Method, system and device for synergistically regulating surface quality of inner flow channel and storage medium

Abstract

The present application belongs to the field of additive manufacturing, and particularly relates to a kind of inner runner surface quality synergic control method, system, equipment and storage medium, method includes: obtaining training part three-dimensional model and extracting inner runner geometric characteristic parameter, constructs inner runner digital twin model;According to flow field characteristics, roughness prediction result and geometric feature, inner runner is zoned, and printing parameter and polishing process are matched;In SLM printing process, based on molten pool state data and profile deviation determines local remelting parameter and scanning strategy;After printing, according to the surface roughness of each region, implement differential polishing and adjust polishing parameter;According to processing data and prediction deviation, update theoretical prediction model and machine learning model, determine the processing parameter of the part to be formed using the updated digital twin model.The present application realizes the synergic closed loop of planning before printing, regulation in printing, zoned polishing after printing and model self-learning update, improves the consistency and accuracy of inner runner surface quality regulation.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing, and specifically relates to a method, system, device and storage medium for the coordinated control of the surface quality of internal flow channels. Background Technology

[0002] Selective Laser Melting (SLM) is an additive manufacturing technology that achieves near-net-shape forming of parts by melting and depositing metal powder layer by layer. It features high forming freedom, high material utilization, and suitability for manufacturing complex structures, and has been widely used in the manufacturing of complex components in aerospace and other fields. For parts with complex internal flow channel structures, SLM can overcome the limitations of traditional machining on the shape and accessibility of the flow channel space, achieving integrated forming of complex structures such as curved flow channels, variable cross-section flow channels, and micro-flow channels. However, due to factors such as the step effect, spheroidization effect, partial adhesion of sintered powder, and irregular overlap of the molten pool, the surface of the internal flow channel formed by SLM usually has a large surface roughness, which affects fluid transport performance and part service performance. Therefore, effective control of the surface quality of complex internal flow channels in SLM is of great significance.

[0003] Existing methods for improving the surface quality of complex internal flow channels in SLM (Self-Drawing Molding Machine) mainly fall into two categories: printing parameter optimization and post-printing polishing. Printing parameter optimization typically involves adjusting process parameters such as laser power, scanning speed, scanning spacing, and layer thickness to improve molten pool stability and interlayer overlap, thereby reducing the initial roughness of the formed surface. Post-printing polishing primarily employs abrasive flow polishing, electrochemical polishing, and multi-energy field assisted polishing to further treat the internal flow channel surface. For different flow channel structures, appropriate treatment methods can be selected based on the flow channel geometry and the accessibility of the polishing medium. For example, abrasive flow polishing can be used for straight areas, ultrasonic and magnetic field coupled abrasive flow polishing can be used for curved and difficult-to-reach areas, and electrochemical polishing or electrolyte plasma polishing can be used for micro- or blind-end areas to further reduce the surface roughness of the internal flow channels.

[0004] However, existing technologies still have certain shortcomings. On the one hand, there are significant differences in the geometry, polishing medium flow state, and surface quality of different regions in complex internal channels. Especially for regions with large aspect ratios, small radii of curvature, variable cross-sections, and micro-blind ends, the polishing medium is difficult to reach uniformly, which can easily lead to local under-polishing or over-polishing. Furthermore, there is a risk of abrasive residue and blockage in micro-channels. On the other hand, existing printing parameter optimization and subsequent polishing processes are usually independent of each other. It is difficult to coordinate the adjustment of printing parameters, local remelting, scanning strategies, and polishing parameters based on the melt pool state, actual contour deviation, and actual surface roughness after printing. At the same time, there is a lack of a mechanism to continuously correct the roughness prediction results and update the process model using actual processing data. This makes it difficult to form an effective closed loop between pre-printing process planning, in-situ control during printing, and post-printing differentiated polishing, thus limiting the accuracy and stability of surface quality control in complex internal channels. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, system, device and storage medium for coordinated control of the surface quality of internal flow channels, so as to reduce the local surface quality differences caused by complex structures such as bending, variable cross-section and micro blind ends, and improve the consistency and control accuracy of the surface quality of complex internal flow channels.

[0006] A method for synergistic control of the surface quality of an internal flow channel includes: Obtain the 3D model of the training part and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part. A digital twin model of the inner flow channel is constructed based on the 3D model of the training part and the geometric feature parameters of the inner flow channel. The digital twin model of the inner flow channel includes a parametric geometric model of the inner flow channel, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. The flow state of the polishing medium in the inner flow channel is simulated based on the flow field simulation model to obtain the flow field characteristics. The surface roughness of the inner flow channel is predicted based on the surface roughness theoretical prediction model to obtain the roughness prediction result before printing. The inner flow channel is divided into partitions based on the flow field characteristics, pre-printing roughness prediction results, and internal flow channel geometric characteristic parameters to obtain partitioning results; the printing parameters and polishing processes of the corresponding regions are matched according to the partitioning results to obtain pre-printing process planning results; Acquire the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area based on the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model to obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the roughness prediction result of the printing process based on the theoretical roughness prediction value and the residual correction value. The actual contour data of the printed layer is obtained, and the contour deviation is calculated based on the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel. The local remelting parameters are determined based on the contour deviation, the scanning strategy is determined based on the roughness prediction results of the printing process, and the SLM printing process is controlled based on the local remelting parameters and the scanning strategy to obtain the in-situ control results of the printing process. After SLM printing, the actual surface roughness of each region of the inner flow channel is obtained. Based on the actual surface roughness and the pre-printing process planning results, the polishing process corresponding to each region is determined to obtain a partitioned polishing scheme. The inner flow channel is polished differently according to the partitioned polishing scheme, and the surface roughness detection results during the polishing process are obtained. The polishing parameters are adjusted according to the comparison between the surface roughness detection results and the target roughness threshold of the corresponding region until the corresponding region reaches the target roughness threshold, and the polishing processing result is obtained. The process data is obtained by acquiring monitoring data, in-situ control results, and polishing results during the SLM printing process. The prediction deviation is calculated based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel. The model parameters of the surface roughness theoretical prediction model are updated based on the prediction deviation, and the machine learning correction model is updated based on the processing data to obtain the updated digital twin model of the inner flow channel. Obtain a 3D model of the part to be formed, determine the printing parameters and polishing process corresponding to the part to be formed based on the 3D model of the part to be formed and the updated internal flow channel digital twin model, and determine the corresponding local remelting parameters and scanning strategy based on the updated internal flow channel digital twin model during the SLM printing process of the part to be formed.

[0007] Optionally, obtain a 3D model of the training part, and extract the geometric feature parameters of the internal flow channel from the 3D model of the training part, including: Acquire a 3D model of the training part and identify the spatial orientation and cross-sectional topology of the internal flow channels in the 3D model of the training part. Based on the spatial orientation and cross-sectional topology of the internal flow channel, the internal flow channel diameter, radius of curvature, bending angle, depth-to-diameter ratio, and cross-sectional change rate are extracted to obtain the geometric characteristic parameters of the internal flow channel.

[0008] Optionally, a digital twin model of the internal flow channel is constructed based on the 3D model of the training part and the geometric feature parameters of the internal flow channel. Flow field characteristics are obtained based on the flow field simulation model, and the roughness prediction results before printing are obtained based on the surface roughness theory prediction model, including: Based on the 3D model of the training part and the geometric feature parameters of the internal flow channel, a parametric geometric model of the internal flow channel is established. The spatial orientation, cross-sectional topology and theoretical contour data of each position of the internal flow channel are characterized by the parametric geometric model of the internal flow channel. A flow field simulation model is constructed based on the parametric geometric model of the inner flow channel. The flow field simulation model is used to perform computational fluid dynamics simulation on the flow state of the polishing medium in the inner flow channel, and the flow velocity, turbulent kinetic energy and wall shear stress corresponding to each region of the inner flow channel are obtained to obtain the flow field characteristics. Based on the step effect, spheroidizing effect and powder adhesion effect of the inner flow channel surface roughness generated during the SLM forming process, a surface roughness theoretical prediction model is established. Based on the surface roughness theoretical prediction model, the theoretical roughness corresponding to each region of the inner flow channel is determined, and the roughness prediction result before printing is obtained. Set up a machine learning correction model to perform residual correction on the theoretical roughness prediction values ​​output by the surface roughness theoretical prediction model based on the melt pool state data during the SLM printing process. The parametric geometric model of the internal flow channel, the flow field simulation model, the surface roughness theoretical prediction model, and the machine learning correction model are linked together, and historical printing and polishing data are imported to construct a digital twin model of the internal flow channel.

