Metal powder abnormal particle sorting equipment based on multi-mode imaging

Abstract

The invention provides abnormal metal powder particle sorting equipment based on multi-modal imaging, which relates to the technical field of metal powder material processing and comprises a feeding system, a vacuum conveying system, a metallographic imaging system, a scanning electron microscope imaging system, a sorting execution system, a collecting system, an equipment control system, a vacuum and environment control system and a vacuum chamber, the vacuum system comprises a vacuum chamber, a conveying belt, a driving device, a vacuum maintaining system and a cleaning device, and an intelligent analysis unit is installed in the middle of the front face of the vacuum chamber. The comprehensive deep analysis of the metal powder from macroscopic to microscopic and from morphology to components is realized, and the recognition precision and coverage rate of the abnormal particles are greatly improved. In order to solve the problems of powder layer and positioning grabbing, a solution of quasi-single-layer powder laying and multi-sensor fusion positioning is provided, and the engineering feasibility is enhanced.

Description

Technical Field

[0001] This invention relates to the field of metal powder material processing technology, specifically to a metal powder abnormal particle sorting device based on multimodal imaging. Background Technology

[0002] High-performance metal powders are the cornerstone of modern advanced manufacturing processes. Their quality directly determines the mechanical properties, fatigue life, and reliability of the final parts. In additive manufacturing processes such as selective laser melting (SLM), the uniformity of the powder bed and the cleanliness of the powder itself are crucial. Abnormal particles in the powder, such as high-temperature oxides, foreign impurities (sand, fibers from raw materials or the preparation process), abnormal morphology (satellite powder, agglomerated powder), and out-of-tolerance particles, are the core causes of defects such as porosity, thermal cracks, and inclusions in printed parts. In severe cases, they can directly lead to printing failure and cause significant economic losses.

[0003] Currently, the industrial sector primarily relies on sieving and manual sampling for quality control of metal powders. Sieving can only classify particles based on physical size, and is ineffective for particles with similar particle sizes to qualified powders but abnormal composition or morphology. Manual sampling involves observation under a microscope, a method that is not only inefficient and labor-intensive, but also highly susceptible to the subjective experience and fatigue of operators, resulting in poor repeatability and a high rate of missed detections. This approach cannot meet the stringent requirements of large-scale industrial production for consistent quality and efficient quality control.

[0004] Although automated inspection equipment based on machine vision has developed in recent years, most devices still have significant shortcomings. These devices typically have a single imaging mode or only use ordinary optical cameras, making it difficult to simultaneously capture macroscopic morphology and microscopic details, and even more so, they cannot perform quantitative analysis of the chemical composition of particles, resulting in low accuracy in identifying key defects such as oxides and complex inclusions. Furthermore, while scanning electron microscopy (SEM) offers extremely high resolution and component analysis capabilities, traditional SEM sample processing and inspection processes cannot be integrated into online production lines, requiring process interruptions, sample preparation, and vacuuming, resulting in extremely low efficiency. Currently, there is a lack of an automated solution that seamlessly integrates atmospheric feeding, high-precision imaging (including metallography and SEM) under high vacuum conditions, intelligent analysis and decision-making, and precise execution of sorting throughout the entire process.

[0005] Therefore, a metal powder abnormal particle sorting device based on multimodal imaging is needed to solve the above problems. Summary of the Invention

[0006] Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a metal powder abnormal particle sorting device based on multimodal imaging, which solves the problems of existing technologies.

[0008] Technical solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: a metal powder abnormal particle sorting device based on multimodal imaging, comprising a feeding system, a vacuum conveying system, a metallographic imaging system, a scanning electron microscope imaging system, a sorting execution system, a collection system, an equipment control system, a vacuum and environmental control system, and a vacuum chamber. The vacuum system includes a vacuum chamber, a conveyor belt, a drive device, a vacuum maintenance system, and a cleaning device. An intelligent analysis unit is installed in the center of the front of the vacuum chamber. A conveyor belt, which is circular, is installed inside the vacuum chamber. The back of the conveyor belt is the return section, and the front is the feeding section. A lifting controller is installed at the top of the vacuum chamber. The bottom of the device is connected to a lifting cylinder assembly, and a rotating head is installed at the bottom of the lifting cylinder assembly. A rotating brush is installed at the bottom of the rotating head. A negative pressure vacuum cleaner is installed on the side of the lifting controller, and a negative pressure suction pipe is installed at the bottom of the negative pressure vacuum cleaner. The negative pressure suction pipe and the rotating brush are both located on the return section of the conveyor belt. The negative pressure vacuum cleaner has a built-in high-efficiency filter. The negative pressure vacuum cleaner and the rotating brush constitute a cleaning device. The cleaning device is located at the top of the return section of the conveyor belt. The drive device uses a servo motor in conjunction with a magnetohydrodynamic sealed feed device, with a speed control accuracy of ±0.05mm / s and a speed range adjustable from 0.01-0.5m / s. It also integrates a high-precision grating ruler for position feedback.

[0010] The feeding system is used for the storage, quantitative conveying, and non-destructive transition of metal powder from atmospheric to vacuum environments.

[0011] Vacuum transfer systems are used to carry and transport spread powder samples in an ultra-high vacuum environment;

[0012] Metallographic imaging system is used to acquire macroscopic morphological images of powder on a conveyor belt with a large field of view and high depth of field.

[0013] Scanning electron microscope imaging system is used for high-resolution microscopic morphology imaging, composition analysis and particle statistics of powders in a vacuum environment;

[0014] The intelligent analysis unit is used to process and analyze images and data from the metallographic imaging system and the scanning electron microscope imaging system, and to identify and classify abnormal particles through multimodal data fusion and deep learning algorithms.

[0015] The sorting and execution system is used to accurately capture and separate abnormal particles according to the instructions of the intelligent analysis unit through multi-sensor fusion positioning and negative pressure adsorption.

