Full-automatic metallographic sample preparation method

Abstract

The invention discloses a full-automatic metallographic sample preparation method, and relates to the technical field of metallographic sample preparation. The method sequentially comprises the links of automatic water jet cutting, multi-stage intelligent sample grinding, self-adaptive polishing, cleaning and blow-drying, AI auxiliary corrosion, metallographic photographing and automatic storage, and is characterized in that a sample clamping head integrated with an XYZ omnidirectional mechanical sensor is adopted, and whether the next step of grinding is started or not is judged by monitoring fluctuation of contact force in real time; a high-precision line laser sensor is used for scanning the surface of the sample, and whether scratches meet the grinding or polishing requirements of the next step or not is accurately judged; the corrosion stage is based on an AI image recognition technology, compares sample surface tissue characteristics with a standard image, and automatically judges whether corrosion is in place or not. According to the invention, full-flow automation of metallographic sample preparation is realized, traditional manual operation errors are avoided, and the sample preparation quality stability and efficiency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metallographic sample preparation, in particular to a full-automatic metallographic sample preparation method integrating automatic cutting, intelligent sample grinding, self-adaptive polishing, AI-assisted etching, automatic detection and storage, which is suitable for sample preparation before microstructure analysis of metal materials. The method uses a mechanical replacement mechanism to realize automatic replacement of consumables, reducing equipment cost and maintenance difficulty, and can be widely used in fields such as mechanical manufacturing, automobile industry, aerospace, shipbuilding engineering and material science research, especially suitable for batch metallographic sample preparation scenarios that require sample preparation quality stability, automation degree and focus on equipment economy. BACKGROUND

[0002] Metallographic analysis is a core means for evaluating the performance of metal materials, judging the quality of materials and optimizing processing technology, and the precision of metallographic sample preparation directly determines the reliability of the analysis results. The traditional metallographic sample preparation process relies on manual operation, and the existing semi-automatic equipment also has many human decision-making links, which have the following key problems: 1. Grinding stage judgment relies on experience without reference standard: In the traditional sample grinding process, the operator judges whether to proceed to the next grinding step by observing the sample surface with the naked eye or relying on hand feeling, which cannot be quantified and is easy to lead to insufficient or excessive grinding; although some semi-automatic equipment is equipped with a downward pressure sensor, it cannot comprehensively monitor the contact state of the sample and the grinding disc, and there is still a judgment error.

[0003] 2. Surface scratch detection is missing, affecting the subsequent process: The traditional method lacks precise detection means for sample surface scratches, and if the scratches left in the previous stage exceed the removal capacity of the grinding disc or polishing cloth in the next stage, it will cause defects on the final sample surface, affecting metallographic observation; there is no uniform scratch detection standard after polishing, which completely relies on subjective judgment by human.

[0004] 3. Etching stage control is extensive and consistency is poor: Traditional etching requires manual immersion of the sample in etching solution, and the etching degree is judged by timing or visual observation, which cannot accurately identify whether the organizational features are clear, and is prone to problems of insufficient or excessive etching, and the quality consistency of batch samples is poor.

[0005] 4. Consumable replacement structure is complex and has low economic efficiency: The existing semi-automatic or automatic equipment mostly uses manual replacement of consumables, which is low in efficiency and difficult to adapt to batch sample preparation requirements.

[0006] 5. Low degree of automation in the whole process and low efficiency: The cutting, grinding, polishing and etching links are independent of each other, and manual sample transfer and intervention are required, which is easy to introduce human pollution; there is no associated optimization mechanism for parameters in each link, which cannot be adjusted adaptively according to the sample material, resulting in low sample preparation efficiency and large quality fluctuations.

[0007] To address the aforementioned issues, there is an urgent need in this field for a fully automated metallographic sample preparation method that combines "precise judgment with economic automation." This method would achieve precise judgment through omnidirectional mechanical monitoring, line laser scratch detection, and AI image recognition, while employing a mechanical replacement mechanism to reduce equipment costs and fill the gaps in precision control and economic automation offered by traditional methods. Summary of the Invention

[0008] The objective of this invention is achieved through the following technical solutions.

[0009] To address the problems in traditional metallographic sample preparation processes, such as reliance on experience-based judgment for grinding, lack of surface scratch detection, coarse control of corrosion levels, and low automation, especially the large fluctuations in sample quality and low efficiency caused by manual operation, the core objective of this invention is to provide a fully automated metallographic sample preparation method based on "precise quantitative judgment + economical automation." This improves the stability and efficiency of metallographic sample preparation quality, meeting the needs of batch sample preparation scenarios.