[0009] Optionally, the inner flow channel is partitioned based on flow field characteristics, pre-printing roughness prediction results, and internal flow channel geometric characteristic parameters to obtain partitioning results; the printing parameters and polishing processes of the corresponding regions are matched according to the partitioning results to obtain pre-printing process planning results, including: Based on the flow field characteristics, the roughness prediction results before printing, and the geometric characteristic parameters of the inner flow channel, the inner flow channel is divided into a straight and easily accessible area, a curved and difficult-to-reach area, and a micro-blind end area, thus obtaining the partitioning results; Among them, the straight-through easily accessible zone is determined according to the flow state of the polishing medium in the inner channel, the curved difficult-to-access zone is determined according to the relationship between the inner channel curvature radius and the inner channel diameter, and the micro-blind end zone is determined according to the inner channel diameter and the inner channel end structure. For different regions, with the goal of reducing the surface roughness of the corresponding region after printing, at least one corresponding printing parameter is determined from laser power, scanning speed, scanning spacing and layer thickness; Based on the geometric characteristics of the internal flow channels and the accessibility of the polishing medium in different regions, the corresponding polishing process is matched, and the printing parameters, polishing process and regional information of each region are correlated to obtain the pre-printing process planning results.

[0010] Optionally, acquire molten pool state data during the SLM printing process, and obtain printing process roughness prediction results based on the surface roughness theoretical prediction model and the machine learning correction model, including: High-speed cameras or photoelectric sensors are used to acquire the molten pool radiation signals corresponding to each scanning path during the SLM printing process; The molten pool temperature and size are determined based on the molten pool radiation signal, thus obtaining molten pool state data; Based on the printing parameters and internal flow channel geometry of the current printing area, the theoretical roughness prediction value of the current printing area is determined by the surface roughness theoretical prediction model. The molten pool state data is input into the machine learning correction model, which then determines the residual correction value corresponding to the theoretical roughness prediction value based on the molten pool state data. By fusing the theoretical roughness prediction value with the residual correction value, the surface roughness of the inner flow channel in the current printing area is determined, and the roughness prediction result of the printing process is obtained.

[0011] Optionally, the actual contour data of the printed layer is obtained, the contour deviation is calculated based on the actual contour data and the theoretical contour data, and the SLM printing process is adjusted based on the contour deviation and the roughness prediction results of the printing process to obtain the in-situ adjustment results of the printing process, including: After the printing layer is completed, the inner flow channel contour corresponding to the printing layer is scanned in situ to obtain the actual contour data. Obtain the theoretical contour data corresponding to the printing layer in the parametric geometric model of the inner flow channel, and determine the contour deviation corresponding to each contour position based on the position difference between the actual contour data and the theoretical contour data; The contour deviation threshold is determined based on the layer thickness of the printed layer. The contour deviation is compared with the contour deviation threshold to identify abnormal locations where the contour deviation exceeds the contour deviation threshold. The local remelting power and the number of local remelting cycles are determined based on the contour deviation corresponding to the abnormal location, thus obtaining the local remelting parameters. The roughness prediction results of the printing process are compared with the target roughness threshold. When the roughness prediction results of the printing process do not meet the target roughness threshold, the scanning direction deflection angle of the adjacent printing layer is determined according to the surface roughness theory prediction model to obtain the scanning strategy. Local remelting is performed on abnormal locations based on local remelting parameters, and the scanning direction of adjacent printing layers is adjusted according to the scanning strategy to obtain the in-situ control results of the printing process.

[0012] Optionally, the actual surface roughness of each region of the inner flow channel is obtained after SLM printing, and differentiated polishing is performed based on the pre-printing process planning results to obtain the polishing results, including: After SLM printing is completed, the actual surface roughness of each region of the inner channel is obtained. The actual surface roughness is compared with the target roughness threshold of the corresponding region. Combined with the process planning results before printing, the processing priority and corresponding polishing process of each region are determined. For the easily accessible area, a liquid metal-driven abrasive flow polishing process was determined; for the curved and difficult-to-reach area, an ultrasonic and magnetic field coupled abrasive flow polishing process was determined; and for the micro-blind end area, an electrochemical polishing process was determined, resulting in a zoned polishing scheme. Differential polishing is performed on the corresponding area of ​​the inner flow channel according to the partitioned polishing scheme, and the surface roughness detection results of the corresponding area are obtained during the polishing process; The surface roughness detection result is compared with the target roughness threshold of the corresponding area. When the surface roughness detection result does not reach the target roughness threshold, at least one of the polishing parameters, polishing time, polishing pressure and polishing power is adjusted, and the corresponding area is polished. When the surface roughness detection result reaches the target roughness threshold, polishing of the corresponding area is stopped, and the polishing result is obtained.

[0013] A system for coordinated control of surface quality in an internal flow channel, comprising: The geometric feature extraction module is used to obtain the 3D model of the training part and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part. The digital twin construction and prediction module is used to construct a digital twin model of the inner flow channel based on the 3D model of the training part and the geometric feature parameters of the inner flow channel. The digital twin model of the inner flow channel includes a parametric geometric model of the inner flow channel, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. Based on the flow field simulation model, the flow state of the polishing medium in the inner flow channel is simulated to obtain the flow field characteristics. Based on the surface roughness theoretical prediction model, the surface roughness of the inner flow channel is predicted to obtain the roughness prediction result before printing. The pre-printing process planning module is used to partition the inner flow channel according to the flow field characteristics, pre-printing roughness prediction results, and inner flow channel geometric feature parameters to obtain partitioning results; and to match the printing parameters and polishing processes of the corresponding regions according to the partitioning results to obtain the pre-printing process planning results. The printing process roughness prediction module is used to acquire the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area based on the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model to obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the printing process roughness prediction result based on the theoretical roughness prediction value and the residual correction value. The in-situ control module for the printing process is used to acquire the actual contour data of the printed layer, calculate the contour deviation based on the actual contour data and the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel, determine the local remelting parameters based on the contour deviation, determine the scanning strategy based on the roughness prediction results of the printing process, and control the SLM printing process based on the local remelting parameters and the scanning strategy to obtain the in-situ control results of the printing process. The partitioned polishing control module is used to obtain the actual surface roughness of each region of the inner flow channel after SLM printing. Based on the actual surface roughness and the pre-printing process planning results, the corresponding polishing process for each region is determined to obtain the partitioned polishing scheme. The inner flow channel is polished differently according to the partitioned polishing scheme, and the surface roughness detection results during the polishing process are obtained. The polishing parameters are adjusted according to the comparison between the surface roughness detection results and the target roughness threshold of the corresponding region until the corresponding region reaches the target roughness threshold, and the polishing processing result is obtained. The digital twin model update module is used to acquire monitoring data, in-situ control results of the printing process, and polishing results during the SLM printing process to obtain processing data; calculate the prediction deviation based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel; update the model parameters of the surface roughness theoretical prediction model based on the prediction deviation, and update the machine learning correction model based on the processing data to obtain the updated digital twin model of the inner flow channel. The processing parameter determination module is used to acquire the three-dimensional model of the part to be formed, determine the printing parameters and polishing process corresponding to the part to be formed based on the three-dimensional model of the part to be formed and the updated internal flow channel digital twin model, and determine the corresponding local remelting parameters and scanning strategy based on the updated internal flow channel digital twin model during the SLM printing process of the part to be formed.

[0014] A terminal device includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it employs a method for coordinated control of the surface quality of the internal flow channel.

[0015] A computer-readable storage medium storing a computer program, wherein when the computer program is loaded and executed by a processor, an internal flow channel surface quality coordinated control method is employed.

[0016] The beneficial effects of this invention are: 1. By constructing a digital twin model of the internal flow channel, the geometric features of the internal flow channel, the flow field characteristics of the polishing medium, and the surface roughness prediction results are correlated. The internal flow channel is divided into zones according to the flow field state, geometric features, and roughness state of different regions. Then, corresponding SLM printing parameters and polishing processes are matched for different zones. At the same time, during the SLM printing process, the surface roughness of the internal flow channel is dynamically predicted by combining the melt pool state data. The local remelting parameters are determined based on the contour deviation between the actual contour and the theoretical contour of the printed layer. The scanning strategy is determined based on the roughness prediction results. This achieves synergy between pre-printing process planning, in-situ control during printing, and post-printing differentiated polishing. It can implement differentiated quality control for the forming state and polishing medium accessibility of different regions of complex internal flow channels, reduce the local surface quality differences caused by complex structures such as bending, variable cross-section, and micro blind ends, and improve the consistency and control accuracy of the surface quality of complex internal flow channels.