[0016] The collection system, connected to the main chamber via a vacuum isolation valve, is used to collect normal powder and various abnormal particles separately, and to achieve weight measurement and information traceability.

[0017] The equipment master control system is used to coordinate and control the timing actions and data interactions of various subsystems;

[0018] Vacuum and environmental control systems are used to establish and maintain the ultra-high vacuum required by the equipment and the stability of the internal temperature, humidity, and vibration environment.

[0019] Preferably, a powder storage bin is provided on one side of the top of the vacuum chamber, a vacuum lock hopper is installed at the bottom of the powder storage bin, and a vibrating feeder is installed at the bottom of the vacuum chamber near the vacuum lock hopper.

[0020] Preferably, the feeding system includes a powder storage bin, a vacuum lock hopper, a vibrating feeder, and a powder spreading device. The powder spreading device consists of a precision scraper and a leveling roller, both located at the bottom of the vibrating feeder. The height of the scraper and the pressure between the rollers can be precisely adjusted to form a quasi-single-layer powder layer with controllable thickness on the vacuum conveyor belt. The entire feeding system is located at the top of the end of the feed section of the conveyor belt.

[0021] The powder storage silo is made of low-oxygen stainless steel with a mirror-polished inner wall, has a capacity of 5-50 liters, and is protected by inert gas.

[0022] The vacuum lock hopper is located below the powder storage bin and is connected to the powder storage bin via a high-vacuum gate valve. It has vacuuming and gas filling functions to achieve a smooth transition from atmospheric to vacuum.

[0023] The vibrating feeder 8 is electromagnetically driven, with a vibration frequency that is infinitely adjustable from 10 to 100 Hz and an amplitude that is infinitely adjustable from 0.1 to 5 mm. It is used to uniformly disperse powder. The powder spreading device is located below the outlet of the vibrating feeder.

[0024] Preferably, the vacuum chamber is provided with a first station, a second station and a third station arranged sequentially from left to right, and the first station, the second station and the third station are all located at the top of the feed section of the conveyor belt.

[0025] Preferably, the metallographic imaging system is located in the first workstation, and the metallographic imaging system includes:

[0026] An infinity-corrected optical metallurgical microscope equipped with 5×, 10×, 20×, and 50× long working distance objectives, featuring motorized focusing and Z-axis stacking functions.

[0027] High-resolution color CMOS camera with a resolution of no less than 4096×2160 pixels and a frame rate of no less than 60fps;

[0028] Multi-mode LED lighting system, including ring light, coaxial light and oblique lighting, with adjustable color temperature from 3000K to 6500K and adjustable brightness from 100 to 10000 lux, supports bright field, dark field and differential interference observation modes;

[0029] The first image acquisition card supports multi-channel synchronous trigger acquisition;

[0030] The metallographic imaging system achieves automatic focusing and large-area image stitching through software control.

[0031] Preferably, the scanning electron microscope imaging system is located in the second station, and the scanning electron microscope imaging system 7 includes:

[0032] Field emission scanning electron microscope with a resolution of up to 1.0 nm @ 15 kV and an adjustable accelerating voltage of 0.5-30 kV;

[0033] Energy dispersive spectrometer with silicon drift detector, effective area ≥100mm², used for point and area scanning elemental composition analysis;

[0034] Backscattered electron detectors and secondary electron detectors;

[0035] The second image acquisition card is used to acquire microscopic morphology and component distribution maps;

[0036] The microscope tube of the field emission scanning electron microscope is directly mounted on the top of the vacuum chamber 1 via a flange, sharing the same vacuum environment with the vacuum transmission system.

[0037] Preferably, a sorting execution system is provided in the third workstation. The sorting execution system is used to accurately grasp and separate abnormal particles according to the instructions of the intelligent analysis unit through multi-sensor fusion positioning and negative pressure adsorption. The sorting execution system includes a multi-axis robotic arm and a negative pressure suction nozzle device. The bottom end of the negative pressure suction nozzle device is provided with a suction nozzle. The sorting execution system also includes a visual positioning assistance system and a motion controller. The multi-axis robotic arm has a repeatability positioning accuracy of ±5μm and adopts a full vacuum compatible design, including a vacuum servo motor and a harmonic reducer.

[0038] The negative pressure suction nozzle device is controlled by a piezoelectric ceramic high-speed valve with a response time of <1ms and a diameter of 0.1-1mm that can be quickly replaced. It is connected to an external high vacuum pump group through a vacuum feed interface. The multi-axis robotic arm will separate powders in different states to different positions, and then the negative pressure suction nozzle device at the corresponding position will suck the separated powders into the corresponding collection tank through the suction nozzle. Each axis of the robotic arm is equipped with a negative pressure suction nozzle device at the corresponding position.

[0039] The visual positioning assistance system uses a high frame rate miniature camera and laser displacement sensor integrated into the end of a multi-axis robotic arm to achieve visual servoing and gripping point compensation.

[0040] The motion controller communicates in real time with the equipment's main control system and the conveyor belt encoder. It dynamically calculates the target point based on a coordinate mapping model and coordinates the synchronous movement of the multi-axis robotic arm and the conveyor belt.

[0041] Preferably, the intelligent analysis unit includes:

[0042] Industrial computers equipped with high-performance multi-core processors;

[0043] GPU computing cards with a video memory capacity of no less than 32GB are used to accelerate deep learning inference.

[0044] The image processing module performs image preprocessing, noise reduction, segmentation, morphological operations, and feature extraction.

[0045] A multimodal data fusion algorithm module is used for spatiotemporal registration and feature-level fusion of image and composition data from metallography and SEM.

[0046] The deep learning recognition module uses a deep convolutional neural network model, trained on more than 500,000 labeled images, and can identify oxides, impurities, satellite powder, agglomerated powder, and shaped powder.

[0047] The real-time decision-making system processes metallographic images at a speed of ≥60 frames / second and performs comprehensive decision-making after processing SEM data.