[0010] The fully automated metallographic sample preparation method described in this invention integrates functional modules such as automatic cutting, intelligent grinding, adaptive polishing, AI-assisted etching, and automatic cleaning and storage to form a coherent automated sample preparation process. The specific technical implementation of each step is as follows: 1. Automatic water jet cutting process: This process aims to achieve cold cutting of samples, avoiding the heat damage caused by traditional abrasive wheel cutting. The high-pressure water jet device starts cutting according to the preset path. After cutting, the sample is directly transferred to the grinding station by the robotic arm. There is no human contact throughout the process, which avoids sample contamination.

[0011] The specific process of automatic waterjet cutting is as follows: based on the material and thickness of the workpiece to be cut, the waterjet cutting pressure is preset to 100-400MPa, the cutting speed is 5-50mm / s and the cutting path is preset, an initial metallographic sample with a thickness of 15~20mm is obtained and transferred to the grinding station.

[0012] 2. Multi-stage intelligent sample grinding process (core innovation); This process achieves precise control and automation of the sample grinding process through "omnidirectional mechanical monitoring + line laser scratch detection".

[0013] (1) Fix the initial metallographic sample on the sample clamping head, which integrates an XYZ omnidirectional force sensor and an automatic downward pressure adjustment device; the clamping head slowly and gently touches the grinding disc of the current stage (the initial grinding disc is fixed at 180 mesh), and the XYZ sensor collects the X, Y, and Z three-axis contact force data in real time; when the three-axis force fluctuation is ≤ ±0.5N, it is determined that the sample and the grinding disc are in stable contact and the sample grinding stage is allowed; if the fluctuation exceeds the threshold, the position of the clamping head or the flatness of the grinding disc is automatically adjusted until the contact requirements are met.

[0014] (The above is completed in two stages, without the need for an AI model. 1. Automatic adjustment of Z-axis pressure: Based on the results of previous experiments, the system has collected enough data. Under the same pressure, that is, under the same Z-axis pressure, since the roughness of the grinding disc sandpaper is known, the pressure or tension values ​​brought by the rotation of the disc to the sensors in the X and Y directions can be judged. The range of values ​​can roughly correspond to the corresponding sample hardness range, thereby adjusting the Z-axis pressure. 2. Collect the fluctuation of the X and Y direction sensors. If the fluctuation range meets the requirements, continue grinding the sample. If it does not meet the requirements, the sample is re-stressed.) (2) The initial grinding parameters are preset according to the sample. During the grinding process, the XYZ sensor continuously monitors the change in contact force. When the sensor reaches the threshold range, the bearing head is lifted, and after cleaning and drying, the surface scratch is accurately detected to obtain the three-dimensional morphology data of the surface. The maximum scratch depth of the surface is calculated by the image analysis algorithm. If it is less than the grinding capacity of the next stage grinding disc, the current stage of grinding is determined to be completed. If the requirements are not met, the grinding time is extended by 0.5-2 minutes and then scanned and detected again until the scratch depth meets the standard.

[0015] (The above does not require an AI model. A high-precision line laser scanner is used to scan the polished surface of the sample. The image algorithm synthesizes the line laser data into 3D surface data to determine the surface scratches, including scratch length, scratch depth, and scratch location.) (3) During the current grinding process, the next stage grinding disc inspection is carried out simultaneously. A fast laser scanner is used to scan the surface of the ordinary magnetic grinding disc to be replaced to detect whether there are impurities such as metal chips, abrasive residue, or cracks, gaps, etc. If the grinding disc does not meet the standards, a spare grinding disc is automatically retrieved from the grinding disc storage rack.

[0016] (This step also uses a high-precision line laser scanner to scan the surface of the grinding disc, mainly to determine the presence of debris, impurities, and damage.) (4) AI parameter optimization: Input the sample material information (hardness, composition), XYZ sensor data (average force, fluctuation range), line laser scanning data (scratch depth, surface smoothness), grinding parameters (downward pressure, rotation speed, time) and grinding effect data of each grinding process into the AI ​​learning system; The system adopts a neural network algorithm to establish a "sample material-grinding parameters-grinding effect" correlation model, automatically optimize the grinding parameters of the same type of sample, and realize the adaptive iteration of grinding parameters.

[0017] This step uses a PSO-BP neural network. First, the BP model has a strong nonlinear mapping capability, which can establish a complex relationship between "sample material - grinding parameters - grinding effect". At the same time, the particle swarm optimization algorithm PSO is added to avoid local optima and improve the optimization efficiency. The sample material information (hardness, composition), XYZ sensor data (average force, fluctuation range), and line laser scanning data (scratch depth, surface smoothness) are used as inputs, and the output is the grinding parameters (Z-axis pressure, grinding speed, and grinding time for different grinding disc meshes). The results are fed back to steps (1, 2).