[0017] 2. Monitoring data, in-situ control results, and polishing results during the SLM printing process are fed back to the internal flow channel digital twin model as processing data. The parameters of the digital twin model are updated based on the prediction deviation between the actual surface roughness and the corresponding roughness prediction results. The machine learning correction model is continuously updated using new processing data, enabling the digital twin model to continuously correct the surface roughness prediction and process decision relationship based on the actual processing results. When processing the parts to be formed, the updated internal flow channel digital twin model can be used to determine the corresponding printing parameters, polishing process, local remelting parameters, and scanning strategy. This forms a closed loop of processing data feedback and model iterative update, improving the adaptability of the digital twin model to different internal flow channel structures and actual processing conditions, and improving the accuracy of subsequent part process parameter determination and surface quality control. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of a method for coordinated control of surface quality in an internal flow channel according to the present invention. Detailed Implementation

[0019] A method for synergistic control of the surface quality of an internal flow channel includes: S1. Obtain the 3D model of the training part and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part. Specifically, the geometric characteristic parameters of the internal flow channel include a set of parameters used to quantitatively describe the internal flow channel size, curvature, and cross-sectional changes, including the internal flow channel diameter D, radius of curvature R, curvature angle θ, depth-to-diameter ratio L / D, and cross-sectional change rate η.

[0020] Training parts: Known or sample parts used to build, calibrate, or update digital twin models. Training parts can be historical parts that have already been SLM printed and have printing, inspection, and polishing records, or sample parts specifically manufactured to build a process database.

[0021] Three-dimensional model: Three-dimensional digital geometric data used to describe the external shape and internal structure of a part. For this invention, the three-dimensional model can at least represent the boundary surface, spatial location, and cross-sectional shape of the internal flow channel, and can be derived from a CAD design model or three-dimensional data that can be converted into equivalent solid / mesh geometry.

[0022] Internal flow channels: channel or cavity structures located inside the solid part for fluid transport or passage of functional media, including straight flow channels, curved flow channels, variable cross-section flow channels, micro flow channels, and flow channels with blind ends.

[0023] Internal flow channel geometric characteristic parameters: a set of parameters used to quantify the internal flow channel morphology and serve as inputs for subsequent flow field simulation, region division, and process matching. These parameters include at least the flow channel diameter D, radius of curvature R, bending angle θ, depth-to-diameter ratio L / D, and cross-sectional change rate η.

[0024] First, the boundary of the inner flow channel cavity is identified from the 3D model of the training part, and the centerline or equivalent spatial orientation is extracted along the main extension direction of the inner flow channel. Then, cross-sectional sampling is performed along the centerline according to a preset step size to obtain the cross-sectional profile, equivalent diameter, spatial coordinates, and local tangential direction at each sampling position.

[0025] The flow channel diameter D can be taken as the actual diameter of a circular cross-section; for non-circular cross-sections, an equivalent diameter can be used according to the actual engineering definition. The radius of curvature R is obtained by the reciprocal of the local curvature of the centerline; the bending angle θ is determined according to the angle between the starting and ending tangential directions of the bending segment; in the depth-to-diameter ratio L / D, L is the axial or centerline length of the target flow channel segment; the cross-sectional change rate η is used to characterize the degree of dimensional change between adjacent cross-sections or target sections.

[0026] Geometric feature parameters are bound to spatial coordinates, layer numbers, and region numbers in the 3D model, so that subsequent CFD flow field results, roughness prediction results, theoretical profiles, actual profiles, and polishing results can be mapped to the same internal flow channel location.

[0027] Specifically, the spatial orientation of the internal flow channel is the extension path of the centerline or equivalent axis of the internal flow channel in the three-dimensional coordinate system, reflecting the straight segments, curved segments, branch segments of the flow channel, and the spatial connection relationship between each segment.

[0028] Cross-sectional topology: The shape of the cross section taken along the direction of the inner flow channel and its connection / change relationship, including structural information such as circular, non-circular, expansion, contraction, branching, merging and blind ends.

[0029] Inner channel diameter D: A parameter describing the local cross-sectional dimensions of the channel; the actual diameter is used for circular channels, while the equivalent diameter obtained according to the actual engineering definition can be used for non-circular channels.

[0030] Radius of curvature R: A parameter describing the degree of curvature of the flow channel centerline. When the local curvature of the centerline is κ, R = 1 / κ can be used. The smaller R is, the more steep the curvature.

[0031] Bending angle θ: The angle between the inlet direction and the outlet direction of the bending segment, or the direction change angle calculated from the tangential vectors at the start and end of the bending segment.

[0032] The depth-to-diameter ratio L / D is the ratio of the characteristic length L of the flow channel to the diameter D of the flow channel. It is used to indicate the degree of slenderness or the difficulty of the polishing medium reaching deep areas.

[0033] Cross-sectional change rate η: A parameter used to quantify the degree of change in cross-sectional dimensions along the flow direction. It can be calculated based on the relative changes in the characteristic diameter or area of ​​adjacent cross-sections and should be defined consistently within the same project.

[0034] The 3D model is imported into the geometry processing module, where topological identification is performed on the solids and cavities to extract the closed boundary surfaces belonging to the internal flow channels. The flow channel centerline is obtained through a centerline extraction algorithm, discretized into continuous sampling points, and a normal section is established at each sampling point.

[0035] Record the center point coordinates, local tangential vector, cross-sectional profile, and cross-sectional characteristic dimensions at each sampling location. Calculate R based on the local curvature of the centerline, calculate θ based on the start and end tangential vectors of the curved section, calculate L / D based on the centerline lengths L and D of the target segment, and calculate η based on the dimensional changes of adjacent cross-sections or target sections.

[0036] When the rate of change of cross section is expressed using the equivalent diameter, it can be defined as follows: ;in The equivalent diameter of the upstream reference section. The equivalent diameter of the downstream target section is used. The extracted geometric parameters are written into the digital twin model according to their coordinate positions, forming a spatial index for subsequent CFD mesh generation, surface tilt angle calculation, region partitioning, and generation of the theoretical contour of the printing layer.

[0037] Obtain the 3D model of the training part, and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part, including: Acquire a 3D model of the training part and identify the spatial orientation and cross-sectional topology of the internal flow channels in the 3D model of the training part. Based on the spatial orientation and cross-sectional topology of the internal flow channel, the internal flow channel diameter, radius of curvature, bending angle, depth-to-diameter ratio, and cross-sectional change rate are extracted to obtain the geometric characteristic parameters of the internal flow channel.

[0038] S2. Construct a digital twin model of the inner flow channel based on the 3D model of the training part and the geometric feature parameters of the inner flow channel. The digital twin model of the inner flow channel includes a parametric geometric model of the inner flow channel, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. Simulate the flow state of the polishing medium in the inner flow channel based on the flow field simulation model to obtain the flow field characteristics. Predict the surface roughness of the inner flow channel based on the surface roughness theoretical prediction model to obtain the roughness prediction result before printing. Specifically, digital twins are a technical system that links physical parts and their processing with geometric, physical, technological, and data models in virtual space, forming a two-way mapping through processing data feedback and model output. The digital twin in this invention is not simply a three-dimensional display model, but a composite model capable of flow field calculation, roughness prediction, process decision-making, and parameter updates.

[0039] The internal flow channel digital twin model is a composite virtual model built around the internal flow channel, including at least parametric geometric representation of the internal flow channel, simulation of the polishing medium flow field, theoretical prediction of surface roughness, and correction functions based on historical / real-time data. To support subsequent machine learning residual correction, a machine learning model is pre-set in the digital twin model and initially trained using historical processing data.

[0040] Polishing media: The working medium that enters the inner flow channel during the post-processing and participates in material removal or surface leveling, including the carrier fluid and abrasive particles in the abrasive flow, multiphase abrasive media, etc.

[0041] Flow state: The motion and force state of the polishing medium as its spatial position changes within the inner flow channel, mainly described by velocity, pressure, turbulence, and wall action.

[0042] Flow field characteristics: Data extracted from computational fluid dynamics simulation and used for regional evaluation. In this embodiment, at least the flow velocity distribution, turbulent kinetic energy distribution, and wall shear stress distribution are included.

[0043] Roughness prediction results: The predicted surface roughness values ​​of the inner flow channel obtained before printing from the multi-mechanism theoretical model of surface roughness in the digital twin can be output by location or region.