[0048] The data storage and management system uses a RAID 10 array.

[0049] Preferably, the collection system includes a collection tank located on one side of the vacuum chamber. The collection tank has five groups, including normal powder collection, oxide collection, impurity collection, abnormal morphology collection, and abnormal size collection. The collection tank is equipped with a high-precision weight sensor and an RFID tagging system for automatic weighing, batch identification, and quality data traceability.

[0050] Preferably, the equipment control system adopts a distributed real-time control architecture, including a main PLC, motion control card, image processing embedded system, vacuum gauge and vacuum pump controller, and human-machine interface, realizing full-process automated control and status monitoring through EtherCAT bus. The vacuum and environmental control system includes:

[0051] The temperature and humidity control module maintains the chamber temperature at 22±0.5℃ and humidity at <10%RH through a circulating water cooling system and a Peltier temperature control device.

[0052] The vibration isolation device uses an active air-floating vibration isolation platform with an isolation frequency as low as 1.5Hz, ensuring the stability of the imaging system.

[0053] The online quality monitoring and fault diagnosis module is used to monitor the sorting effect and key equipment parameters in real time and issue early warnings.

[0054] Beneficial effects

[0055] This invention provides a metal powder anomalous particle sorting device based on multimodal imaging. It has the following beneficial effects:

[0056] By innovatively integrating metallographic and SEM systems into a single online vacuum environment, comprehensive and in-depth analysis of metal powders from macroscopic to microscopic levels, and from morphology to composition, was achieved, significantly improving the accuracy and coverage of abnormal particle identification. Addressing the challenges of powder layering and positioning, a solution combining "quasi-single-layer" powder spreading and multi-sensor fusion positioning was proposed, enhancing engineering feasibility. The fully automated design significantly improved detection and sorting efficiency, reduced labor costs and subjective errors, and provided a complete, reliable, and efficient solution for the quality control of high-end metal powders. Attached Figure Description

[0057] Figure 1 This is a diagram showing the overall internal structure of the present invention;

[0058] Figure 2 This is an overall structural diagram of the present invention;

[0059] Figure 3 This is a front view of the entire invention;

[0060] Figure 4 This is an enlarged view of section B of the present invention;

[0061] Figure 5 This is an enlarged view of point A in the present invention;

[0062] Figure 6 This is a hardware architecture diagram of the intelligent analysis unit of the present invention;

[0063] The components include: 1. Vacuum chamber; 2. Conveyor belt; 3. First station; 4. Metallographic imaging system; 5. Second station; 6. Third station; 7. Scanning electron microscope imaging system; 8. Vibrating feeder; 9. Powder spreading device; 10. Negative pressure vacuum cleaner; 11. Collection tank; 12. Lifting controller; 13. Intelligent analysis unit; 14. Powder storage bin; 15. Vacuum lock hopper; 16. Lifting cylinder assembly; 17. Rotating head; 18. Rotating brush; 19. Negative pressure suction tube; 20. Multi-axis robotic arm; 21. Negative pressure suction nozzle device; 22. Suction nozzle. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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. Specific Implementation Example 1:

[0066] like Figure 1-6 As shown, the metal powder abnormal particle sorting equipment based on multimodal imaging includes a feeding system, a vacuum conveying system, a metallographic imaging system 4, a scanning electron microscope imaging system 7, a sorting execution system, a collection system, an equipment control system, a vacuum and environmental control system, and a vacuum chamber 1. The vacuum system includes the vacuum chamber 1, a conveyor belt 2, a drive device, a vacuum maintenance system, and a cleaning device. An intelligent analysis unit 13 is installed in the center of the front of the vacuum chamber 1. The conveyor belt 2 is installed inside the vacuum chamber 1. The conveyor belt 2 is circular, with a return section on the back and a feeding section on the front. A lifting controller 12 is installed at the top of the vacuum chamber 1, and a lifting cylinder assembly 16 is connected to the bottom of the lifting controller 12. A rotating head 17 is installed at the bottom of the lifting cylinder assembly 16, and a rotating brush 18 is installed at the bottom of the rotating head 17. A negative pressure vacuum cleaner 10 is installed on the side of the lifting controller 12, and a negative pressure suction pipe 19 is installed at the bottom of the negative pressure vacuum cleaner 10. Both the negative pressure suction pipe 19 and the rotating brush 18 are located in the return section of the conveyor belt 2. The negative pressure vacuum cleaner 10 has a built-in high-efficiency filter. The negative pressure vacuum cleaner 10 and the rotating brush 18 constitute a cleaning device. The cleaning device is located at the top of the return section of the conveyor belt 2. The drive device adopts a servo motor in conjunction with a magnetic fluid sealed feed device. The speed control accuracy is ±0.05mm / s, and the speed range is adjustable from 0.01 to 0.5m / s. A high-precision grating ruler is integrated for position feedback.

[0067] The feeding system is used for the storage, quantitative conveying, and non-destructive transition of metal powder from atmospheric to vacuum environments.

[0068] Vacuum transfer systems are used to carry and transport spread powder samples in an ultra-high vacuum environment;

[0069] Metallographic imaging system 4 is used to acquire macroscopic morphological images of powder on conveyor belt 2 with a large field of view and high depth of field.

[0070] Scanning electron microscope imaging system 7 is used for high-resolution microscopic morphology imaging, composition analysis and particle statistics of powder in a vacuum environment;

[0071] The intelligent analysis unit 13 is used to process and analyze images and data from the metallographic imaging system 4 and the scanning electron microscope imaging system 7, and to identify and classify abnormal particles through multimodal data fusion and deep learning algorithms.

[0072] The sorting execution system is used to accurately capture and separate abnormal particles according to the instructions of the intelligent analysis unit 13 through multi-sensor fusion positioning and negative pressure adsorption.

[0073] The collection system, connected to the main chamber via a vacuum isolation valve, is used to collect normal powder and various abnormal particles separately, and to achieve weight measurement and information traceability.