[0018] 3. Adaptive polishing stage: This stage determines the sample hardness based on the mechanical data from the grinding stage, enabling precise selection and automated replacement of polishing cloths. Based on the force data collected by the XYZ omnidirectional mechanical sensors during the fine grinding stage, combined with the "force data-sample hardness" correlation model pre-established in the AI ​​system, different types of polishing cloths and polishing materials are used for polishing.

[0019] This step uses the "force data - sample hardness" correlation model data that has been established in the system. The model data was collected from a large number of standard samples in the early stage and established through AI model. The model is mainly based on the XYZ mechanical data of the grinding stage and outputs the sample hardness level.

[0020] The model is implemented in two steps: hardness prediction model and polishing parameter optimization.

[0021] I. The process of establishing the XGBoost hardness prediction model: The XGBoost model can combine multiple weak learning sets into a strong learning set, which is suitable for situations like this system where the amount of standard sample collection is limited in the early stages.

[0022] (1) Use the XYZ force statistics corresponding to different standard hardness (such as HRC20, 30, 40, 50, 60, etc.) as the “calibration sample” for model training, and divide the mechanical feature threshold of hardness range from 1 to 10.

[0023] (2) Through multiple verifications using standard samples, the core features strongly related to hardness (XY force fluctuation coefficient) were screened out, invalid features were eliminated, and generalization was improved.

[0024] (3) The "mechanical characteristics-hardness grade" benchmark library constructed from standard data allows the model to achieve high prediction accuracy without a large amount of unknown sample data.

[0025] II. Polishing Parameter Optimization Based on a preset "standard hardness - optimal polishing scheme" benchmark combination as a "decision table", scheme matching and fine-tuning are completed through simple logical judgment.

[0026] (1) Compare the predicted hardness of the unknown sample (e.g., HRC 48) with the preset standard hardness (e.g., HRC 40, 50) and match the benchmark polishing scheme corresponding to the closest standard hardness (e.g., "flocked hard polishing cloth + 0.5μm polishing material + 1.8N pressure" for HRC 50).

[0027] (2) Adjust the parameters according to the hardness deviation (e.g., HRC 48 is 2 lower than HRC 50, the pressure is reduced by 0.1N, and the speed is reduced by 50rpm). The fine-tuning rules are based on the empirical threshold of the benchmark data (determined in advance through verification with standard samples).

[0028] 4. Polishing effect detection: After polishing, the high-precision line laser sensor scans the sample surface again to detect the surface roughness. If Ra≤0.02μm, the polishing is deemed qualified; if it does not meet the standard, the polishing time is extended by 0.5-2min and the test is performed again.

[0029] This step does not require an AI model; the data collected from the laser sensor can be processed and then used to make a judgment.

[0030] 5. Cleaning and Drying: This step involves alternating cleaning with deionized water and alcohol to remove residual polishing agent and impurities from the sample surface, preventing any impact on subsequent etching. After cleaning, the robotic arm transfers the sample to the drying station, where the sample surface is dried for 1-3 minutes using air at room temperature and a speed of 1-5 m / s. This ensures that no moisture remains on the sample surface.

[0031] The specific process of cleaning and drying is as follows: first, rinse the sample surface with deionized water for 10-15 seconds, then rinse with alcohol for 10-15 seconds; after cleaning, use room temperature air with a wind speed of 1-5 m / s to dry the sample surface for 1-3 minutes.

[0032] 6. AI-Assisted Corrosion Process (Core Precision Control Process): This process utilizes AI image recognition technology to automatically determine the degree of corrosion, avoiding errors associated with traditional manual assessment. It automatically selects the appropriate etching solution based on the sample material, transfers the sample to the metallographic microscope station, activates the automatic dispensing device, and uses a 10x microscope to acquire real-time corrosion images of the sample surface. These images are then transmitted to the AI ​​image analysis system. The system extracts key features of the sample surface (such as color changes, clarity, and metallographic structure proportions) using image segmentation algorithms and matches them with a preset "optimal corrosion standard image" (established based on standard metallographic spectra for different materials). When the feature matching degree is ≥90%, the corrosion is considered complete, and the rinsing device is immediately activated. The rinsing time is 10-30 seconds, terminating the corrosion process. After rinsing, the sample is immediately dried.

[0033] This step uses image processing and the MobileNetV2 model.