[0044] A digital twin model of the internal flow channel is constructed based on the 3D model of the training part and the geometric feature parameters of the internal flow channel. Flow field characteristics are obtained based on the flow field simulation model, and the roughness prediction results before printing are obtained based on the surface roughness theory prediction model, including: Based on the 3D model of the training part and the geometric feature parameters of the internal flow channel, a parametric geometric model of the internal flow channel is established. The spatial orientation, cross-sectional topology and theoretical contour data of each position of the internal flow channel are characterized by the parametric geometric model of the internal flow channel. A flow field simulation model is constructed based on the parametric geometric model of the inner flow channel. The flow field simulation model is used to perform computational fluid dynamics simulation on the flow state of the polishing medium in the inner flow channel, and the flow velocity, turbulent kinetic energy and wall shear stress corresponding to each region of the inner flow channel are obtained to obtain the flow field characteristics. Based on the step effect, spheroidizing effect and powder adhesion effect of the inner flow channel surface roughness generated during the SLM forming process, a surface roughness theoretical prediction model is established. Based on the surface roughness theoretical prediction model, the theoretical roughness corresponding to each region of the inner flow channel is determined, and the roughness prediction result before printing is obtained. Set up a machine learning correction model to perform residual correction on the theoretical roughness prediction values ​​output by the surface roughness theoretical prediction model based on the melt pool state data during the SLM printing process. The parametric geometric model of the internal flow channel, the flow field simulation model, the surface roughness theoretical prediction model, and the machine learning correction model are linked together, and historical printing and polishing data are imported to construct a digital twin model of the internal flow channel.

[0045] Specifically, firstly, a parametric geometric model of the internal flow channel is established in virtual space based on the 3D model of the training part, corresponding to the spatial position of the physical part. The spatial orientation, cross-sectional topology, layer number, and theoretical contour of the internal flow channel are then defined. This geometric model serves two purposes: firstly, it defines the CFD computational domain, and secondly, it provides theoretical contour data for each layer during the printing process.

[0046] Secondly, the internal flow channel cavity is used as the fluid computation domain. The inlet, outlet and wall boundary conditions of the polishing medium are set. CFD solution is performed based on the continuity equation and Navier-Stokes equation of incompressible Newtonian fluid to obtain the flow velocity, turbulent kinetic energy and wall shear stress.

[0047] Furthermore, a multi-mechanism coupling theoretical model for surface roughness is established based on the step effect, spheroidization effect, and powder adhesion effect. Finally, key parameters from historical printing and polishing data are imported into the aforementioned geometric, flow field, and roughness models, and the correlation between region, layer number, process parameters, and detection results is established, forming a composite digital twin model that integrates geometric, fluid dynamic, and process information.

[0048] Based on the 3D model of the training part, a parametric geometric model of the internal flow channel is generated, and the obtained parameters such as D, R, θ, L / D, and η are written into the corresponding flow channel segment or wall element. A unique number is assigned to each internal flow channel region, and a mapping relationship is established between the region number and the 3D coordinates, printing layer number, and section number.

[0049] The theoretical profile is obtained by slicing the three-dimensional geometric model at each printing layer height. This serves as the benchmark for subsequent layer-by-layer profile deviation calculations; simultaneously, the inner flow channel cavity boundary is exported as the CFD calculation domain.

[0050] Historical printing parameters, molten pool status, post-printing roughness, polishing parameters, and post-polishing roughness are imported into the database by region number and associated with the geometric model, flow field simulation model, and roughness theory model to form a predictable and updatable digital twin model of the internal flow channel.

[0051] This method uses discretization and solution of the governing equations for fluid mass conservation and momentum conservation to numerically calculate the velocity, pressure, and wall effects of the polishing medium in the internal flow channel.

[0052] Incompressible Newtonian fluid: A fluid model in which the density is approximately independent of pressure during calculation, and the shear stress and velocity gradient satisfy the Newtonian viscosity relationship.

[0053] Flow rate: The magnitude or velocity vector of the polishing medium at a certain spatial location.

[0054] Turbulent kinetic energy: A flow field quantity that characterizes the intensity of fluid velocity fluctuations and is used to evaluate local flow disturbances and the degree of medium mixing.

[0055] Wall shear stress: The shear force per unit area exerted by the fluid on the wall tangentially along the inner channel wall, denoted as It can be used to evaluate the strength of the polishing medium's effect on the wall surface and the medium's accessibility.

[0056] In the digital twin model, the internal flow channel cavity is set as a fluid domain, with flow rate, velocity, or pressure boundaries for the polishing medium at the inlet and pressure or outflow boundaries at the outlet. The internal flow channel walls are configured with no-slip wall conditions. The underlying carrier fluid for the polishing medium can be approximated as an incompressible Newtonian fluid.

[0057] The CFD solution employs the continuity equation and the Navier-Stokes equation. After the solution is obtained, the velocity field, turbulent kinetic energy, and wall shear stress are mapped to the wall elements or region numbers of the parametric geometric model for subsequent four types of region division and accessibility determination of the polishing process.

[0058] The continuity equation for incompressible fluids is:

[0059] in, For vector differential operators, Let be the fluid velocity vector of the polishing medium. 0 indicates that the velocity field divergence is zero under incompressible conditions, which means that mass conservation is satisfied.

[0060] The momentum conservation of the polishing medium can be expressed by the Navier-Stokes equations as follows:

[0061] in, The density of the polishing medium fluid. This is the fluid velocity vector. For time. Let be the local rate of change of velocity over time. This is the fluid convection acceleration term. This refers to fluid pressure. This represents the pressure gradient. This refers to dynamic viscosity. For the Laplace operator. This is the volume force vector acting on a unit volume of fluid.

[0062] Step effect: When SLM is discretized layer by layer, the inclined surface is formed by the superposition of finite layer thicknesses to form a step-like profile, which causes the roughness component.

[0063] Spheroidization effect: Surface undulations caused by the fracture or spherical contraction of the melt channel under the influence of surface tension and Rayleigh instability.

[0064] Powder adhesion effect: The roughness contribution caused by the adhesion of incompletely melted or partially sintered powder particles to the surface near the inner flow channel wall.

[0065] Multi-mechanism coupling model: The roughness components generated by the step effect, spheroidization effect and powder adhesion effect are quantitatively combined to obtain the theoretical prediction value of the surface roughness of the internal flow channel.

[0066] Surface roughness Ra: the arithmetic mean deviation of the profile, used to characterize the degree of micro-undulation on the surface of the internal flow channel.

[0067] The theoretical value of the surface roughness of the internal flow channel is expressed as the superposition of three physical mechanism components. In the calculation, the step effect component is obtained based on the printing layer thickness and surface tilt angle; the spheroidizing effect component is obtained based on the surface tension of the liquid metal, the density of the liquid metal, the scanning spacing and the laser line energy density; and the powder adhesion effect component is obtained based on the powder particle size, overlap rate and reference energy density.

[0068] For the same internal flow channel region, the actual printing parameters P, V, h, a and material parameters are substituted into the following formula to obtain the roughness prediction result before printing; during the printing process, this theoretical value is also used as the basis value for machine learning residual correction.

[0069] The multi-mechanism coupling relationship is as follows:

[0070] in, This is the theoretical predicted value for the surface roughness of the internal flow channel. This represents the roughness component due to the step effect. This represents the roughness component due to the spheroidizing effect. This represents the roughness component due to powder adhesion effect.

[0071] The staircase effect roughness component is expressed as:

[0072] in, : SLM printing layer thickness. : The surface tilt angle of the inner flow channel surface to be predicted relative to the forming direction or a specified reference; the definition should be kept consistent within the same model. : The tangent function of the surface tilt angle α. : Tilt angle correction function; when When approaching 90°, The value of approaches 1. Roughness component caused by step effect.

[0073] The roughness component of the spheroidizing effect is expressed as:

[0074] in, Roughness component caused by spheroidization effect. The sphericity effect proportionality coefficient can be calibrated using historical processing data and updated during the model's autonomous learning phase. Surface tension of liquid metal at temperature T. Temperature of the molten pool or liquid metal. : Density of liquid metal. : Gravitational acceleration. Scanning interval. Laser line energy density. : Threshold energy density of spheroidization effect.

[0075] The laser line energy density can be determined by the following formula:

[0076] in, This refers to the laser power. This refers to the laser scanning speed. This refers to the laser energy input per unit scan length.

[0077] The roughness component of the powder adhesion effect is expressed as:

[0078] in, This represents the roughness component caused by powder adhesion effect. This is the proportionality coefficient for powder adhesion effect, which can be calibrated using historical data. To represent the powder particle size. λ is the melt channel overlap ratio. This represents the laser line energy density. For reference energy density.

