Metal additive manufacturing molten pool monitoring method and system, electronic equipment and storage medium

Abstract

The invention discloses a metal additive manufacturing molten pool monitoring method, system and equipment and a medium. The method comprises the following steps: acquiring a processing optical image sequence and real-time laser process parameters of time sequence alignment; calculating a temperature field based on parameters by using a thermophysical model, and extracting coordinate mapping meeting a threshold value to generate a space prompt point set; calling a visual segmentation large model, and respectively inputting the image and a prompt point set serving as spatial position priori into a model encoder; fusing the features to generate a molten pool segmentation mask; and performing closed-loop control based on the mask generation instruction. According to the method, the physical certainty priori is combined with the perception capability of the visual model, and high-precision segmentation and closed-loop control of the molten pool are realized under complex interference.

Description

Technical Field

[0001] This invention relates to the field of technology, and particularly to the field of technology. Background Technology

[0002] Laser powder bed fusion (LPBF), as a mainstream metal additive manufacturing technology, is widely used in aerospace, biomedicine, and other fields. During the processing, the morphology of the molten pool directly reflects the stability of the energy input and the forming quality. Therefore, achieving real-time, high-precision monitoring of the molten pool morphology is crucial for ensuring printing quality.

[0003] Existing molten pool monitoring methods primarily rely on high-speed cameras mounted on optical systems to acquire visible light or infrared images, and then extract molten pool features using image processing algorithms. Traditional image segmentation algorithms struggle to adapt to drastic lighting changes during processing. While deep learning-based visual models, which have emerged in recent years, have improved segmentation performance to some extent, these purely data-driven methods often suffer from false positives or false negatives when faced with complex conditions such as strong laser reflections, metal spatter, and plume interference due to a lack of physical constraints. Furthermore, pure visual models typically require large amounts of labeled data for training, resulting in limited generalization ability and difficulty maintaining high robustness under unfamiliar process parameters. Although finite element method (FEM) numerical simulations can provide accurate physical field information, their computational complexity is enormous and cannot meet the real-time requirements of online monitoring.

[0004] Therefore, how to effectively combine the physical deterministic prior knowledge in the processing with the powerful perception capabilities of visual models to achieve high-precision segmentation and closed-loop control of the molten pool under complex interference has become an urgent technical challenge. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, electronic device, and storage medium for monitoring the molten pool in metal additive manufacturing. The aim is to effectively combine the physical deterministic prior knowledge of the processing with the powerful perception capabilities of the visual model to achieve high-precision segmentation and closed-loop control of the molten pool under complex interference.

[0006] To achieve the above objectives, this invention proposes a method for monitoring the molten pool in metal additive manufacturing, comprising the following steps: The optical image sequence during the metal additive manufacturing process is acquired, and real-time laser process parameters that are time-aligned with the optical image sequence are obtained; the real-time laser process parameters include at least laser power, laser scanning speed, and laser focus coordinates. Using a pre-set thermophysical model, the temperature field distribution of the processing area corresponding to the optical image sequence is calculated based on the real-time laser process parameters; Extract the coordinates of the regions in the temperature field distribution whose temperature values ​​meet a preset threshold condition, and map the coordinates of the regions to the pixel coordinate system of the optical image sequence through coordinate transformation to generate a set of spatial cue points; The large visual segmentation model is invoked, the optical image sequence is input into the image encoder of the large visual segmentation model, and the spatial cue point set is input into the cue encoder of the large visual segmentation model as prior information of spatial location that can indicate the target area. Using the decoder of the visual segmentation large model, the output features of the image encoder and the output features of the cue encoder are fused to generate a segmentation mask for the molten pool region in the optical image sequence; Based on the process control instructions for generating the segmented mask, the metal additive manufacturing process is controlled in a closed loop.

[0007] Preferably, the step of acquiring an optical image sequence during the metal additive manufacturing process specifically includes: The visible light image sequence is acquired using a high-speed camera equipped with a narrowband filter; The step of obtaining real-time laser process parameters that are time-aligned with the optical image sequence specifically includes: directly reading the laser power, the laser focus coordinates, and the laser scanning speed from the central controller of the equipment via an industrial fieldbus or Ethernet protocol.

[0008] Preferably, the visible light image sequence and the real-time laser process parameters are kept in time alignment through hard synchronization or soft synchronization, and a unified timestamp is marked for each set of aligned data.