[0034] Most relevant literature focuses on tissue identification, tissue characteristics, and morphology. This is relatively easy to implement for samples with only a single or easily distinguishable tissue, but it is much more difficult to implement when metallographic structures are intertwined or have various deformations, requiring a large number of samples for training (tens of thousands of samples per tissue type). The core function of this step is to determine whether the etching is complete and to provide a similar metallographic structure; therefore, a lightweight model is sufficient. The implementation steps are as follows: (1) Image acquisition and processing: 3-5 images are acquired per second, and Gaussian filtering, grayscale and other algorithms are used for image processing after acquisition.

[0035] (2) Organize and separate different tissue regions.

[0036] (3) Gradient calculation and comparison to determine the clarity of corrosion: For tissue samples, the corresponding standard gradient value is called to calculate the similarity; for tissue samples, the overall standard gradient interval is called to determine the similarity.

[0037] MobileNetV2 tissue segmentation and extraction (implementation process of steps 2 and 3 above): (1) This system adopts the MobileNetV2 model and freezes 60-80% of the weights, which can reduce data dependence.

[0038] (2) Use the collected standard metallographic images to carry out classification training and learning.

[0039] (3) Analyze the real-time acquired images to achieve separation of multiple tissues from a single image.

[0040] (4) Perform gradient calculations on the separated tissues.

[0041] (5) Compare the calculated value with the standard value to determine the clarity.

[0042] (6) For a single image with multiple tissues, each tissue must reach a clarity threshold. If the gradient value of a tissue far exceeds the upper limit of the overall interval, erosion should be stopped immediately.

[0043] (7) For tissues without samples, the clarity is judged by the overall standard gradient judgment principle.

[0044] 7. Metallographic Photography and Automated Storage: This stage achieves standardized acquisition of sample images and orderly storage of samples. After etching and drying, a robotic arm transfers the sample to the metallographic photography station. The metallographic microscope automatically switches lenses according to preset magnifications (typically 100x, 200x, and 500x, covering both macroscopic and microscopic observation needs), taking pictures according to preset sample positions (one central area + four surrounding areas). 3-5 clear images are captured at each magnification. The captured metallographic images and sample information (sample number, material, sample preparation time, sample preparation parameters at each stage, and test data) are linked and stored in a database system, supporting retrieval by keywords such as sample number, material, and sample preparation date, facilitating subsequent analysis and quality traceability.

[0045] 8. Sample vacuum sealing storage: The robotic arm puts the sample into a transparent sealed bag, and the vacuum packaging machine removes the air from the bag to make the vacuum degree ≤-0.08MPa. Then the bag is heat-sealed and the sample information is printed on the surface of the sealed bag to complete the entire sample preparation process.

[0046] The advantages of this invention are: it automates the entire metallographic sample preparation process, avoids errors from traditional manual operation, and significantly improves the stability and efficiency of sample quality. Attached Figure Description

[0047] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Appendix Figure 1 A flowchart of a fully automated metallographic sample preparation process according to an embodiment of the present invention is shown.

[0048] Appendix Figure 2 The diagram shows the test results of the grinding process according to Embodiment 1 of the present invention.

[0049] Appendix Figure 3 The diagram shows photographs and analysis of the corrosion process according to Embodiment 1 of the present invention.

[0050] Appendix Figure 4 A schematic diagram of a portion of the metallographic recording photographs according to Embodiment 1 of the present invention is shown.

[0051] Appendix Figure 5 The diagram shows the test results of the grinding process according to Embodiment 2 of the present invention.

[0052] Appendix Figure 6 The diagram shows photographs and analysis of the corrosion process according to Embodiment 2 of the present invention.

[0053] Appendix Figure 7A partial metallographic recording photograph schematic diagram is shown according to Embodiment 2 of the present invention. Detailed Implementation

[0054] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0055] To further illustrate the technical solution and implementation effect of the present invention, detailed embodiments are provided below in conjunction with the sample preparation process of two typical metal materials (45# steel and 6061 aluminum alloy). The parameters and operations of each step are based on the above technical solution to ensure repeatability.

[0056] Example 1: Fully automated metallographic sample preparation of 45# steel samples 1. Sample Information The material to be prepared is 45# hot-rolled steel plate with a thickness of 20mm. A metallographic sample with a size of Φ20mm×20mm needs to be prepared.

[0057] 2. Automatic waterjet cutting Based on the hardness of 45# steel (HB190-230), the preset waterjet cutting pressure is 320MPa, the cutting speed is 18mm / s, and the cutting path is a Φ20mm circle; after cutting, it is directly transferred to the grinding station by a robotic arm.

[0058] 3. Multi-stage intelligent sample grinding Appendix Figure 2 The diagram shows the test results of the grinding process according to Embodiment 1 of the present invention. (Attached) Figure 3 The accompanying diagram shows photographs and analysis of the corrosion process according to Embodiment 1 of the present invention. Figure 4 A schematic diagram of a portion of the metallographic recording photographs according to Embodiment 1 of the present invention is shown.