[0079] S3. Divide the inner flow channel into partitions based on the flow field characteristics, pre-printing roughness prediction results, and inner flow channel geometric characteristic parameters to obtain partitioning results; match the printing parameters and polishing processes of the corresponding regions based on the partitioning results to obtain pre-printing process planning results; Specifically, based on the geometric conditions at different locations, the accessibility of the polishing medium, the strength of the flow field, and the roughness risk, the continuous inner flow channel is divided into regions with different processing characteristics.

[0080] Zoning results: The region category and boundary information corresponding to each spatial location. This invention adopts three types of regions: easily accessible areas, difficult-to-reach areas, and micro-blind areas.

[0081] Printing parameters: Process parameters for controlling local energy input and melt flow formation during SLM forming, including at least laser power P, scanning speed V, scanning spacing h, and layer thickness. One or more of them.

[0082] Polishing process: After printing, the processing route used to reduce the surface roughness of the internal flow channel includes liquid metal driven abrasive flow, ultrasonic-magnetic field coupled three-phase abrasive flow, overpotential electrochemical polishing or electrolyte plasma polishing, etc.

[0083] Pre-printing process planning results: The regional-level process plan formed before the start of solid SLM printing, which records at least the category of each region, the corresponding printing parameters, the pre-selected polishing process, the expected roughness, and the optional virtual test polishing results.

[0084] The inner flow channel is divided into partitions based on flow field characteristics, pre-printing roughness prediction results, and internal flow channel geometric parameters, yielding partitioning results. Based on these partitioning results, printing parameters and polishing processes for the corresponding regions are matched to obtain pre-printing process planning results, including: Based on the flow field characteristics, the roughness prediction results before printing, and the geometric characteristic parameters of the inner flow channel, the inner flow channel is divided into a straight and easily accessible area, a curved and difficult-to-reach area, and a micro-blind end area, thus obtaining the partitioning results; Among them, the straight-through easily accessible zone is determined according to the flow state of the polishing medium in the inner channel, the curved difficult-to-access zone is determined according to the relationship between the inner channel curvature radius and the inner channel diameter, and the micro-blind end zone is determined according to the inner channel diameter and the inner channel end structure. For different regions, with the goal of reducing the surface roughness of the corresponding region after printing, at least one corresponding printing parameter is determined from laser power, scanning speed, scanning spacing and layer thickness; Based on the geometric characteristics of the internal flow channels and the accessibility of the polishing medium in different regions, the corresponding polishing process is matched, and the printing parameters, polishing process and regional information of each region are correlated to obtain the pre-printing process planning results.

[0085] Specifically, during partitioning, geometric parameters, CFD flow field characteristics, and pre-printing roughness prediction results are mapped to the same grid or sampling location in the parametric geometric model. Preferably, the region where the polishing medium can pass smoothly and the wall shear stress reaches the threshold is defined as the straight-through easily accessible region; the bend region with a radius of curvature R less than 5 times the channel diameter D is defined as the bend-difficult-to-access region; and the inner channel region with a diameter D not greater than 2 mm or with blind ends is defined as the micro-blind-end region.

[0086] For each partition, candidate SLM printing parameter combinations are established, and parameter combinations are screened within the parameter range allowed by the equipment and materials with the goal of minimizing the expected surface roughness of the corresponding area.

[0087] Simultaneously, a rule library for mapping printing parameters, flow channel characteristics, and polishing processes is established. For easily accessible areas, liquid metal-driven abrasive flow polishing is pre-selected; for difficult-to-reach areas, ultrasonic-magnetic field coupled three-phase abrasive flow polishing is pre-selected; and for micro-blind end areas, overpotential electrochemical polishing or electrolyte plasma polishing is pre-selected.

[0088] Virtual polishing simulations can be performed in a digital twin environment before physical polishing, and the roughness reduction trend, processing priority, and expected polishing time of different areas can be evaluated based on the material removal model.

[0089] Directly accessible zone: This is the area where the polishing medium can pass through relatively smoothly, with continuous local flow and a set level of wall shear stress. Preferably, it utilizes wall shear stress. As one of the criteria for accessibility.

[0090] Difficult-to-reach areas due to curvature: regions with small radii of curvature where conventional polishing media are difficult to reach stably or form a uniform wall effect due to the influence of curved flow. The preferred criterion is R < 5D.

[0091] Micro-blind end region: A region with a small local size or a closed end, where polishing media is difficult to remove. The preferred criteria are D≤2mm or the presence of a blind end.

[0092] : Wall shear stress threshold, used to distinguish whether the effect of the polishing medium on the wall has reached a preset achievable level; Computational cells or sampling points are established on the walls of the parametric geometric model. For each cell, geometric features such as D, R, θ, L / D, and η, as well as predicted values ​​of CFD flow velocity, turbulent kinetic energy, wall shear stress, and roughness are read. Cells are classified according to preset criteria, and then spatially continuous cells of the same category are merged into regions.

[0093] Preferably: Furthermore, areas with relatively straight flow paths are classified as straight-through easily accessible areas; areas with bends where R < 5D are classified as curved and difficult-to-access areas; and areas with D ≤ 2 mm or blind ends are classified as micro-blind end areas.

[0094] When multiple criteria are met at the same location, composite labels can be retained and prioritized according to the category with higher media accessibility risk during process selection; alternatively, priority rules can be established, such as prioritizing micro-blind end risk over bending and difficult-to-reach risk, and prioritizing bending and difficult-to-reach risk over ordinary straight-through areas.

[0095] Laser power P: The power output by the SLM laser during scanning, which affects the heat input per unit time and the size of the molten pool.

[0096] Scanning speed V: The linear velocity of the laser spot moving along the predetermined scanning trajectory, which, together with P, affects the laser line energy density.

[0097] Scan spacing h: The center distance between adjacent scan tracks, which affects the overlap of melt tracks and surface continuity.

[0098] Layer thickness The thickness of a single layer corresponding to each powder layer laid and melted by SLM directly affects the step effect.

[0099] Polishing media accessibility: The ability of polishing media to enter, pass through and effectively act on the wall of the target area, which can be comprehensively evaluated by local flow velocity, wall shear stress, geometric curvature, depth-to-diameter ratio and blind end structure.

[0100] Process mapping rule base: A data structure that stores the correspondence between regional geometric / flow field characteristics, printing parameters, and post-processing polishing processes, which can be constructed from historical processing data and virtual simulation results.

[0101] Candidate printing parameter combinations are established for each partition, and the resulting expected surface roughness is used as the objective function. The upper and lower limits of each parameter are determined jointly by the SLM equipment capabilities, material processing window, and historical database.

[0102] According to the mapping rules, liquid metal-driven abrasive flow polishing is pre-selected for the easily accessible zone; ultrasonic-magnetic field coupled three-phase abrasive flow polishing is pre-selected for the curved and difficult-to-reach zone; and overpotential electrochemical polishing or electrolyte plasma polishing is pre-selected for the micro-blind end zone.

[0103] After completing the parameter and process matching, the area number, area category, printing parameter combination, pre-selected polishing process, and virtual pre-polishing results are saved as the pre-printing process planning results.

[0104] The combination of printing parameters, aimed at minimizing the expected roughness of the corresponding region, can be expressed as:

[0105]

[0106] in, For the first The internal flow channel region has parameters P, V, h, Expected surface roughness under the given conditions. Number the internal flow channel area. This refers to the laser power. , The minimum and maximum permissible laser power are given. V represents the scanning speed. , The minimum and maximum allowable scan rates are given. h represents the scan interval. , The minimum and maximum allowable scan spacing. The thickness is the layer thickness. , These represent the minimum and maximum allowable layer thicknesses. `st` is an abbreviation for `subject to`, indicating that the objective optimization must satisfy the subsequent parameter range constraints.

[0107] S4. Obtain the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area based on the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model, obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the roughness prediction result of the printing process based on the theoretical roughness prediction value and the residual correction value. Specifically, the molten pool is a transient liquid or semi-liquid region formed by metal powder under laser irradiation during SLM scanning. Its temperature and geometry are related to local energy input, molten channel stability, and surface quality.

[0108] Molten pool status data: molten pool process quantities acquired and calibrated by high-speed camera devices or photoelectric sensors, including molten pool temperature field or molten pool temperature, as well as molten pool dimensions such as molten width, molten depth, and molten length.

[0109] Machine learning model: A data-driven regression model set within the digital twin model to learn the residual relationship between theoretical predictions and actual results of surface roughness. Its input is molten pool state data, and its output is a roughness residual correction term; the specific algorithm structure is not a necessary limitation.

[0110] Internal flow channel surface roughness prediction result: The expected surface roughness of the current layer or current region is obtained by fusing theoretical roughness prediction value and machine learning residual correction term during the printing process.