[0074] The equipment master control system is used to coordinate and control the timing actions and data interactions of various subsystems;

[0075] Vacuum and environmental control systems are used to establish and maintain the ultra-high vacuum required by the equipment and the stability of the internal temperature, humidity, and vibration environment.

[0076] A powder storage bin 14 is provided on one side of the top of the vacuum chamber 1. A vacuum lock hopper 15 is installed at the bottom of the powder storage bin 14. A vibrating feeder 8 is installed in the vacuum chamber 1 near the bottom of the vacuum lock hopper 15.

[0077] The feeding system includes a powder storage bin 14, a vacuum lock hopper 15, a vibrating feeder 8, and a powder spreading device 9. The powder spreading device 9 consists of a precision scraper and a leveling roller, both located at the bottom of the vibrating feeder 8. The scraper height and roller pressure can be precisely adjusted to form a quasi-single-layer powder layer with controllable thickness on the vacuum conveyor belt. The entire feeding system is located at the top of the feeding section of the conveyor belt 2.

[0078] The powder storage chamber 14 is made of low-oxygen stainless steel with mirror-polished inner walls, has a capacity of 5-50 liters, and is filled with inert gas for protection.

[0079] The vacuum lock hopper 15 is located below the powder storage chamber 14 and is connected to the powder storage chamber 14 through a high vacuum gate valve. It has the functions of vacuuming and filling, realizing a smooth transition from atmospheric to vacuum.

[0080] The vibrating feeder 8 is electromagnetically driven, with a vibration frequency that is infinitely adjustable from 10 to 100 Hz and an amplitude that is infinitely adjustable from 0.1 to 5 mm. It is used to uniformly disperse powder. The powder spreading device 9 is located below the outlet of the vibrating feeder 8.

[0081] The vacuum chamber 1 is equipped with a first station 3, a second station 5 and a third station 6, arranged sequentially from left to right. The first station 3, the second station 5 and the third station 6 are all located at the top of the feeding section of the conveyor belt 2.

[0082] Metallographic imaging system 4 is located in the first station 3, and metallographic imaging system 4 includes:

[0083] An infinity-corrected optical metallurgical microscope equipped with 5×, 10×, 20×, and 50× long working distance objectives, featuring motorized focusing and Z-axis stacking functions.

[0084] High-resolution color CMOS camera with a resolution of no less than 4096×2160 pixels and a frame rate of no less than 60fps;

[0085] Multi-mode LED lighting system, including ring light, coaxial light and oblique lighting, with adjustable color temperature from 3000K to 6500K and adjustable brightness from 100 to 10000 lux, supports bright field, dark field and differential interference observation modes;

[0086] The first image acquisition card supports multi-channel synchronous trigger acquisition;

[0087] Metallographic imaging system 4 achieves automatic focusing and large-area image stitching through software control.

[0088] The scanning electron microscope imaging system 7 is located in the second station 5, and the scanning electron microscope imaging system 7 includes:

[0089] Field emission scanning electron microscope with a resolution of up to 1.0 nm @ 15 kV and an adjustable accelerating voltage of 0.5-30 kV;

[0090] Energy dispersive spectrometer with silicon drift detector, effective area ≥100mm², used for point and area scanning elemental composition analysis;

[0091] Backscattered electron detectors and secondary electron detectors;

[0092] The second image acquisition card is used to acquire microscopic morphology and component distribution maps;

[0093] The microscope tube of the field emission scanning electron microscope is directly mounted on the top of the vacuum chamber 1 via a flange, sharing the same vacuum environment with the vacuum delivery system.

[0094] The third station 6 is equipped with a sorting execution system. The sorting execution system is used to accurately grasp and separate abnormal particles according to the instructions of the intelligent analysis unit through multi-sensor fusion positioning and negative pressure adsorption. The sorting execution system includes a multi-axis robotic arm 20 and a negative pressure suction nozzle device 21. The bottom end of the negative pressure suction nozzle device 21 is equipped with a suction nozzle 22. The sorting execution system also includes a vision positioning assistance system and a motion controller. The multi-axis robotic arm 20 has a repeatability positioning accuracy of ±5μm and adopts a full vacuum compatible design, including a vacuum servo motor and a harmonic reducer.

[0095] The negative pressure suction nozzle device 21 is controlled by a piezoelectric ceramic high-speed valve with a response time of <1ms and a hole diameter of 0.1-1mm that can be quickly replaced. It is connected to an external high vacuum pump group through a vacuum feed interface. The multi-axis robotic arm 20 will separate powders in different states to different positions. Then, the negative pressure suction nozzle device 21 at the corresponding position sucks the separated powders into the corresponding collection tank 11 through the suction nozzle 22. Each axis of the robotic arm 20 is equipped with a negative pressure suction nozzle device 21 at the corresponding position.

[0096] The visual positioning assistance system uses a high frame rate miniature camera and laser displacement sensor integrated into the end of the multi-axis robotic arm 20 to achieve visual servoing and gripping point compensation.

[0097] The motion controller communicates in real time with the equipment's main control system and the encoder of conveyor belt 2. It dynamically calculates the target point based on the coordinate mapping model and coordinates the synchronous movement of the multi-axis robotic arm 20 and conveyor belt 2.

[0098] The intelligent analysis unit 13 includes:

[0099] Industrial computers equipped with high-performance multi-core processors;

[0100] GPU computing cards with a video memory capacity of no less than 32GB are used to accelerate deep learning inference.

[0101] The image processing module performs image preprocessing, noise reduction, segmentation, morphological operations, and feature extraction.

[0102] A multimodal data fusion algorithm module is used for spatiotemporal registration and feature-level fusion of image and composition data from metallography and SEM.

[0103] The deep learning recognition module uses a deep convolutional neural network model, trained on more than 500,000 labeled images, and can identify oxides, impurities, satellite powder, agglomerated powder, and shaped powder.