[0059] (1) The XYZ omnidirectional force sensor (accuracy ±0.1N) integrated in the sample clamping head is activated. The sample is clamped and lightly touches the 180# grinding disc. The sensor collects the force fluctuation in the X direction of 0.3N, the Y direction of 0.2N, and the Z direction of 0.4N. All of them are ≤±0.5N. The contact is determined to be stable, and the sample is then ground. (2) The automatic pressure adjustment device is set to a pressure of 20N and the electric grinding machine speed is 400r / min. During the grinding process, the sensor monitors the contact force in real time. When the force in the XY direction is less than 1N, the grinding ends. (3) After the grinding is paused, the high-precision line laser sensor scans the sample surface to obtain the three-dimensional surface morphology data. The analysis shows that the maximum scratch depth is 1.8 μm, which meets the requirements of 400# grinding disc (can remove scratches ≤3 μm). It is determined that the grinding is completed at this stage.

[0060] (4) Replace the grinding disc. A fast laser scanner was used to scan the 400# grinding disc and found three damaged areas on the surface. Replace the grinding disc with a spare one. (5) Set the downward pressure to 15N and the rotation speed to 400r / min. Start grinding with 400# grinding disc. When the force in the XY direction is less than 1N, the grinding sample ends. The maximum scratch depth is 1μm obtained by line laser scanning, which meets the requirements of 600# grinding disc (can remove scratches ≤1.5μm). Grinding with 400# is completed.

[0061] (6) Inspect the 600# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 15N and the rotation speed to 400r / min. Start grinding with the 600# grinding disc. When the force in the XY direction is less than 1N, the grinding is finished. The maximum scratch depth is 0.7μm obtained by line laser scanning, which meets the requirements of the 800# grinding disc (can remove scratches ≤1μm). The 600# grinding is completed.

[0062] (7) Inspect the 800# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 15N and the rotation speed to 600r / min. Start grinding with the 800# grinding disc. When the force in the XY direction is less than 1N, the grinding is finished. The maximum scratch depth is 0.4μm obtained by line laser scanning, which meets the requirements of the 1000# grinding disc (can remove scratches ≤0.6μm). The 800# grinding is completed.

[0063] (8) Check the 1000# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 10N and the rotation speed to 800r / min. Start grinding with the 1000# grinding disc. When the force in the XY direction is less than 1N, the grinding sample ends. The maximum scratch depth is 0.4μm obtained by line laser scanning, which meets the polishing requirements.

[0064] (9) Based on the force data collected by the XYZ sensor during the grinding stage, combined with the "force data-hardness" correlation model in the AI ​​system, the sample is determined to be hard. High-density velvet polishing cloth and diamond polishing liquid are selected. The mechanical replacement mechanism fixes the polishing cloth on the grinding machine worktable to complete the replacement.

[0065] (10) Set the downward pressure to 10N, the rotation speed to 1500r / min, and the preset polishing time to 6min; after polishing, scan the surface with a line laser sensor and measure the roughness Ra=0.02μm, and judge that the polishing is qualified.

[0066] (11) First rinse the sample surface with deionized water (rinsing time 30s) to remove residual polishing agent; then rinse the sample surface with alcohol (rinsing time 10s). (12) The robotic arm transfers the sample to the drying station and dries it with room temperature air (wind speed 3m / s) for 1 minute. There is no moisture residue on the sample surface.

[0067] (13) AI-assisted corrosion: Based on the material of 45# steel, 4% nitric acid alcohol solution is automatically selected as the corrosion liquid. The sample is transferred to the stage of the metallographic microscope (magnification 10x). The automatic dripping device adds 2mL of corrosion liquid to cover the sample surface to be corroded. The microscope collects the sample surface image in real time and transmits it to the AI ​​image analysis system. The system extracts the color change and metallographic structure ratio features through the image segmentation algorithm and compares them with the preset "45# steel best corrosion standard image". After 12 seconds of corrosion, the feature matching degree is 92%≥90%, and the corrosion is determined to be in place. The deionized water rinsing device is immediately started to rinse for 20s and then dried with room temperature air (wind speed 3m / s) for 1min.

[0068] (14) The microscope automatically switches between preset magnifications of 100x, 200x and 500x, adjusts the sample position according to the preset positioning, takes pictures of 5 different areas (1 in the center and 4 around the perimeter), and obtains a total of 15 metallographic images (5 images at each magnification), with an image resolution of 2048×2048 pixels.

