresponding method to describe the process parameter optimization flow. The number of optimization iterations required for each group to reach its optimal process window was recorded. Final pure copper components were prepared using the optimal process window found by each group. The final density of the samples was tested using the Archimedes displacement method (using a Mettler Toledo MS105DU analytical balance, Switzerland), and the average grain size was measured using metallographic image analysis (using a Zeiss Axio Imager M2m microscope, Germany). Three samples were tested in each group, and the average value of the results was taken. Table 1: Comparison of optimization efficiency and final molding quality of each embodiment and comparative example

[0027] As can be seen from the data in Table 1, the present invention (Examples 1-5) has significant technical advantages over comparative examples (1-3); The comparison between Example 1 (complete scheme, 50 iterations) and Comparative Examples 1, 2, and 3 (all 50 iterations) clearly reveals the synergistic necessity of the two core self-made raw materials of this invention; Comparative Example 1 lacks sensing powder, and its density is only 99.65%, which is much lower than 99.91% in Example 1. This confirms that the lack of sensing powder will result in an extremely low signal-to-noise ratio of the molten pool thermal signal, and the model cannot establish an accurate physical mapping between molten pool fluctuations and pores, leading to optimization failure. Comparative Example 2 lacked the dual-sensitized etchant, and its density was 99.71%; this confirms that ordinary etchants prevent the AI ​​from distinguishing between pores and grain boundaries, providing junk data and causing the model training to be fundamentally inaccurate. Comparative Example 3 used only a single ratio of sensing powder, and its density was 99.80%, which was higher than that of Comparative Examples 1 and 2, but still lower than that of Example 1. This verifies the core point of the technical disclosure: using a series of sensing powders is the key to forcing the model to learn the impact of raw material batch differences on the physical process of the molten pool. The model of Comparative Example 3 had insufficient generalization ability, which led to a decrease in robustness when migrating to commercially available pure copper. Example 1 achieved a density of 99.91% and a grain size of 68.5 μm with only 50 iterations, fully meeting the core objectives. Examples 2-5 show that the density can be further optimized to 99.98% with increasing iteration count. Figure 4 As shown, the grain size remains stable within the target range (50-80 μm); this proves that the proposed method has a fast convergence speed and the results are robust and controllable. In summary, the complete solution of the present invention (Examples 1-5) constructs a high-quality data closed loop through the synergistic effect of a series of sensing powders and dual-sensitized etchants, ensuring that the model can learn profound physical insights, thereby efficiently and robustly locking the optimal process window for commercially available pure copper after transfer.

[0028] (2) Comparative test of the accuracy of microstructure quantification This test aims to quantify the effect of dual-sensitized etchant on improving the accuracy of AI image recognition and to verify its necessity as a source of high-quality data. Metallographic specimens were cut from the printed blocks of Example 1 (using a dual-sensitized etchant) and Comparative Example 2 (using a common etchant), respectively. Both sets of specimens were mounted and polished, and then etched using their respective etchants. One hundred metallographic images were randomly acquired from each set at the same magnification using a Zeiss Axio Imager M2m microscope (Germany). Three experienced materials engineers manually segmented all 200 images to create a ground truth dataset for pores and grain boundaries. The same AI image recognition algorithm (based on the U-Net architecture) was used to automatically segment both sets of images. The AI ​​segmentation results were compared pixel-level with the ground truth datasets to calculate the Intersection over Union (IoU) for pore identification and the Intersection over Union (IoU) for grain boundary identification. A higher IoU value (closer to 100%) indicates higher AI recognition accuracy. The data represents the average of 100 images. Table 2: Effects of different etchants on the accuracy of AI microstructure recognition