[0009] Preferably, the preset threshold condition is: The calculated temperature value in the two-dimensional temperature field of the processing area is higher than the liquidus temperature of the target material.

[0010] Preferably, the step of inputting the spatial cue point set as prior information indicating the spatial location of the target region into the cue encoder of the large visual segmentation model specifically includes: The set of spatial cue points mapped to the pixel coordinate system is encapsulated into the data structure required by the large visual segmentation model, and each coordinate point in the set of spatial cue points is identified as a positive sample cue to indicate that the image region corresponding to the coordinate point belongs to the melt pool foreground to be segmented.

[0011] Preferably, before the step of generating process control instructions based on the segmentation mask, a step of calculating the geometric feature parameters of the molten pool is further included; The steps for calculating the geometric characteristic parameters of the molten pool specifically include: The molten pool area is obtained by counting pixels, the molten pool perimeter is calculated by extracting the contour boundary pixels of the segmentation mask, and the molten pool roundness is obtained by calculating the ratio of four times pi multiplied by the square of the molten pool area and the molten pool perimeter.

[0012] Preferably, the step of generating process control instructions based on the segmentation mask specifically includes performing at least one of the following operations according to the molten pool geometric feature parameters: (1) When the area of ​​the molten pool continuously exceeds the stable threshold, the laser power is adjusted according to a preset strategy; (2) When the roundness of the molten pool is lower than the shape threshold, optimize the scanning path; (3) When the molten pool is abnormal or splits, send an emergency stop command to the machine tool control system.

[0013] 8. The method for monitoring the molten pool in metal additive manufacturing as described in claim 1, characterized in that the thermophysical model is constructed based on the principle of heat superposition; in the step of calculating the temperature field distribution of the processing area using a preset thermophysical model, the temperature field calculation formula is as follows:

[0014] In the formula, In time ,Location Temperature at that location; It is a constant related to the thermal conductivity of the material; The laser absorption coefficient; The laser power; These are the beam distribution parameters; These are the beam distribution parameters; The density of the material; These are material constants related to specific heat capacity; The contribution of the current scan pass to heat; The cumulative contribution of historical scan passes to heat; This refers to the initial temperature or ambient temperature.

[0015] Preferably, the current scan channel contributes to the heat. The calculation formula is:

[0016] In the formula, For integration variables; and These represent the current time and the start time of the current scan track, respectively. For auxiliary time parameters; These are the relative coordinates of the laser center as it moves to the current point in the current scan pass; Thermal diffusivity; The laser spot radius; Current time With integral variables The time difference between them.

[0017] Preferably, the cumulative contribution of the historical scan passes to heat... The calculation formula is:

[0018] In the formula, and These represent the start and end times of the historical scans, respectively. The number of historical scan passes to be considered; An index for historical scan tracks; In the first The relative coordinates of the laser center when it moves to the current point in each historical scan pass.

[0019] Preferably, the auxiliary time parameter The calculation formula is:

[0020] Among them, the normalized melt pool depth parameter The calculation formula is:

[0021] In the formula, The laser scanning speed; The solidus temperature of the material; This represents the specific heat capacity of the material.

[0022] This application also discloses an intelligent monitoring system for molten pools in metal additive manufacturing, including: An image acquisition module is used to acquire optical image sequences during the metal additive manufacturing process and associate them with real-time laser process parameters that are time-aligned with the optical image sequences; the real-time laser process parameters include at least laser power, laser scanning speed, and laser focus coordinates; The physical calculation module has a built-in preset thermophysical model, which is used to calculate the temperature field distribution of the processing area corresponding to the optical image sequence based on the real-time laser process parameters, extract the coordinates of the area in the temperature field distribution whose temperature value meets the preset threshold condition, and map the area coordinates to the pixel coordinate system of the optical image sequence through coordinate transformation to generate a set of spatial prompt points; The intelligent segmentation module is used to invoke a large-scale visual segmentation model. It inputs the optical image sequence into the image encoder of the large-scale visual segmentation model, and inputs the spatial cue point set as prior information indicating the spatial location of the target region into the cue encoder of the large-scale visual segmentation model. Using the decoder of the large-scale visual segmentation model, it fuses the output features of the image encoder and the output features of the cue encoder to generate a segmentation mask for the molten pool region in the optical image sequence. The control output module is used to generate process control instructions based on the segmentation mask to perform closed-loop control of the metal additive manufacturing process.