[0069] (15) Associate the image with the sample information (number: 45#-202512345-001, material: 45# steel, grinding parameters, polishing parameters, corrosion parameters, etc.) and store it in the local database.

[0070] (16) The robotic arm puts the sample into a transparent sealed bag, and the vacuum packaging machine removes the air from the bag and seals it; then it is transferred to the inkjet printer to print the sample number information, thus completing the sample preparation.

[0071] Example 2: Fully Automated Metallographic Sample Preparation of Industrial Pure Iron Samples 1. Sample Information The material to be prepared is industrial pure iron with a thickness of 20 mm. A metallographic sample with dimensions of Φ20 mm × 20 mm needs to be prepared.

[0072] 2. Automatic waterjet cutting Based on the hardness of industrial pure iron (HV125~185), the preset waterjet cutting pressure is 260MPa, the cutting speed is 18mm / s, and the cutting path is a Φ20mm circle; after cutting, it is directly transferred to the grinding station by a robotic arm.

[0073] 3. Multi-stage intelligent sample grinding Appendix Figure 5 The diagram shows the test results of the grinding process according to Embodiment 2 of the present invention. (Attached) Figure 6 The accompanying photographs and analysis diagrams illustrate the corrosion process according to Embodiment 2 of the present invention. Figure 7A partial metallographic recording photograph schematic diagram is shown according to Embodiment 2 of the present invention.

[0074] (1) The XYZ omnidirectional force sensor (accuracy ±0.1N) integrated in the sample clamping head is activated. The sample is clamped and lightly touches the 180# grinding disc. The sensor collects the force fluctuation in the X direction of 0.3N, the Y direction of 0.2N, and the Z direction of 0.4N. All of them are ≤±0.5N. The contact is determined to be stable, and the sample is then ground. (2) The automatic pressure adjustment device is set to a pressure of 15N and the electric grinding machine speed is 400r / min. During the grinding process, the sensor monitors the contact force in real time. When the force in the XY direction is less than 1N, the grinding ends. (3) After the grinding is paused, the high-precision line laser sensor scans the sample surface to obtain the three-dimensional surface morphology data. The analysis shows that the maximum scratch depth is 2.5 μm, which meets the requirements of 400# grinding disc (can remove scratches ≤3 μm). It is determined that the grinding is completed at this stage.

[0075] (4) Check the 400# grinding disc. Replace the grinding disc after it is qualified. Set the downward pressure to 15N and the rotation speed to 400r / min. Start grinding with the 400# grinding disc. When the force in the XY direction is less than 1N, the grinding is finished. The maximum scratch depth is 1.5μm obtained by line laser scanning, which meets the requirements of the 600# grinding disc (can remove scratches ≤1.5μm). The 400# grinding is completed.

[0076] (5) Check the 600# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 10N and the rotation speed to 400r / min. Start grinding with the 600# grinding disc. When the force in the XY direction is less than 1N, the grinding sample ends. The maximum scratch depth is 0.7μm obtained by line laser scanning, which meets the requirements of the 800# grinding disc (can remove scratches ≤1μm). The 600# grinding is completed.

[0077] (6) Inspect the 800# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 10N and the rotation speed to 600r / min. Start grinding with the 800# grinding disc. When the force in the XY direction is less than 1N, the grinding is finished. The maximum scratch depth is 0.4μm obtained by line laser scanning, which meets the requirements of the 1000# grinding disc (can remove scratches ≤0.6μm). The 800# grinding is completed.

[0078] (7) Check the 1000# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 10N and the rotation speed to 600r / min. Start grinding with the 1000# grinding disc. When the force in the XY direction is less than 1N, the grinding is finished. The maximum scratch depth is 0.4μm obtained by line laser scanning, which meets the requirements of the 1200# grinding disc (can remove scratches ≤0.4μm). The 1000# grinding is completed.

[0079] (8) Check the 1200# grinding disc. If it is qualified, replace the grinding disc. Set the downward pressure to 10N and the rotation speed to 800r / min. Start grinding with the 1000# grinding disc. When the force in the XY direction is less than 1N, the grinding sample ends. The maximum scratch depth is 0.3μm obtained by line laser scanning, which meets the polishing requirements.

[0080] (9) Based on the force data collected by the XYZ sensor during the grinding stage, combined with the "force data-hardness" correlation model in the AI ​​system, it is determined that the sample is soft. Select the navy wool polishing cloth and W2.5 diamond polishing liquid. The mechanical replacement mechanism fixes the polishing cloth on the grinding machine worktable to complete the replacement.