[0111] Acquire molten pool state data during SLM printing, and obtain printing process roughness prediction results based on surface roughness theoretical prediction models and machine learning correction models, including: High-speed cameras or photoelectric sensors are used to acquire the molten pool radiation signals corresponding to each scanning path during the SLM printing process; The molten pool temperature and size are determined based on the molten pool radiation signal, thus obtaining molten pool state data; Based on the printing parameters and internal flow channel geometry of the current printing area, the theoretical roughness prediction value of the current printing area is determined by the surface roughness theoretical prediction model. The molten pool state data is input into the machine learning correction model, which then determines the residual correction value corresponding to the theoretical roughness prediction value based on the molten pool state data. By fusing the theoretical roughness prediction value with the residual correction value, the surface roughness of the inner flow channel in the current printing area is determined, and the roughness prediction result of the printing process is obtained.

[0112] Specifically, during printing, a high-speed camera or photoelectric sensor collects the molten pool radiation signal corresponding to each scanning path in real time, and synchronizes the signal with the layer number, scanning path, and inner flow channel region number. The high-speed camera method can estimate the temperature and extract the molten pool boundary through the calibrated image grayscale or multi-band radiation information; the photoelectric sensor method can obtain the radiation intensity corresponding to the scanning position through time synchronization, and then calculate the molten pool temperature characteristics after calibration.

[0113] The digital twin model first calculates the theoretical prediction value of surface roughness based on the current printing parameters and geometric features. Then, it inputs the melt pool temperature, melt width, melt depth and melt length into a pre-trained machine learning model to obtain the residual correction value. The two are then superimposed to form the surface roughness prediction result during the printing process.

[0114] Melt pool radiation signal: The thermal or optical radiation response of the molten pool during the SLM process can be expressed as image grayscale, radiation intensity, spectral intensity, or corresponding electrical signals.

[0115] Molten pool temperature: A thermal state parameter of the molten pool obtained from radiation signals through equipment calibration or a radiation thermometry model, denoted as .

[0116] Molten pool size: The geometric dimensions of the molten pool, including at least the weld width. Melting depth and melting length One or more of them.

[0117] During SLM printing, the sampling clock of the high-speed camera or photoelectric sensor is synchronized with the laser scanning controller, and the path number, layer number, sampling time and spatial coordinates are saved for each scanning path.

[0118] When using high-speed imaging, background subtraction, radiation calibration, and boundary segmentation are performed on the molten pool image to obtain the molten pool temperature field, as well as the molten pool width and length; the melt depth can be obtained by combining historical calibration, coaxial imaging features, or a calibration model associated with process parameters. When using photoelectric sensors, the radiation intensity signal is calibrated to obtain temperature characteristics, and the molten pool size is obtained by combining it with the scanning path and historical molten pool size calibration relationship.

[0119] Ultimately, a molten pool state vector is formed that corresponds one-to-one with the current scan path and the inner flow channel region. , , , The data is uploaded to the digital twin model via a data chain.

[0120] S5. Obtain the actual contour data of the printed layer, calculate the contour deviation based on the actual contour data and the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel; determine the local remelting parameters based on the contour deviation, determine the scanning strategy based on the roughness prediction results of the printing process, and regulate the SLM printing process based on the local remelting parameters and the scanning strategy to obtain the in-situ regulation results of the printing process. Specifically, the printed layer is the single layer or the current layer that has been scanned and melted during the SLM layer-by-layer forming process.

[0121] Actual contour data: The actual height or shape data obtained by scanning the flow channel boundary in the current printing layer using a confocal ranging device or an equivalent in-situ contour detection device.

[0122] Theoretical profile data: Design profile / theoretical height data obtained by cutting or calculating at the corresponding floor height based on the parametric geometric model of the internal flow channel.

[0123] Contour deviation: The difference between the actual contour and the theoretical contour at the same coordinates, used to identify local protrusions or out-of-tolerance locations.

[0124] Local remelting parameters: Parameters used when performing low-power fixed-point laser remelting at abnormal locations, including at least the remelting power and the number of remelting cycles.

[0125] Scanning strategy: Control rules used to change the scanning direction of adjacent printed layers, including the scanning direction deflection angle and its selection method.

[0126] In-situ control: Quality repair or process adjustment is implemented based on the inspection results while the part is still in the SLM manufacturing process, without waiting for all printing to be completed.

[0127] The actual contour data of the printed layer is obtained. The contour deviation is calculated based on the actual contour data and the theoretical contour data. The SLM printing process is then adjusted based on the contour deviation and the roughness prediction results of the printing process to obtain the in-situ adjustment results of the printing process, including: After the printing layer is completed, the inner flow channel contour corresponding to the printing layer is scanned in situ to obtain the actual contour data. Obtain the theoretical contour data corresponding to the printing layer in the parametric geometric model of the inner flow channel, and determine the contour deviation corresponding to each contour position based on the position difference between the actual contour data and the theoretical contour data; The contour deviation threshold is determined based on the layer thickness of the printed layer. The contour deviation is compared with the contour deviation threshold to identify abnormal locations where the contour deviation exceeds the contour deviation threshold. The local remelting power and the number of local remelting cycles are determined based on the contour deviation corresponding to the abnormal location, thus obtaining the local remelting parameters. The roughness prediction results of the printing process are compared with the target roughness threshold. When the roughness prediction results of the printing process do not meet the target roughness threshold, the scanning direction deflection angle of the adjacent printing layer is determined according to the surface roughness theory prediction model to obtain the scanning strategy. Local remelting is performed on abnormal locations based on local remelting parameters, and the scanning direction of adjacent printing layers is adjusted according to the scanning strategy to obtain the in-situ control results of the printing process.

[0128] Specifically, after each layer is printed, a confocal ranging device is used to perform in-situ scanning of the flow channel profile within that layer, and then compares it point-by-point with the theoretical profile of the same layer in the digital twin model, or after registration. Anomalies with protrusion heights exceeding 150% of the layer thickness are identified.

[0129] When the abnormal protrusion satisfies When the thickness is >1.5a, a low-power laser of 50-100W is used to perform targeted remelting at abnormal locations before the next layer of powder is laid. The number of scans / remeltings is 1-3.

[0130] For regions where the predicted roughness value during the printing process exceeds the target roughness threshold, the scanning direction of adjacent layers is adjusted, and the candidate direction deflection angle is determined according to the principle of minimizing the sum of the roughness caused by the step effect and the spheroidization effect.

[0131] After performing local remelting and scanning direction adjustment, the corresponding layer's molten pool monitoring data, contour scanning data, abnormal locations, remelting parameters, and scanning adjustment parameters are recorded as in-situ control results of the printing process, and then transmitted back to the digital twin model via data link.

[0132] Theoretical prediction of surface roughness: Based on the current region's geometric features and actual printing parameters P, V, h, using a multi-mechanism coupled model. Calculated .

[0133] Machine learning residual correction value: The value output by the machine learning model based on the melt pool state data. It is used to compensate for the effects of equipment differences, powder batch differences, and transient molten pool fluctuations that are not explicitly represented in the theoretical model.

[0134] Residual learning: A machine learning model is trained with the actual surface roughness minus the theoretical prediction as the supervised target, so that the model does not directly replace the theoretical model, but learns the prediction error of the theoretical model.

[0135] Current printed layer: The SLM forming layer corresponding to the layer number currently undergoing quality evaluation; roughness prediction can be further output according to the internal flow channel region within the current layer.

[0136] The machine learning model is initially configured during the digital twin model construction phase. Initial training data is obtained from historical processing records: the input consists of historical melt pool temperature, melt width, melt depth, and melt length; the supervision target is the difference between the actual surface roughness of the same region and the theoretically predicted value. Any model capable of continuous numerical regression can achieve this function, without limiting the specific network structure.

[0137] When printing online, the theoretical model first calculates the current area. Then, the real-time molten pool status is input into the machine learning model to obtain... Adding the two together gives . After comparison with the target roughness threshold, it can be used as a basis for adjusting the scanning strategy.

[0138] The roughness fusion prediction relationship during the printing process is as follows:

[0139] in, This is the predicted surface roughness for the current printed layer or the current inner flow channel region. This represents the theoretical prediction value of the surface roughness model under the current printing parameters. P is the laser power. V is the scanning speed. h is the scanning spacing. The thickness is the layer thickness. This is the roughness residual correction term output by the machine learning model. This refers to the temperature of the molten pool. This represents the width of the molten pool. Indicates the depth of the molten pool / melt depth; is the length of the molten pool.