[0104] The real-time decision-making system processes metallographic images at a speed of ≥60 frames / second and performs comprehensive decision-making after processing SEM data.

[0105] The data storage and management system uses a RAID 10 array.

[0106] The collection system includes a collection tank 11, which is located on one side of the vacuum chamber 1. The collection tank 11 has five groups, including normal powder collection, oxide collection, impurity collection, abnormal morphology collection and abnormal size collection. The collection tank 1 is equipped with a high-precision weight sensor and RFID tagging system for automatic weighing, batch identification and quality data traceability.

[0107] The equipment's central control system adopts a distributed real-time control architecture, including a main PLC, motion control card, image processing embedded system, vacuum gauge and vacuum pump controller, and human-machine interface. It achieves full-process automated control and status monitoring via EtherCAT bus. The vacuum and environmental control system includes:

[0108] The temperature and humidity control module maintains the chamber temperature at 22±0.5℃ and humidity at <10%RH through a circulating water cooling system and a Peltier temperature control device.

[0109] The vibration isolation device uses an active air-floating vibration isolation platform with an isolation frequency as low as 1.5Hz, ensuring the stability of the imaging system.

[0110] The online quality monitoring and fault diagnosis module is used to monitor the sorting effect and key equipment parameters in real time and issue early warnings.

[0111] In this technical solution, the feeding system is used to store, quantitatively transport, and seamlessly transition metal powder from an atmospheric environment to a high vacuum environment. Its core lies in the vacuum lock hopper 15 structure, which, through the sequential opening and closing of a high-vacuum gate valve, achieves continuous and automatic powder replenishment without disrupting the vacuum level of the main chamber, thus solving the technical bottleneck of continuous feeding in vacuum equipment. The powder spreading device 9, through the cooperation of a precision scraper and a leveling roller, forms a quasi-single-layer powder layer with controllable thickness, ensuring both imaging quality and processing efficiency.

[0112] The vacuum conveying system forms the main framework and operational basis of the equipment. It consists of a sealed stainless steel vacuum chamber 1 and a ring conveyor belt 2 running within it. This conveying system not only needs to transport powder smoothly and precisely, but its chamber also provides the necessary high-vacuum environment for high-resolution observation by metallurgical microscopes and normal operation of scanning electron microscopes. The drive unit uses magnetohydrodynamic sealed transmission to ensure that the vacuum is not disrupted during power transmission. An integrated high-precision grating ruler provides a unified positional reference for the entire system.

[0113] Metallurgical imaging system 4 employs a metallurgical microscope with an infinity optical system, coupled with a high-resolution color camera and a multi-mode LED illumination system, to rapidly scan the powder on conveyor belt 2. Its main task is to acquire macroscopic information about the powder, including the precise shape of the particles (for calculating sphericity, identifying satellite powders and aggregated powders), color (for preliminary judgment of oxidation discoloration), and size distribution, completing the first round of rapid screening and preliminary classification. This system features autofocus and Z-axis stacking capabilities, which can compensate for some of the undulations in the powder layer, acquiring clear images of more particles.

[0114] The scanning electron microscope imaging system 7 shares the same high-vacuum environment with the vacuum transport system, enabling true "in-situ" online detection. Utilizing a high-resolution electron beam from a field emission electron gun, it can clearly reveal the microscopic surface morphology of powder particles. More importantly, its equipped backscattered electron detector provides compositional contrast images, and the integrated energy dispersive spectrometer can perform targeted elemental composition analysis on suspicious points, thus providing decisive evidence for identifying particles with abnormal chemical composition, such as oxides and ceramic inclusions. Its analytical targets are primarily derived from suspected areas identified by the metallographic imaging system 4, implementing "precision targeting" to improve overall efficiency.

[0115] The intelligent analysis unit 13 is the "brain" of the device. It receives and processes image and composition data from both metallographic and SEM systems. Through image preprocessing, segmentation, and feature extraction algorithms, it accurately identifies each particle. Its core innovation lies in a multimodal data fusion algorithm, which correlates and fuses the macroscopic morphological features, color features, microscopic texture features, and composition features of each particle to construct its complete digital profile. Finally, a high-precision model trained on a deep convolutional neural network intelligently identifies and classifies the fused multidimensional features, outputting the type of abnormal particles and their precise coordinates on the conveyor belt.

[0116] The sorting execution system, based on instructions from the intelligent analysis unit 13 and the real-time encoder position of the conveyor belt 2, dynamically calculates the target point of the robotic arm using a coordinate mapping model. The vision assistance system at the end of the robotic arm performs final gripping point correction. The negative pressure suction nozzle, controlled by a high-speed solenoid valve, generates instantaneous negative pressure to precisely pick up identified abnormal particles and transfer them to the corresponding collection tank. The entire sorting process is completed in a vacuum environment, avoiding external contamination.

[0117] The collection system is connected to the bottom of the main chamber via a vacuum gate valve and includes a normal powder collection container and multiple collection containers for classifying abnormal particles. Each collection container is equipped with a high-precision weight sensor and RFID tag for automatic weighing, identification, and quality traceability, providing quantitative data support for powder quality.

[0118] The overall equipment control system and the vacuum and environmental control system serve as the nerve center and life support system of the equipment. They are responsible for coordinating the timing of all subsystems, maintaining high vacuum in the chamber, controlling internal temperature and humidity, suppressing vibration, and ensuring the safety and stability of equipment operation. Specific Implementation Example 2:

[0120] like Figure 1-6 As shown, the following is a further supplement to this technical solution:

[0121] The powder spreading device 9 includes a main scraper and an auxiliary leveling roller. The main scraper is made of hard alloy material and its gap with the upper surface of the conveyor belt 2 is precisely adjustable from 20 to 500 μm through double wedge blocks and a stepper motor (minimum step distance 1 μm). The auxiliary leveling roller is a silicone-coated roller and its pressure with the surface of the conveyor belt 2 is adjustable from 0.5 to 5 N through a precision pressure reducing valve and a cylinder. The two work together to stabilize the powder layer thickness at 50-200 μm and the thickness uniformity within ±15 μm, achieving true quasi-single-layer spreading and ensuring that each powder particle is exposed to the imaging system's field of view without obstruction.