[0023] This application also discloses an electronic device including a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the method as described in any of the preceding claims.

[0024] This application also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0025] The above technical solution has the following advantages: This invention employs the aforementioned intelligent monitoring method for molten pools in metal additive manufacturing, offering the following advantages: By introducing an analytical thermophysical model, this invention calculates the temperature field using real-time process parameters and generates a spatial cue point set containing physical priors. This point set is then used as strong guiding information input into a large visual segmentation model. This method effectively solves the problem of pure visual models easily losing targets or having inaccurate segmentation edges under complex interference such as strong light and splashing, significantly improving the accuracy and robustness of molten pool segmentation. Simultaneously, leveraging the fast calculation speed of the analytical model ensures the real-time performance of the monitoring system, achieving an efficient closed loop from physical perception to process control. Attached Figure Description

[0026] The present invention will now be described in detail with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of the intelligent monitoring method for the molten pool in metal additive manufacturing as described in Embodiment 1 of the present invention; Figure 2 This is a structural block diagram of the intelligent monitoring system for the metal additive manufacturing molten pool described in Embodiment 3 of the present invention. Detailed Implementation

[0027] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0028] Example 1 like Figure 1 As shown, this embodiment provides a method for monitoring molten pools in metal additive manufacturing. This method aims to address the problems of existing visual monitoring technologies, such as difficulty in accurately segmenting the molten pool relying solely on visual features under complex interferences like strong light and splashing, and the weak generalization ability of the model due to a lack of prior physical knowledge. By introducing an analytical thermophysical model as "deterministic guidance" and combining it with a large visual model, such as the "general segmentation capability" of SAM, high-precision and robust molten pool morphology monitoring and closed-loop control are achieved.

[0029] This method mainly relies on a metal additive manufacturing monitoring system. The core hardware of the system includes a vision sensor installed on the top of the processing room, a central controller connected to the laser and galvanometer, and an industrial edge computing device that executes the core algorithm.

[0030] The specific process of this method is as follows: Step S1: Obtain the optical image sequence during the metal additive manufacturing process, and associate it with the real-time laser process parameters that are time-aligned with the optical image sequence.

[0031] In actual laser powder bed melting (LPBF) processing, the vision sensor employs a high-speed camera equipped with a narrow-band filter. The center wavelength of the narrow-band filter is selected in a band where the thermal radiation of the molten pool is strong and avoids the laser wavelength, such as the near-infrared band, to filter out laser reflection and highlight the thermal radiation signal of the molten pool. The high-speed camera continuously captures images of the processing area at a fixed frame rate, such as 1000fps or higher, generating a sequence of optical images.

[0032] Simultaneously, the system must acquire key process parameters affecting the thermal behavior of the molten pool. These real-time laser process parameters include at least laser power, laser scanning speed, and laser focus coordinates. These parameters are not estimated from images, but are read directly from the device's central controller via industrial fieldbus, such as EtherCAT or Ethernet protocols. This direct reading method ensures data accuracy and low latency.

[0033] To ensure temporal consistency between "visual data" and "physical parameters," thereby guaranteeing that the temperature field calculated by the subsequent physical model accurately corresponds to the current image frame, the visible light image sequence and the real-time laser process parameters are time-aligned using either hard or soft synchronization. In this embodiment, hard synchronization is preferred: the central controller sends a laser activation signal while simultaneously sending a trigger pulse to the high-speed camera via the I / O trigger line, ensuring that the exposure time of each image frame corresponds to a defined set of process parameter states. The system marks each set of aligned image data and process parameter data with a unified timestamp, forming a synchronized data stream of <image frame, process parameter set, timestamp>.

[0034] Step S2: Using a preset thermophysical model, calculate the temperature field distribution of the processing area corresponding to the optical image sequence based on the real-time laser process parameters.

[0035] This step is the core of achieving "physical guidance." The system has a built-in pre-set thermophysical model. Unlike traditional, time-consuming numerical simulations, such as finite element analysis, this embodiment uses a heat conduction model based on analytical solutions, which can meet the real-time requirements of online monitoring. This model quickly calculates the two-dimensional temperature field distribution of the current processing area based on the current laser power, scanning speed, position, and the material's thermophysical parameters. This temperature field distribution reflects the theoretical temperature distribution of the material surface under the current energy input.