[0140] Theoretical profile data Theoretical design profile obtained by slicing the three-dimensional geometric model at the current layer height.

[0141] Contour deviation The difference between the actual profile height and the theoretical profile height at the same coordinate position on the plane. Positive deviation can characterize local bulges, while negative deviation can characterize local under-formation or depressions.

[0142] Registration: Before calculating the difference, the actual measured coordinates are unified with the theoretical coordinates to make their spatial positions consistent.

[0143] right and After coordinate registration, the difference is calculated point by point to obtain the profile deviation field. Further calculate the maximum positive deviation within the target flow channel region. It is used to identify local abnormal protrusions.

[0144] Profile deviation is calculated using the following formula:

[0145] in, coordinates The contour deviation at the location. x and y are the coordinates of the printing layer plane. The actual profile height measured at coordinates (x, y). The theoretical profile height given for the parameterized geometric model of the digital twin at the same location.

[0146] The contour deviation threshold is used to determine whether a protrusion requires in-situ repair, based on layer thickness. 150% is used as the anomaly criterion, that is >1.5 .

[0147] Abnormal location: The coordinate location or continuous area where the contour deviation exceeds the threshold.

[0148] Local remelting power: Low-power laser output used for targeted remelting of abnormal protrusions. Preferably 50–100 W.

[0149] Number of remelting cycles: The number of times a fixed-point remelting scan is performed on the same abnormal location, preferably 1 to 3 times.

[0150] Find in the target area And compare it with 1.5a. When No more than 1.5 If the toner is applied normally, printing can continue as usual; if the toner is applied normally, printing can continue as usual. Greater than 1.5 When the maximum deviation location and the continuous area exceeding the threshold are marked as abnormal locations.

[0151] Select the local remelting power from a low power range of 50–100W based on the abnormal deviation amplitude and area size, and choose 1–3 remelting cycles. Remelting is preferably performed before the next layer of powder is laid to reduce the impact of abnormal protrusions on subsequent powder laying and the formation of the inner flow channel surface.

[0152] The specific power and number of times can be established by looking up a table or regression relationship based on the repair effect of similar anomalies in history, and will continue to be updated after the processing data is returned.

[0153] The layers that are adjacent to the current layer in the forming sequence mainly refer to the next layer or several adjacent layers that will be printed after the current layer.

[0154] Scanning direction: The direction angle of the SLM laser when performing the main filling scan within a layer.

[0155] Directional deflection angle θ: The change in angle between the scanning direction of adjacent layers and the current layer or reference direction.

[0156] Scanning strategy: A method for controlling the deflection angle of the scanning direction of adjacent layers based on the roughness prediction results.

[0157] The result With the regional target roughness threshold Comparison. When No more than When, the preset scanning strategy can be maintained; when Exceed When the sum of the step effect component and the spherical effect component is calculated in the candidate direction deflection angle set, the direction deflection angle corresponding to the smallest sum is selected.

[0158] By changing the scanning direction of adjacent layers, the staircase effect related to surface tilt and the spheroidization effect related to melt channel stability are suppressed together, thus forming a scanning strategy for subsequent layers.

[0159] The direction deflection angle is determined by the following formula:

[0160] in, The preferred deflection angle for adjacent layer scanning directions. This represents the candidate scan direction deflection angle. The roughness component of the step effect under the condition of candidate scanning direction deflection angle θ. The roughness component of the spheroidization effect under the condition of candidate scanning direction deflection angle θ.

[0161] Corresponding printing area: The physical printing area determined by the location of the abnormal contour, the current layer region number, and the location where the roughness prediction exceeds the threshold.

[0162] In-situ control results during the printing process: the actual local remelting location, remelting power, number of remelting cycles, scanning direction deflection angle, and records of monitoring data before and after execution.

[0163] The equipment controller calls a local low-power scanning program to complete fixed-point remelting based on the abnormal position coordinates output by the digital twin model; for areas where the roughness prediction exceeds the threshold, the scanning direction parameters of the next layer or adjacent layer are modified.

[0164] After the adjustment is completed, the monitoring data of each layer of molten pool, the scanning data of the contour layer, and the corresponding local remelting and scanning strategy parameters are transmitted back to the digital twin model via the data link and stored in the process database as processing data for the model update stage.

[0165] S6. Obtain the actual surface roughness of each region of the inner flow channel after SLM printing. Determine the polishing process corresponding to each region based on the actual surface roughness and the pre-printing process planning results to obtain a partitioned polishing scheme. Perform differentiated polishing on the inner flow channel according to the partitioned polishing scheme and obtain the surface roughness detection results during the polishing process. Adjust the polishing parameters based on the comparison between the surface roughness detection results and the target roughness threshold of the corresponding region until the corresponding region reaches the target roughness threshold to obtain the polishing processing result. Specifically, actual surface roughness: After SLM printing is completed and the support is removed, the actual roughness distribution of the region is obtained by industrial CT, endoscopy, or other detection methods that can obtain the surface roughness of the internal flow channel.

[0166] Zoned polishing scheme: The zone-level polishing route, processing sequence and parameter scheme is formed by jointly evaluating the zone type formed before printing, the process pre-matching result and the actual roughness after printing.

[0167] Differentiated polishing: different polishing mechanisms, equipment combinations or process parameters are used for different internal flow channel areas, rather than using a single process for the entire internal flow channel.

[0168] Real-time surface roughness: The current roughness detection value obtained continuously or periodically during the polishing process, used to determine whether the target has been achieved.

[0169] Target roughness threshold: The maximum allowable roughness value preset based on part design, functional location, or region category; for example, complex curved flow channels typically require... Not greater than 5μm, critical throttling parts may require No larger than 1.6 μm; specific project settings should be based on design requirements.

[0170] Polishing results: A collection of data recording the final roughness achieved in each area, the process used, polishing time, pressure, power, and processing sequence.

[0171] After SLM printing, the actual surface roughness of each region of the inner flow channel is obtained, and differentiated polishing is performed based on the pre-printing process planning results to obtain the polishing results, including: After SLM printing is completed, the actual surface roughness of each region of the inner channel is obtained. The actual surface roughness is compared with the target roughness threshold of the corresponding region. Combined with the process planning results before printing, the processing priority and corresponding polishing process of each region are determined. For the easily accessible area, a liquid metal-driven abrasive flow polishing process was determined; for the curved and difficult-to-reach area, an ultrasonic and magnetic field coupled abrasive flow polishing process was determined; and for the micro-blind end area, an electrochemical polishing process was determined, resulting in a zoned polishing scheme. Differential polishing is performed on the corresponding area of ​​the inner flow channel according to the partitioned polishing scheme, and the surface roughness detection results of the corresponding area are obtained during the polishing process; The surface roughness detection result is compared with the target roughness threshold of the corresponding area. When the surface roughness detection result does not reach the target roughness threshold, at least one of the polishing parameters, polishing time, polishing pressure and polishing power is adjusted, and the corresponding area is polished. When the surface roughness detection result reaches the target roughness threshold, polishing of the corresponding area is stopped, and the polishing result is obtained.

[0172] Specifically, after printing is completed and the support is removed, the actual roughness distribution of each region is obtained, compared with the target threshold, and the processing priority and process route are determined by combining the pre-printing zoning planning and virtual pre-splashing results in the digital twin model.

[0173] The easily accessible zone is polished using liquid metal-driven abrasive flow, with a polishing time of 5–15 min. After polishing, high-pressure gas is introduced for alternating forward and reverse purging at least three times. The difficult-to-reach curved zone is polished using ultrasonic-magnetic coupled three-phase abrasive flow, with an ultrasonic frequency of 20–50 kHz and a polishing time of 15–30 min. During polishing, the flow field can be monitored in real time using particle image velocimetry and optimized in a closed-loop manner by a computer numerical control system. The micro-blind end zone is polished using overpotential electrochemical polishing or electrolyte plasma polishing. A conformal flexible electrode can be used, and the distance between the electrode and the flow channel wall should be maintained at approximately 0.5–1 mm.

[0174] For complex internal flow channels containing multiple region types, the processing follows a progressive step-by-step sequence: micro-blind end region – difficult-to-reach curved region – easily accessible straight region, gradually reducing roughness. In-situ or periodic inspections are performed after each polishing step. When the actual roughness reaches the target threshold, polishing of that region is immediately stopped. If the target is not met, parameters are adjusted according to the material removal model, polishing continues for 1–2 minutes, and the inspection is repeated, cycling until the target is met. After all polishing is completed, industrial CT is used for final non-destructive testing to confirm that the surface roughness of the entire flow channel meets the target and that there is no abrasive or processing media residue.