[0122] The vacuum lock hopper has a capacity of 8 liters and uses two pneumatic high-vacuum gate valves (DN100) to switch on and off alternately. Under the premise that the vacuum of the main vacuum chamber 1 is not broken, it can automatically replenish powder every 4-6 minutes, with a single replenishment amount of 0.8-2kg, which can fully meet the requirement of continuous operation for more than 8 hours without manual intervention.

[0123] Conveyor belt 2 adopts a continuous low-speed operation + dynamic coordinate mapping working mode, with a stepless adjustable speed of 0.01-0.1m / s; the drive unit uses a 60kW AC servo motor and harmonic reducer mounted on the atmospheric side, and feeds through a magnetohydrodynamic sealed rotary feeder (leakage rate ≤1×10). -11 The power is transmitted into the cavity at a speed of Pa·L / s, and the active roller has a shaft diameter of 80mm. A high-precision grating ruler (resolution 0.1μm) is attached to the edge of the conveyor belt throughout the entire process, providing a unified global coordinate reference for metallographic imaging, SEM positioning, and robotic arm grasping, with a position repeatability accuracy of ±5μm.

[0124] The cleaning device is located directly above the return section of conveyor belt 2. It consists of a conductive PEEK rotating brush 18 (adjustable speed 200-800rpm, brush bristle-to-belt gap 0.2-0.5mm) controlled by the lifting cylinder assembly 16 and a negative pressure vacuum cleaner 10 (vacuuming speed ≥600L / min). The negative pressure vacuum cleaner 10 has a built-in two-stage HEPA (99.997%@0.3μm) + ULPA (99.9995%@0.12μm) filter to ensure that the residual powder on the belt surface in the return section is ≤1 particle / cm². The rotating brush 18 is raised and lowered by the lifting cylinder assembly 16 and rotated by the rotating head 17, which is internally driven by a motor.

[0125] The third workstation 6 is equipped with a sorting execution system, preferably consisting of four vacuum-compatible six-axis robotic arms 20 (preferably Staubli RX160VAC or Mitsubishi RH-6FRH-DV). The base is fixed to the top plate of the vacuum chamber 1 via flanges. The working radius of a single arm is ≥1100mm, the repeatability is ±5μm, and the end-effector load is ≤3kg. The entire machine uses vacuum-specific lubricating grease and harmonic reducers. Each robotic arm 20 integrates a negative pressure suction nozzle device 21 (with four quick-change nozzle diameters: 0.1 / 0.3 / 0.6 / 1.0mm), a piezoelectric ceramic high-speed valve (response time ≤0.8ms), a high frame rate miniature camera (≥500fps), and a laser triangulation rangefinder (resolution 0.5μm) at its end.

[0126] Four robotic arms are arranged symmetrically in pairs along the width of the conveyor belt, working together to cover a 500mm powder-spreading width, achieving redundant and high-speed parallel grasping. It should be noted that this solution does not limit the specific number of robotic arms.

[0127] The sorting process is as follows: the intelligent analysis unit 13 outputs the global coordinates (X, Y), category, and confidence level of abnormal particles in real time → the motion controller dynamically maps the coordinates according to the position of the grating ruler → the corresponding robotic arm 20 directly grabs the abnormal particle during the conveyor belt operation → the suction nozzle 22 instantly generates a negative pressure of -80kPa to suck up the single abnormal particle → the high-speed valve is closed to release the negative pressure, allowing the abnormal particle to fall freely into the corresponding collection tank 11 through the pipe. Normal powder is not touched by any robotic arm and falls directly into the normal powder collection tank at the end of the conveyor belt 2.

[0128] Each collection tank in the collection system has a volume of 5 liters and is controlled by an independent pneumatic high-vacuum gate valve (DN50, leakage rate ≤1×10⁻⁶). -9 The pressure sensor (Pa·L / s) is connected to the corresponding discharge port at the bottom of the chamber; the bottom of the tank integrates a high-precision weighing sensor (resolution 0.01g, range 10kg) and an RFID electronic tag to achieve automatic weight accumulation, batch identification, and full-process data traceability. When removing the tank, only the corresponding gate valve is closed and the tank is independently purged with argon gas, without affecting the vacuum of the main chamber.

[0129] The specific data flow and algorithm process of the intelligent analysis unit 13 are as follows:

[0130] The metallographic imaging system 4 outputs ≥60 frames of 4096×2160 stitched full-frame images per second, with simultaneous grating ruler position coding and timestamp;

[0131] The image processing module first performs flat field correction and multi-frame fusion noise reduction, and then uses an improved Mask R-CNN (backbone ResNet101-FPN) instance segmentation algorithm to obtain the pixel-level mask, minimum bounding moment, area, circularity, color HSV histogram, and surface roughness estimate of each particle, totaling 18 macroscopic features.

[0132] Particles with a roundness ≤0.88 or abnormal color (too yellow / too black) are automatically marked as suspected abnormalities, and their global coordinates (X,Y) are pushed to the SEM control system.

[0133] The scanning electron microscope imaging system 7 adopts a step-scan hybrid mode to perform 1000-8000× secondary electron imaging + backscatter imaging + point / area energy dispersive spectroscopy analysis on suspected particles one by one, and outputs 24-dimensional micro-texture features (GLCM contrast, entropy, etc.) and 12-dimensional elemental atomic percentages (O, N, C, Al, Si, Ca, etc.).