[0036] Step S3: Extract the coordinates of the regions in the temperature field distribution whose temperature values ​​meet the preset threshold conditions, and map the coordinates of the regions to the pixel coordinate system of the optical image sequence through coordinate transformation to generate a spatial cue point set.

[0037] The temperature field calculated by the physical model contains rich information, which needs to be condensed into "cues" with clear semantics to guide the visual model. In this embodiment, the preset threshold condition is set as follows: the temperature value of the calculated point in the two-dimensional temperature field of the processing area is higher than the liquidus temperature of the target material. The liquidus temperature is the critical temperature at which the material completely melts, and the area above this temperature is theoretically a "molten pool".

[0038] The system iterates through the calculated temperature field, extracting coordinate points where the temperature is above the material's liquidus temperature. These coordinates are initially in a physical world coordinate system, such as a workpiece coordinate system with units in millimeters. To allow the visual model to understand the positions of these points, a coordinate transformation is required. The system is pre-calibrated using a calibration plate to obtain the homography matrix or transformation function from the physical coordinate system to the camera pixel coordinate system. Using this calibration relationship, the coordinates of the high-temperature region in the physical world are projected onto the image plane, resulting in a set of pixel coordinate points, i.e., a set of spatial cue points. This set of points acts like a "cursor," clearly indicating the approximate location where the molten pool should appear in the image.

[0039] Step S4: Call the visual segmentation large model, input the optical image sequence into the image encoder of the visual segmentation large model, and input the spatial cue point set as prior information of spatial location that can indicate the target area into the cue encoder of the visual segmentation large model.

[0040] This embodiment employs a Segment Anything Model (SAM) or a similar large-scale visual segmentation model architecture. This architecture includes an image encoder, a prompt encoder, and a mask decoder.

[0041] In practice, optical images, such as visible light images, are input to an image encoder to extract and embed deep visual features. Simultaneously, the spatial cue point set generated in the previous step is used as "spatial location prior information" input to the cue encoder. To conform to the model's input specifications, the spatial cue point set mapped to the pixel coordinate system is encapsulated into the data structure required by the large-scale visual segmentation model, and each coordinate point in the spatial cue point set is identified as a positive sample cue. This means telling the model, "The location of these points is the target (melt pool) I want to segment." This physical cue mechanism cleverly utilizes thermal priors to solve the problem that pure visual models are prone to losing targets or making false detections under splash interference and changes in lighting.

[0042] Step S5: Using the decoder of the visual segmentation large model, fuse the output features of the image encoder and the output features of the cue encoder to generate a segmentation mask for the molten pool region in the optical image sequence.

[0043] The mask decoder receives image features from the image encoder and cue features from the cue encoder. It uses an attention mechanism to inject cue information into the image features, thereby focusing on the region near the cue point in the feature map. Combined with visual edge information, it ultimately outputs a high-precision pixel-level segmentation mask. This mask is a binary image where regions with a value of 1 represent the melt pool and regions with a value of 0 represent the background.

[0044] Step S6: Based on the segmentation mask, generate process control instructions to perform closed-loop control of the metal additive manufacturing process.

[0045] After obtaining the segmentation mask, the system enters the feature extraction and decision-making stage. First, the geometric feature parameters of the molten pool are calculated. The specific calculation methods include: obtaining the molten pool area by counting the number of pixels with a value of 1 in the mask and combining it with the pixel equivalent; extracting the contour boundary pixels of the mask using an edge detection algorithm to calculate the molten pool perimeter; and obtaining the molten pool roundness by calculating the ratio of four times pi multiplied by the molten pool area to the square of the molten pool perimeter.

[0046] Subsequently, control commands are generated based on these geometric features. The specific strategy includes performing at least one of the following operations based on the molten pool geometric feature parameters: (1) When the area of ​​the molten pool exceeds the stable threshold continuously, for example, when the area is too large for 5 consecutive frames, indicating excessive heat accumulation, the system generates an instruction to reduce the laser power and adjusts the laser power according to a preset strategy to prevent overheating.

[0047] (2) When the roundness of the molten pool is lower than the shape threshold, for example, the molten pool is stretched, indicating that the scanning speed is too fast and causes keyhole effect or instability, an instruction to optimize the scanning path is generated or the scanning speed is finely adjusted.

[0048] (3) When the molten pool morphology is abnormal, such as sudden changes in area or splitting, an emergency stop command is immediately sent to the machine tool control system to avoid the generation of scrap.