[0029] As shown in Table 2, Example 1 (using the dual-sensitized etchant of the present invention) has an overwhelming advantage over Comparative Example 2 (using a common etchant) in terms of accuracy in microstructure quantification. (See Table 2 for more details.) Figure 3 Left image; In Example 1, the AI ​​achieved an accuracy rate of approximately 99% in identifying pores and grain boundaries, verifying the effectiveness of the etchant of the present invention; while in Comparative Example 2, the AI's identification accuracy was only slightly over 80%. The huge difference stems from the formulation design of the etchant; the dual-sensitizing etchant in Example 1 contains liquid A and liquid B; liquid A preferentially sensitizes grain boundaries, while liquid B passivates the matrix but has a poor passivation effect on the interior of pores; the two work together to make the pore edges deeply etched and the grain boundaries light gray in the final image, which is easily distinguishable by AI. In Comparative Example 2, the ordinary etchant indiscriminately etched both pores and grain boundaries, resulting in similar morphology and contrast in the images. (See Figure 2.) Figure 2 (c) AI algorithms therefore frequently misidentify tiny pores as grain boundaries or coarse grain boundaries as pores. The low accuracy (81.35%) of Comparative Example 2 means that the machine learning model (Model A) was provided with heavily contaminated garbage data; Model A was trained based on the wrong target (e.g., the model was told that a region with pores was 99.9% dense), which fundamentally undermined its predictive ability and led to the failure of the final optimization process (denseness of only 99.71%); while the >99% accuracy of Example 1 ensured the absolute accuracy of the model training target (output); (3) Signal-to-noise ratio (SNR) test of molten pool sensing signal This test aims to quantify the effect of indium-cerium oxide@copper composite sensing powder on improving the quality of molten pool thermal signals and to verify its necessity as a high-fidelity data probe. The indium-cerium oxide@copper composite sensing powder used in Example 1 and the commercially available high-purity spherical pure copper powder used in Comparative Example 1 were used respectively. Single-track printing was performed on both powders using the exact same process parameters on an LMF (Low-Medium-Density) device. During the printing process, a high-speed thermal imager (FLIR X6900sc, USA) was used to simultaneously capture thermal radiation images of the molten pool. The acquired thermal image sequences were processed to extract the thermal signal intensity of the molten pool boundary region and the noise of the background powder bed, and the signal-to-noise ratio (SNR) of the thermal signal was calculated. Ten molten tracks were tested for each powder, with 1000 frames of images acquired for each track, and the average SNR value of the 10 tests was calculated. Table 3: Effect of different powders on the signal-to-noise ratio (SNR) of the molten pool thermal signal

[0030] As shown in Table 3, the SNR of the molten pool signal in Example 1 is much higher than that in Comparative Example 1, verifying the superiority of the sensing powder of the present invention. (See Table 3 for details.) Figure 3 Right image; This significant improvement in SNR stems from the special design of the indium-cerium oxide@copper composite sensing powder. Pure copper's high reflectivity and thermal conductivity to lasers result in inherently weak and unstable thermal signals in its molten pool. However, in the sensing powder of this invention, indium selectively evaporates at the edge of the molten pool, while cerium oxide remains stable at the high-temperature center. These two components produce distinctly different thermal signal responses in different regions of the molten pool, greatly enhancing the thermal signal contrast at the molten pool boundary. See [link to related documentation]. Figure 2 (a); Comparative Example 1 uses ordinary copper powder, and the signal-to-noise ratio captured by the thermal imager is extremely low; this means that the precise boundaries of the molten pool and transient fluctuations are completely submerged in background noise, see [link to example]. Figure 2 (b); The low SNR data in Comparative Example 1 is malnourished for the machine learning model; the model cannot establish a high-precision physical mapping relationship between molten pool fluctuations and porosity defects based on such fuzzy data; the model cannot learn the correct intermediate physical processes, which leads to its inability to converge effectively in step 3 training and ultimately optimization failure; while the high SNR data in Example 1 ensures high fidelity in the model training process. The data from tests (1), (2) and (3) form a complete chain of evidence. This invention ensures high fidelity at the model input end through a series of sensing powders (test 3), and at the same time ensures high accuracy at the model output end through dual-sensitized etchant. It is this high-quality data closed loop from raw materials to process to molten pool to microstructure that enables this method to solve the process optimization problem of pure copper additive manufacturing with extremely high efficiency and robustness.