[0081] (10) Set the downward pressure to 10N, the rotation speed to 1500r / min, and the preset polishing time to 6min; after polishing, scan the surface with a line laser sensor and measure the roughness Ra=0.02μm, and judge that the polishing is qualified.

[0082] (11) First rinse the sample surface with deionized water (rinsing time 30s) to remove residual polishing agent; then rinse the sample surface with alcohol (rinsing time 10s). (12) The robotic arm transfers the sample to the drying station and dries it with room temperature air (wind speed 3m / s) for 1 minute. There is no moisture residue on the sample surface.

[0083] (13) AI-assisted corrosion: Based on the industrial pure iron material, 4% nitric acid alcohol solution is automatically selected as the corrosion liquid. The sample is transferred to the stage of the metallographic microscope (magnification 10x). The automatic dripping device adds 2mL of corrosion liquid to cover the sample surface to be corroded. The microscope collects the sample surface image in real time and transmits it to the AI ​​image analysis system. The system extracts the color change and metallographic structure ratio features through the image segmentation algorithm and compares them with the preset "best corrosion standard image of industrial pure iron". After 12 seconds of corrosion, the feature matching degree is 92%≥90%, and the corrosion is determined to be in place. The deionized water rinsing device is immediately started to rinse for 20s and then blown dry with room temperature air (wind speed 3m / s) for 1min.

[0084] (14) The microscope automatically switches between preset magnifications of 100x, 200x and 500x, adjusts the sample position according to the preset positioning, takes pictures of 5 different areas (1 in the center and 4 around the perimeter), and obtains a total of 15 metallographic images (5 images at each magnification), with an image resolution of 2048×2048 pixels.

[0085] (15) Associate the image with the sample information (number: DT-202554321-001, material: industrial pure iron, grinding parameters, polishing parameters, corrosion parameters, etc.) and store it in the local database.

[0086] (16) The robotic arm puts the sample into a transparent sealed bag, and the vacuum packaging machine removes the air from the bag and seals it; then it is transferred to the inkjet printer to print the sample number information, thus completing the sample preparation.

[0087] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fully automated metallographic sample preparation method, characterized in that, The steps are as follows: (1) Automatic water jet cutting: A high-pressure water jet cutting device is used to perform cold cutting on the metal workpiece to be sampled to obtain the initial metallographic sample; (2) Multi-stage intelligent grinding: An electric grinding machine equipped with an automatic grinding disc changing device is used to grind the initial metallographic sample in multiple stages. The grinding process is based on a sample clamping head with an integrated XYZ omnidirectional mechanical sensor. The contact force fluctuation is monitored to determine whether to proceed to the next grinding step. At the same time, a line laser sensor is used to scan the sample surface and collect surface scratch data to further determine whether the requirements for the next grinding step are met. The grinding parameters are adjusted in combination with an AI learning system. Before each grinding step, the surface condition of the grinding disc is detected by a fast laser scanner to ensure that the grinding disc conditions meet the requirements. (3) Adaptive polishing: Based on the data collected by the XYZ omnidirectional mechanical sensor during the grinding stage, the sample hardness is determined, and the appropriate polishing cloth type is automatically selected. After the polishing cloth is replaced by the mechanical replacement mechanism, the polishing operation is carried out. After polishing, the surface is scanned by the line laser sensor to determine whether the scratches meet the requirements of subsequent operations. (4) Cleaning and drying: Rinse the sample alternately with deionized water and alcohol, and then dry the sample with cold air; (5) AI-assisted etching: Under a metallographic microscope, select the appropriate etching solution according to the sample material, add the etching solution through an automatic dripping device, and at the same time use an AI image analysis system to collect and compare the surface tissue characteristics of the sample in real time. Use image recognition to determine whether the etching is in place. Once the standard is met, rinse immediately to stop the etching. (6) Metallographic photography and automatic storage: The corroded sample is automatically photographed using a metallographic microscope. After the image is associated with the sample information and stored, the sample is vacuum-sealed and bagged and automatically assigned a storage location.

2. The fully automated metallographic sample preparation method according to claim 1, characterized in that, In step (2), the multi-stage intelligent grinding process based on the XYZ omnidirectional mechanical sensor determines the next grinding step as follows: the initial metallographic sample is fixed on the sample clamping head that integrates the XYZ omnidirectional mechanical sensor and the automatic downward pressure adjustment device. The clamping head drives the sample to lightly touch the grinding disc of the current stage. When the fluctuation of the force data collected by the XYZ omnidirectional mechanical sensor is less than the threshold, it is determined that the current stage of grinding of the sample is over and surface scratch detection can be performed.