[0049] Example 2 This embodiment, based on Embodiment 1, focuses on a detailed explanation of the "thermal physics model" involved in step S2. The accuracy of this model directly determines the quality of the prompts. To balance computational accuracy and real-time performance, this embodiment employs an analytical model based on the principle of heat superposition, considering the thermal effects of the current tread and the thermal accumulation of historical treads.

[0050] The formula for calculating the temperature field distribution in the processing area using the thermophysical model is as follows:

[0051] In the formula, In time ,Location Temperature at that location; It is a constant related to the thermal conductivity of the material; The laser absorption coefficient; The laser power; These are the beam distribution parameters; These are the beam distribution parameters; The density of the material; These are material constants related to specific heat capacity; The contribution of the current scan pass to heat; The cumulative contribution of historical scan passes to heat; This refers to the initial temperature or ambient temperature.

[0052] The core term in the formula lies in the superposition of heat source contributions. Among them, The contribution of the currently scanning laser beam to the temperature at a certain point is calculated using the following formula:

[0053] In the formula, For integration variables; and These represent the current time and the start time of the current scan track, respectively. For auxiliary time parameters; These are the relative coordinates of the laser center as it moves to the current point in the current scan pass; Thermal diffusivity; The laser spot radius; Current time With integral variables The time difference between them.

[0054] This represents the residual thermal effect (i.e., heat accumulation effect) of all previously scanned historical passes on the current temperature. The calculation formula is:

[0055] In the formula, and These represent the start and end times of the historical scans, respectively. The number of historical scan passes to be considered; An index for historical scan tracks; In the first The relative coordinates of the laser center when it moves to the current point in each historical scan pass.

[0056] To more accurately simulate the effect of the three-dimensional morphology of the molten pool on the surface temperature, an auxiliary time parameter is introduced into the above formula. and normalized melt pool depth parameter .

[0057] Auxiliary time parameters The calculation formula is:

[0058] Among them, the normalized melt pool depth parameter The calculation formula is:

[0059] In the formula, The laser scanning speed; The solidus temperature of the material; This represents the specific heat capacity of the material.

[0060] Through the rigorous mathematical model described above, this embodiment can dynamically calculate the heat-affected zone and melting zone within a millisecond timescale based on real-time power P and velocity v, thus providing extremely accurate physical priors for the visual model. For example, when the laser power suddenly increases, the temperature field range calculated by the formula will immediately expand, and the generated spatial cue point set will also expand accordingly, thereby guiding the SAM model to segment a larger region. This effectively prevents segmentation defects caused by image lag or insufficient contrast in the visual model.

[0061] Example 3 This embodiment provides an intelligent monitoring system for the molten pool in metal additive manufacturing, which is used to perform the methods described in the above embodiment.

[0062] like Figure 2 As shown, the system includes: The image acquisition module mainly consists of an industrial high-speed camera, its optical lens group, and a narrowband filter. It is configured to acquire optical image sequences during the metal additive manufacturing process and correlate them with real-time laser process parameters that are time-aligned with the optical image sequences. This module is responsible for the digital conversion of signals from the physical world.

[0063] The physics calculation module runs on a high-performance edge computing unit. It has a built-in pre-defined thermophysics model, namely the complete set of formulas from Example 2. This module is used to calculate the temperature field distribution of the processing area corresponding to the optical image sequence based on the real-time laser process parameters, and extract the coordinates of regions in the temperature field distribution whose temperature values ​​meet preset threshold conditions. Through coordinate transformation, these region coordinates are mapped to the pixel coordinate system of the optical image sequence, generating a set of spatial cue points. This module acts as a translator from the "physical world" to "semantic cues."

[0064] The intelligent segmentation module loads a pre-trained or fine-tuned Visual Segmentation Model (SAM). This module invokes the SAM by inputting the optical image sequence into its image encoder and the spatial cue point set as prior information indicating the spatial location of the target region into its cue encoder. Subsequently, using the SAM's decoder, the output features of the image encoder and the cue encoder are fused to generate a segmentation mask for the molten pool region in the optical image sequence.

[0065] The control output module is connected to the central control system of the additive manufacturing equipment. This module is used to extract geometric features based on the segmented mask and generate process control commands accordingly. Specifically, this module supports sending power adjustment or path correction commands directly to the laser and laser galvanometer system via a real-time control interface using industrial Ethernet technology to perform closed-loop control of the metal additive manufacturing process.