[0031] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing the parameters of pure copper microstructure formed by laser melting, characterized in that, Includes the following steps: (a) Prepare a series of sensing powders and a dual-sensitizing etchant; (b) High-throughput experiment: Using the series of sensing powders mentioned above, an orthogonal experimental matrix printing was performed, which included laser power, scanning speed, scanning spacing and sensing material ratio, and the molten pool sensing signal was collected simultaneously. (c) Microstructure quantification: The printed test block is cut, inlaid and polished, and metallographic etching is performed using the dual-sensitized etchant. Metallographic images are collected and the density and grain morphology are quantified. (d) Basic model training: Establish a basic proxy model, whose input is the process parameters and the ratio of sensing raw materials, and whose output is the density, grain morphology and molten pool sensing signal characteristics. (e) Model migration and calibration: A small-batch verification experiment was performed using commercially available pure copper powder, and the microstructure of the sample was quantified using the dual-sensitized etchant. The obtained real data was used to fine-tune the basic proxy model to obtain a calibration model. (f) Process window search: Using the calibration model, input process parameters to perform high-throughput calculations and search for process windows that meet the target compactness and grain morphology; (g) Determine the robust process window: Based on the process window search results, determine the final robust process window.

2. The method according to claim 1, characterized in that, The sensing powder is an indium-cerium oxide@copper composite sensing powder, and its preparation includes the following steps: (1) Preparation of cerium oxide nanocrystals: Cerium nitrate and citric acid were dissolved in deionized water, stirred and mixed, pH was adjusted, and the mixture was transferred to a reaction vessel for hydrothermal reaction. Then, the mixture was filtered, washed and dried to obtain cerium oxide nanocrystals. (2) Powder composite: The cerium oxide nanocrystals, indium nitrate and PVP dispersant are added to anhydrous ethanol and ultrasonically dispersed; commercially available pure copper powder is added and mechanically stirred at 60°C; the mixed slurry is spray-dried; and finally, heat treatment is carried out in a mixed atmosphere with gradient heating to obtain the sensing powder.

3. The method according to claim 1, characterized in that, The preparation of the dual-sensitized etchant includes the following steps: (1) Preparation of solution A: Dissolve thiourea and potassium dichromate in deionized water and stir magnetically until dissolved; (2) Preparation of solution B: Dissolve benzotriazole in anhydrous ethanol and mix to dissolve; (3) Final mixing: Before use, mix the A solution and B solution with the standard ferric nitrate and copper chloride etching solution, and let it stand to remove bubbles before use.

4. The method according to claim 2, characterized in that, In the step of preparing the cerium oxide nanocrystals, the mass ratio of cerium nitrate, citric acid and deionized water is 10:8:100; the pH is adjusted to 10 using sodium hydroxide solution.

5. The method according to claim 4, characterized in that, The cerium nitrate and citric acid were stirred and mixed at 60°C for 0.5 hours; the hydrothermal reaction was carried out at 150°C for 12 hours.

6. The method according to claim 2, characterized in that, The heat treatment was carried out in a hydrogen / argon (5% / 95%) mixed atmosphere, with the temperature gradually increased to 200°C and held for 2 hours.

7. The method according to claim 3, characterized in that, In solution A, the mass ratio of thiourea, potassium dichromate, and deionized water is 2:1:100; in solution B, the mass ratio of benzotriazole and anhydrous ethanol is 1:

50.

8. The method according to claim 7, characterized in that, In the final mixture, the volume ratio of solution A, solution B, and standard ferric nitrate and copper chloride etching solution is 10:1:

100.

9. The method according to claim 1, characterized in that, After the dual-sensitized etchant etches the printed test block, the edges of the pore defects are deeply etched, while the grain boundaries are light gray, thus presenting a significantly different contrast in the metallographic image.

10. The method according to claim 1, characterized in that, In the training of the basic proxy model, the features of the molten pool sensing signal are used as auxiliary outputs, forcing the model to understand the intermediate processes of molten pool physics while learning the mapping of process parameters and raw material ratios to microstructure indicators.

Images