[0134] The multimodal fusion module performs early fusion (concatting into 54 dimensions) of the macroscopic 18-dimensional vector and the microscopic 36-dimensional vector of the same particle, inputs it into the trained ResNet50+Transformer classifier (training dataset of more than 800,000 labeled real industrial powders, with 5-fold cross-validation accuracy of 99.37%), and finally outputs 6 categories of labels (normal / oxide / ceramic impurity / satellite powder / morphological abnormality / size abnormality) and corresponding confidence scores;

[0135] The real-time decision-making system only pushes the coordinates, category, and optimal nozzle aperture of abnormal particles with a confidence level ≥ 0.95 to the motion controller, with an overall decision-making cycle of ≤ 80ms. Specific Implementation Example 3:

[0137] like Figure 1-6 As shown, this is used for the detection and sorting of titanium alloy (Ti-6Al-4V) powder.

[0138] This embodiment uses Ti-6Al-4V titanium alloy powder as the processing object. This material is extremely sensitive to oxygen content, and oxides are one of its main defects.

[0139] The equipment parameters are set as follows: the working vacuum degree of vacuum chamber 1 is maintained at 5×10⁻⁶. -3Pa. The powder spreading device sets the scraper gap to 1.5 times the average powder particle size (approximately 75 μm), forming a quasi-monolayer powder. Conveyor belt 2 moves in a stepping manner, stopping for 3 seconds at each imaging station to complete image acquisition. Metallurgical imaging system 4 uses a 20× objective lens and ring illumination, with depth-of-field extension enabled, focusing on particle sphericity (set threshold 0.88) and areas of color aberration. Scanning electron microscopy imaging system 7 uses an accelerating voltage of 15 kV, prioritizing the use of a backscattered electron (BSE) detector, as oxides (such as TiO2) exhibit a dark gray contrast in BSE images, making them easy to identify. EDS point analysis is initiated for suspicious particles; if the oxygen (O) atomic percentage exceeds 25%, it is confirmed as an oxide particle. The algorithm model of intelligent analysis unit 13 makes a judgment based on this. The sorting execution system uses a 0.3 mm aperture suction nozzle, with a negative pressure set to -80 kPa. The robotic arm performs final positioning based on encoder signals and visual servo to ensure reliable suction of target oxide particles.

[0140] The process is as follows: After the powder is evenly spread by the feeding system, it is first scanned by the metallographic imaging system 4 to mark suspected areas with poor sphericity and dark color. The conveyor belt 2 then delivers the powder to the SEM station. The SEM first performs BSE imaging verification on the suspected areas, confirming the presence of low-contrast particles, and then performs EDS composition analysis, confirming that they are oxides with high oxygen content. The intelligent analysis unit 13 integrates the two reports and sends instructions and target coordinates to the sorting robot arm. The robot arm, combined with the real-time position of the conveyor belt 2, moves above the target, and the end-effector camera performs final positioning compensation, subsequently precisely adsorbing and removing the oxide particles. Specific Implementation Example 4:

[0142] like Figure 1-6 As shown, this is used for the detection and sorting of nickel-based superalloy (Inconel 718) powder.

[0143] This embodiment focuses on Inconel 718 powder, specifically detecting potential ceramic inclusions (such as Al2O3).

[0144] The equipment was adjusted as follows: the powder layer thickness was slightly increased to twice the average particle size. The SEM imaging system needed to optimize image contrast to better distinguish Al2O3 (low atomic number, very dark) from the matrix. EDS analysis became the key discrimination method. The algorithm of the intelligent analysis unit 13 increased its sensitivity to aluminum (Al) and silicon (Si) elements, and the rule was set as follows: if a particle is detected to be rich in Al and O elements and almost free of matrix main elements such as Ni, Cr, and Fe, it is determined to be an Al2O3 foreign inclusion. The sorting system was therefore replaced with a nozzle with a smaller aperture (0.2 mm) and higher negative pressure (-95 kPa) to deal with potentially smaller inclusions.

[0145] The working process is as follows: Metallographic imaging system 4 detects particles with abnormal morphology (sharp edges). SEM observes extremely dark particles in BSE mode, and EDS point analysis shows that their main components are Al and O. Based on this, intelligent analysis unit 13 determines that they are Al2O3 ceramic inclusions and instructs the sorting system to sort them into the "impurity" collection tank 11, thus achieving precise removal of particles with abnormal composition.

[0146] The above embodiments demonstrate that the device of the present invention, by flexibly adjusting imaging parameters, analysis algorithms and execution components, possesses strong adaptability and precise sorting capabilities, and can meet the quality control requirements of various metal powders and various defect types.

[0147] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. Metal powder abnormal particle sorting equipment based on multi-modal imaging, comprising a feed system, a vacuum conveying system, a metallographic imaging system (4), a scanning electron microscope imaging system (7), a sorting execution system, a collection system, an equipment control system, a vacuum and environment control system and a vacuum chamber (1), characterized in that: The vacuum system comprises a vacuum chamber (1), a conveying belt (2), a driving device, a vacuum maintaining system and a cleaning device, the middle part of the front face of the vacuum chamber (1) is provided with an intelligent analysis unit (13), the vacuum chamber (1) is provided with the conveying belt (2), the conveying belt (2) is annular, the back face of the conveying belt (2) is a return section, and the front face is a feeding section, the top of the vacuum chamber (1) is provided with a lifting controller (12), the bottom end of the lifting controller (12) is connected with a lifting cylinder assembly (16), the bottom end of the lifting cylinder assembly (16) is provided with a rotary head (17), the bottom of the rotary head (17) is provided with a rotary brush (18), the side of the lifting controller (12) is provided with a negative pressure dust collector (10), the bottom end of the negative pressure dust collector (10) is provided with a negative pressure suction pipe (19), the negative pressure suction pipe (19) and the rotary brush (18) are located on the return section of the conveying belt (2), the negative pressure dust collector (10) is provided with a high-efficiency filter, the negative pressure dust collector (10) and the rotary brush (18) constitute the cleaning device, the cleaning device is located at the top position of the return section of the conveying belt (2), the driving device adopts a servo motor cooperating with a magnetic fluid sealing feeding device, the speed control precision is ±0.05 mm / s, the speed range is 0.01-0.5 m / s and is adjustable, and a high-precision grating ruler is integrated for position feedback.

2. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 1, wherein: The top of one side of the vacuum chamber (1) is provided with a powder storage bin (14), the bottom end of the powder storage bin (14) is provided with a vacuum lock hopper (15), and the bottom of the vacuum chamber (1) close to the vacuum lock hopper (15) is provided with a vibrating feeder (8).

3. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 2, wherein: The feeding system comprises the powder storage bin (14), the vacuum lock hopper (15), the vibrating feeder (8) and a powder laying device (9), the powder laying device (9) is a precision scraper and a leveling roller, and is arranged at the bottom position of the vibrating feeder (8), the height of the scraper and the pressure between the rollers can be accurately adjusted, so as to form a quasi-monolayer powder layer with controllable thickness on the vacuum conveying belt, and the feeding system is located at the top position of the end of the feeding section of the conveying belt (2).

4. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 1, wherein: The vacuum chamber (1) is provided with a first station (3), a second station (5) and a third station (6) which are sequentially arranged from left to right, and the first station (3), the second station (5) and the third station (6) are located at the top position of the feeding section of the conveying belt (2).

5. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 4, wherein: The metallographic imaging system (4) is located in the first station (3), and the metallographic imaging system (4) comprises: an infinite distance correction optical metallographic microscope which is provided with 5×, 10×, 20× and 50× long working distance objectives and has electric focusing and Z-axis stacking functions; a high-resolution color CMOS camera with a resolution of not less than 4096×2160 pixels and a frame rate of not less than 60 fps; a multi-mode LED lighting system which comprises annular light, coaxial light and oblique illumination, has adjustable color temperature of 3000K-6500K and adjustable brightness of 100-10000 lux, and supports bright field, dark field and differential interference observation modes; The first image acquisition card supports multi-channel synchronous trigger acquisition.

6. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 4, wherein: The scanning electron microscope imaging system (7) is located in the second station (5), and the scanning electron microscope imaging system (7) comprises: A field emission scanning electron microscope with an acceleration voltage of 0.5-30 kV and adjustable; A spectrometer with a silicon drift detector and an effective area of greater than or equal to 100 mm2, used for point and area scanning element composition analysis; A backscattered electron detector and a secondary electron detector; A second image acquisition card for collecting microscopic morphology and composition distribution maps; The lens barrel of the field emission scanning electron microscope is directly installed on the top of the vacuum chamber (1) through a flange, and shares the same vacuum environment with the vacuum conveying system.

7. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 4, wherein: The third station (6) is provided with a sorting execution system, the sorting execution system comprises a multi-axis mechanical arm (20) and a negative pressure suction nozzle device (21), the bottom end of the negative pressure suction nozzle device (21) is provided with a suction nozzle (22), the sorting execution system simultaneously comprises a visual positioning auxiliary system and a motion controller, the multi-axis mechanical arm (20) has a repeated positioning accuracy of ±5μm, adopts a full vacuum compatible design, and comprises a vacuum servo motor and a harmonic reducer, the nozzle head of the suction nozzle (22) is installed at the end position of the multi-axis mechanical arm (20), and a pipeline is connected to the end of the suction nozzle (22). The negative pressure suction nozzle device (21) is controlled by a piezoelectric ceramic high-speed valve, has a response time of less than 1ms, a pore diameter of 0.1-1mm, and can be quickly replaced, and is connected to an external high vacuum pump set through a vacuum feedthrough interface; The visual positioning auxiliary system adopts a high-frame-rate miniature camera and a laser displacement sensor integrated at the end of the multi-axis mechanical arm (20), and realizes visual servoing and grabbing point compensation; The motion controller communicates with a device general control system and a conveying belt (2) encoder in real time, dynamically calculates a grabbing target point based on a coordinate mapping model, and coordinates synchronous motion of the multi-axis mechanical arm (20) and the conveying belt (2).

8. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 1, wherein: The intelligent analysis unit (13) comprises: An industrial computer loaded with a high-performance multi-core processor; A GPU computing card with a display memory capacity of not less than 32GB, used for accelerating deep learning inference; An image processing module for performing image preprocessing, noise reduction, segmentation, morphological operation and feature extraction; A multi-modal data fusion algorithm module for spatio-temporal registration and feature-level fusion of images and composition data from metallography and SEM; A deep learning recognition module adopting a deep convolutional neural network model, trained based on more than 500,000 labeled images, capable of recognizing oxides, impurities, satellite powder, adhered powder and shaped powder; A real-time decision system for processing metallographic images at a speed of greater than or equal to 60 frames per second, and making comprehensive decisions after processing SEM data; A data storage and management system adopting a RAID 10 array.

9. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 1, wherein: The collection system comprises a collection tank (11), the collection tank (11) is located on one side of the vacuum chamber (1), and the collection tank (11) has five groups including normal powder collection, oxide collection, impurity collection, morphology abnormality collection and size abnormality collection. The collection tank (11) is provided with a high-precision weight sensor and an RFID identification system for automatic weighing, batch identification and quality data tracing.

10. The multi-modal imaging based metal powder anomaly particle sorting apparatus of claim 1, wherein: The device general control system adopts a distributed real-time control architecture, including a main PLC, a motion control card, an image processing embedded system, a vacuum gauge and vacuum pump controller, and a human-machine interface, and realizes full-process automatic control and state monitoring through an EtherCAT bus. The vacuum and environment control system includes: A temperature and humidity control module, which maintains the temperature in the chamber at 22±0.5℃ and the humidity at <10%RH through a circulating water cooling system and a Peltier temperature control device; A vibration isolation device, which adopts an active air-floating vibration isolation platform, has an isolation frequency as low as 1.5Hz, and ensures the stability of the imaging system; An online quality monitoring and fault diagnosis module, which is used for real-time monitoring of the sorting effect and key parameters of the equipment and early warning.

Images