[0066] The human-machine interface module is configured as a user interface. This module supports receiving processing results from the intelligent segmentation module and the control output module, and displaying them visually on the screen. The displayed content includes, but is not limited to: a real-time image of the molten pool overlaid with the segmentation mask, current laser power and scanning speed, and other equipment process parameters, as well as automatically generated process parameter adjustment suggestions. This allows operators to monitor the processing status intuitively in real time and make manual interventions.

[0067] The process parameter acquisition module is configured to acquire data collected by the image acquisition module and send it to the physical calculation module.

[0068] Example 4 This embodiment provides an industrial edge computing device and a computer-readable storage medium.

[0069] The industrial edge computing device can be an industrial PC, edge computing box, or high-performance server deployed on the production site. The industrial edge computing device includes a sensing interface, a real-time control interface, memory, and a processor.

[0070] The sensing interface is configured to support receiving image data streams from a high-speed camera and synchronous process parameters from the device's central controller.

[0071] The real-time control interface supports real-time communication with external actuators, such as lasers and galvanometer controllers, via the EtherCAT bus protocol, and sends control commands.

[0072] The memory is used to store computer programs, which contain various algorithm codes and data required to execute the above-mentioned molten pool monitoring method, specifically including: the weight file of the SAM large model, the preset parameters of the thermophysical model, such as material thermophysical constants, inference logic code, and image processing library.

[0073] The processor may be a CPU, GPU, FPGA, or NPU or a combination thereof, used to run the computer program to enable the industrial edge computing device to perform the method described in Embodiment 1 or Embodiment 2.

[0074] Furthermore, this embodiment also provides a computer-readable storage medium on which a computer program is stored. This storage medium can be a non-volatile memory such as a hard disk, optical disk, USB flash drive, or flash memory. When the computer program is executed by a processor, it implements the metal additive manufacturing melt pool monitoring method as described in Embodiment 1 or Embodiment 2.

[0075] The above description is merely a specific 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 metal additive manufacturing melt pool monitoring method, characterized by, Includes the following steps: The optical image sequence during the metal additive manufacturing process is acquired, and real-time laser process parameters that are time-aligned with the optical image sequence are obtained; the real-time laser process parameters include at least laser power, laser scanning speed, and laser focus coordinates. Using a pre-set thermophysical model, the temperature field distribution of the processing area corresponding to the optical image sequence is calculated based on the real-time laser process parameters; Extract the coordinates of the regions in the temperature field distribution whose temperature values ​​meet a preset threshold condition, and map the coordinates of the regions to the pixel coordinate system of the optical image sequence through coordinate transformation to generate a set of spatial cue points; The large visual segmentation model is invoked, the optical image sequence is input into the image encoder of the large visual segmentation model, and the spatial cue point set is input into the cue encoder of the large visual segmentation model as prior information of spatial location that can indicate the target area. Using the decoder of the visual segmentation large model, the output features of the image encoder and the output features of the cue encoder are fused to generate a segmentation mask for the molten pool region in the optical image sequence; Based on the process control instructions for generating the segmented mask, the metal additive manufacturing process is controlled in a closed loop.

2. The metal additive manufacturing melt pool monitoring method of claim 1, wherein, The steps for acquiring optical image sequences during the metal additive manufacturing process specifically include: The visible light image sequence is acquired using a high-speed camera equipped with a narrow-band filter; The step of obtaining real-time laser process parameters that are time-aligned with the optical image sequence specifically includes: directly reading the laser power, the laser focus coordinates, and the laser scanning speed from the central controller of the equipment via an industrial fieldbus or Ethernet protocol.

3. The metal additive manufacturing melt pool monitoring method of claim 2, wherein, The visible light image sequence and the real-time laser process parameters are kept in time alignment through hard synchronization or soft synchronization, and a unified timestamp is marked for each set of aligned data.

4. The metal additive manufacturing melt pool monitoring method of claim 1, wherein, The preset threshold condition is: The calculated temperature value in the two-dimensional temperature field of the processing area is higher than the liquidus temperature of the target material.

5. The metal additive manufacturing melt pool monitoring method of claim 1 wherein, The step of inputting the set of spatial cue points as prior information indicating the spatial location of the target region into the cue encoder of the large visual segmentation model specifically includes: The set of spatial cue points mapped to the pixel coordinate system is encapsulated into the data structure required by the large visual segmentation model, and each coordinate point in the set of spatial cue points is identified as a positive sample cue to indicate that the image region corresponding to the coordinate point belongs to the melt pool foreground to be segmented.