3. The fully automated metallographic sample preparation method according to claim 1, characterized in that, The specific process of using a line laser sensor to judge scratches in steps (2) and (3) is as follows: After the current stage of grinding is completed in step (2), the sample surface is scanned by a line laser sensor. If the maximum scratch depth on the surface is less than the grinding capacity of the next grinding disc, the current stage of grinding is determined to be completed and the grinding disc can be changed. After polishing is completed in step (3), the same line laser sensor scans the polished surface. If the surface roughness is less than or equal to the threshold, the polishing is determined to meet the requirements and the cleaning process can be allowed. Otherwise, the polishing time is extended.

4. The fully automated metallographic sample preparation method according to claim 1, characterized in that, During the grinding process, the XYZ sensor continuously monitors the changes in contact force. When the sensor reaches the threshold range, the clamping head is lifted, cleaned and dried, and then surface scratch detection is performed to obtain three-dimensional surface morphology data. The maximum scratch depth on the surface is calculated by image analysis algorithm. If it is less than the grinding capability of the next stage grinding disc, the current stage of grinding is considered complete. If the requirements are not met, the grinding time is extended and the sample is scanned and detected again until the scratch depth meets the standard.

5. The fully automated metallographic sample preparation method according to claim 1, characterized in that, Step (2) also includes AI learning and parameter optimization: inputting sample material information, XYZ omnidirectional mechanical sensor data, line laser sensor scanning data, grinding parameters and grinding effect data into the AI ​​learning system, and establishing a sample material-grinding parameter-grinding effect correlation model through neural network to automatically adjust the parameters of similar samples.

6. The fully automated metallographic sample preparation method according to claim 1, characterized in that, In the adaptive polishing step (3), the sample hardness is determined based on the mechanical data of the grinding stage, and the polishing cloth is accurately selected and automatically replaced. According to the force data collected by the XYZ omnidirectional mechanical sensor in the fine grinding stage, combined with the force data-sample hardness correlation model pre-established in the AI ​​system, different types of polishing cloth and polishing material are used for polishing.

7. The fully automated metallographic sample preparation method according to claim 6, characterized in that, The stress data-sample hardness correlation model is implemented in two steps: hardness prediction model and polishing parameter optimization. I. Establishment of the hardness prediction model: (1) Use the statistical data of XYZ force corresponding to different standard hardness as calibration samples for model training, and divide the mechanical feature thresholds of hardness range 1 to 10. (2) After multiple verifications using standard samples, the core features strongly related to hardness were selected, namely the XY force fluctuation coefficient, and invalid features were eliminated. (3) Construct a mechanical characteristic-hardness grade benchmark library based on standard data; II. Polishing Parameter Optimization Based on a preset standard hardness-optimal polishing scheme benchmark combination as a decision table, scheme matching and fine-tuning are completed through logical judgment: (1) Compare the predicted hardness of the unknown sample with the preset standard hardness, and match the benchmark polishing scheme corresponding to the closest standard hardness. (2) Adjust the parameters to control the pressure and speed according to the hardness deviation.

8. The fully automated metallographic sample preparation method according to claim 1, characterized in that, The specific process of determining whether corrosion is in place using image recognition in step (5) is as follows: a metallographic microscope with a magnification of 10x continuously acquires images of the sample surface in real time. The AI ​​image analysis system extracts the color change and tissue feature change of the sample surface through image segmentation algorithm and compares them with the preset best corrosion standard image. When the feature matching degree is greater than or equal to the threshold, it is determined that corrosion is in place, and the corrosion is terminated immediately by rinsing with alcohol several times.

9. The fully automated metallographic sample preparation method according to claim 8, characterized in that, Step (5) includes: (1) Image acquisition and processing: 3-5 images are acquired per second, and Gaussian filtering and grayscale algorithm are used for image processing after acquisition; (2) The image segmentation algorithm is used to segment tissues and separate different tissue regions; (3) Gradient calculation and comparison to determine the clarity of corrosion: For tissue samples, the corresponding standard gradient value is called to calculate the similarity; for tissue samples, the overall standard gradient interval is called to determine the similarity.

10. The fully automated metallographic sample preparation method according to claim 9, characterized in that, The tissue segmentation, gradient calculation, and comparison were performed using the MobileNetV2 model, including: Classification training is conducted using preset standard metallographic images; Analyze real-time acquired images to separate multiple tissues from a single image; Gradient calculations are performed on the separated tissues; Compare the calculated values ​​with the standard values ​​to determine the sharpness; For a single image with multiple tissues, each tissue needs to reach a sharpness threshold. If the gradient value of one tissue far exceeds the upper limit of the overall range, erosion should be stopped immediately. For tissues without samples, the overall standard gradient judgment principle is used to determine clarity.

Images