6. The metal additive manufacturing melt pool monitoring method of claim 1, wherein, Before the step of generating process control instructions based on the segmented mask, the method also includes a step of calculating the geometric feature parameters of the molten pool; The steps for calculating the geometric characteristic parameters of the molten pool specifically include: The molten pool area is obtained by counting pixels, the molten pool perimeter is calculated by extracting the contour boundary pixels of the segmentation mask, and the molten pool roundness is obtained by calculating the ratio of four times pi multiplied by the square of the molten pool area and the molten pool perimeter.

7. The metal additive manufacturing melt pool monitoring method of claim 6 wherein, The step of generating process control instructions based on the segmented mask specifically includes performing at least one of the following operations according to the molten pool geometric feature parameters: (1) When the area of ​​the molten pool continuously exceeds the stable threshold, the laser power is adjusted according to a preset strategy; (2) When the roundness of the molten pool is lower than the shape threshold, optimize the scanning path; (3) When the molten pool is abnormal or splits, send an emergency stop command to the machine tool control system.

8. The metal additive manufacturing melt pool monitoring method of claim 1, wherein, The thermophysical model is constructed based on the principle of heat superposition; in the step of calculating the temperature field distribution of the processing area using the pre-set thermophysical model, the temperature field calculation formula is as follows: wherein is the temperature at time , location ; is a constant related to the thermal conductivity of the material; is the laser absorption coefficient; is the laser power; is the beam profile parameter; is the beam profile parameter; is the material density; is a material constant related to the specific heat capacity; is the contribution of the current scan pass to the heat; is the cumulative contribution of the historical scan passes to the heat; is the initial temperature or ambient temperature.

9. The method for monitoring the molten pool in metal additive manufacturing as described in claim 8, characterized in that, The contribution of the current scan pass to heat The formula for calculating this is: wherein is an integral variable; is a current time and is a start time of a current scan pass, respectively; is an auxiliary time parameter; is a relative coordinate when the laser center moves to the current point in the current scan pass; is a thermal diffusivity; is a laser spot radius; is a current time is a time difference between the current time and the integral variable 10. The method for monitoring the molten pool in metal additive manufacturing as described in claim 9, characterized in that, The cumulative contribution of the historical scan passes to heat The formula for calculating is: In the formula, represents the start time and the end time of the historical scanning pass, respectively; is the number of historical scanning passes considered; is the index of the historical scanning pass; is the relative coordinate when the laser center moves to the current point in the th historical scanning pass.​ 11. The method for monitoring the molten pool in metal additive manufacturing as described in claim 9 or 10, characterized in that, The auxiliary time parameter The calculation formula is: wherein the normalized molten pool depth parameter is calculated by the formula: wherein is the laser scan speed; is the solidus temperature of the material; is the specific heat capacity of the material.

12. A smart monitoring system for a molten pool in metal additive manufacturing, characterized in that, include: An image acquisition module is used to acquire optical image sequences during the metal additive manufacturing process and associate them with real-time laser process parameters that are time-aligned with the optical image sequences; the real-time laser process parameters include at least laser power, laser scanning speed, and laser focus coordinates; The physical calculation module has a built-in preset thermophysical model, which is used to calculate the temperature field distribution of the processing area corresponding to the optical image sequence based on the real-time laser process parameters, extract the coordinates of the area in the temperature field distribution whose temperature value meets the preset threshold condition, and map the area coordinates to the pixel coordinate system of the optical image sequence through coordinate transformation to generate a set of spatial prompt points; The intelligent segmentation module is used to call the visual segmentation large model, input the optical image sequence into the image encoder of the visual segmentation large model, and input the spatial cue point set as prior information of spatial location that can indicate the target area into the cue encoder of the visual segmentation large model. Using the decoder of the visual segmentation large model, the output features of the image encoder and the output features of the cue encoder are fused to generate a segmentation mask of the molten pool region in the optical image sequence. as well as The control output module is used to generate process control instructions based on the segmentation mask to perform closed-loop control of the metal additive manufacturing process.

13. An electronic device comprising a memory and a processor, characterized in that, The memory is used to store a computer program, and the processor runs the computer program to cause the electronic device to perform the method as described in any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 11.

Images