[0175] After each layer of laser scanning is completed, before the next layer of powder is laid or within the interlayer detection window allowed by the equipment, the inner flow channel profile is scanned using a confocal ranging device. The detection coordinate system is calibrated with the SLM equipment coordinate system so that the (x,y,z) data obtained from the scan can correspond to the theoretical profile of the same layer in the digital twin model.

[0176] After outlier removal, coordinate registration, and necessary filtering of the original contour data, the following is formed: And store them by layer number and area number.

[0177] Processing priority: For multiple internal flow channel areas in the same part, the polishing order is determined according to the degree of roughness deviation, the risk of medium accessibility, and the difficulty of the process.

[0178] Industrial CT: A detection method that uses X-ray computed tomography to obtain non-destructive information about the internal structure of parts. It can be used to evaluate the morphology of internal flow channels, residual media, and reconstructable surface conditions.

[0179] Endoscope: A detection device that can enter or observe the interior of the flow channel and acquire images / morphological information. It can be combined with calibration or surface measurement modules to obtain regional roughness information.

[0180] Virtual pre-polishing results: The expected roughness reduction, expected polishing time and process effect calculated based on the material removal model in the digital twin environment before printing.

[0181] After SLM printing and support removal are completed, the actual surface roughness distribution of each zone is obtained using industrial CT or endoscopy. The actual roughness is compared with the corresponding target threshold, and the region type and virtual pre-polishing results in the pre-printing process planning are read.

[0182] The treatment priority is determined based on the roughness deviation, the difficulty of polishing the area, and the risk of media discharge. For complex internal flow channels that contain multiple area types, a progressive step-by-step approach is preferred, from micro-blind end area to curved and difficult-to-reach area to straight and easy-to-reach area, to gradually reduce the roughness.

[0183] Abrasive flow polishing: A process in which a flowing medium containing abrasive particles is pressurized through an internal flow channel, causing the abrasive particles to perform numerous micro-cuts on the wall surface under fluid drive, thereby reducing roughness. Liquid metal-driven abrasive flow is preferred for direct-access areas.

[0184] Ultrasonic-magnetic field coupled abrasive flow polishing: This method simultaneously applies ultrasonic vibration and a magnetic field to an abrasive flow, giving the abrasive grains additional velocity or force, and is used to enhance material removal in hard-to-reach areas of bending. It is called ultrasonic-magnetic field coupled three-phase abrasive flow polishing.

[0185] Electrochemical polishing: a polishing process that removes microscopic protrusions on the surface through electrochemical anodic dissolution or corresponding electrochemical mechanisms; for micro-blind end regions, overpotential electrochemical polishing or electrolyte plasma polishing is preferred.

[0186] Liquid metal drive: A method in which abrasive particles are propelled through a flow channel by using liquid metal or a working medium containing liquid metal as the drive / carrier system.

[0187] Three-phase abrasive flow: an abrasive flow system in which multiple phases, including a carrier phase, an abrasive phase, and a magnetic medium / particle phase under the influence of a magnetic field, participate.

[0188] For the directly accessible area, polishing is performed using a liquid metal-driven abrasive flow, with a preferred polishing time of 5–15 minutes. After polishing, high-pressure gas is introduced for alternating forward and reverse purging at least three times.

[0189] For the difficult-to-reach curved areas, ultrasonic-magnetic coupled three-phase abrasive flow polishing is employed. The ultrasonic transducer generates high-frequency vibrations of 20–50 kHz, producing pulsating pressure and cavitation effects. Simultaneously, a gradient magnetic field acts on the magnetic abrasive grains, enhancing material removal from the curved area walls under the synergistic effect of the ultrasonic and magnetic fields. The flow field is monitored in real-time using particle image velocimetry, and closed-loop optimization is performed by a computer numerical control system, yielding an optimal polishing time of 15–30 minutes.

[0190] For the micro-dead end region, overpotential electrochemical polishing or electrolyte plasma polishing is used. A conformal flexible electrode can be used, with the distance between the electrode and the channel wall maintained at approximately 0.5–1 mm.

[0191] Polishing time: The effective processing time for the corresponding area from the start of polishing to inspection or cessation.

[0192] Polishing pressure: The pressure parameter that drives the polishing medium through the internal flow channel or is applied to the process system, affecting the medium flow rate and the intensity of the wall action.

[0193] Polishing power: The energy output parameter of ultrasonic, electrochemical, plasma or related polishing equipment. The specific physical meaning is determined by the selected process.

[0194] Adaptive termination: After each inspection, the polishing of the area is continued or stopped based on the comparison between the current actual roughness and the target threshold, thereby avoiding under-polishing or over-polishing caused by a fixed time.

[0195] Real-time / periodic detection: The real-time surface roughness referred to in this invention can be obtained by continuous online sensing or by periodic detection immediately after short-term polishing, as long as a closed-loop comparison and parameter adjustment can be formed.

[0196] Before physical polishing, it is preferable to conduct virtual trial polishing in a digital twin environment. The material removal trend of different regions under candidate parameters is predicted by using a model of single abrasive cutting, ultrasonic-magnetic equivalent abrasive velocity and roughness change over time.

[0197] During actual polishing, in-situ or equivalent cycle testing is performed after each polishing step. Polishing of that area is stopped immediately when the actual roughness reaches the target threshold; if the target threshold is not reached, at least one of the polishing time, polishing pressure, and polishing power is adjusted according to the material removal model and historical database. If the target is not met, polishing can continue for 1-2 minutes and then be tested again, repeating the process until the target is met.

[0198] The final polishing parameters are automatically transmitted back to the digital twin model via data link and stored in the process database. After all polishing is completed, industrial CT is used for final non-destructive testing to confirm that the surface roughness of the entire flow channel meets the standards and that there is no abrasive or machining media residue.

[0199] The depth of cut of a single abrasive grain in virtual test polishing is expressed as:

[0200] in, The cutting depth or equivalent material removal depth of a single abrasive grain on the inner flow channel wall. The energy conversion coefficient mainly depends on the abrasive grain shape, effective cutting edge height, and friction coefficient between the abrasive grain and the wall surface; its value ranges from 0.02 to 0.15. : Quality of a single abrasive grain. : The velocity of abrasive particles in the basic flow field. Vickers hardness of the workpiece material. : The contact area between the abrasive grains and the wall surface. For ultrasonic-magnetic coupled abrasive flow, ultrasonic vibration introduces an additional velocity:

[0201] in, , Additional velocity of abrasive particles introduced by ultrasonic vibration. : Ultrasonic amplitude. : Ultrasonic vibration angular frequency. t: Time. : Ultrasonic frequency, ranging from 20 to 50 kHz. π: Pi (circular diameter). The acceleration of magnetic abrasive particles by a magnetic force can be expressed as... The magnetic force experienced by magnetic abrasive particles in a gradient magnetic field is expressed as:

[0202] in, The abrasive acceleration is generated by the magnetic field force. This represents the magnetic force experienced by magnetic abrasive particles in a gradient magnetic field. For abrasive grain quality; The permeability of free space is approximately 4π × 10⁻⁶. -7 ; The magnetic susceptibility range for ferromagnetic abrasives is approximately 0.1 to 10. H represents the volume of a single abrasive grain; H represents the magnetic field strength. The gradient of the square of the magnetic field strength is given, with a magnitude of approximately 10. 9 ~10¹ 0 .

[0203] The contribution of the magnetic field to the abrasive particle velocity is expressed as: ,in The additional velocity introduced by the magnetic field. The duration of the magnetic field's influence.

[0204] The equivalent abrasive velocity is obtained by superimposing the base abrasive velocity, the ultrasonic additional velocity, and the magnetic field additional velocity, as shown in the formula:

[0205] in, The equivalent abrasive particle velocity under ultrasonic-magnetic field coupling is denoted as . The velocity of the abrasive particles in the basic flow field is denoted as . Additional velocity introduced by ultrasonic vibration. The additional velocity introduced by the magnetic field. The change in surface roughness with polishing time can be expressed as the cumulative effective abrasive cutting amount:

[0206] in, Let t be the surface roughness at polishing time t. This represents the initial surface roughness at the start of polishing. The effective cutting abrasive grain number. This refers to the number of abrasive grains that effectively participate in cutting within the evaluation range. For the first The cutting depth or equivalent material removal contribution of each effective abrasive grain within time t. The length is used to evaluate surface roughness.

[0207] The roughness evolution of the directly accessible region under conditions of no ultrasound and no magnetic field coupling can be expressed as:

[0208] in, For the first abrasive grain quality, For the first The square of the velocity of each abrasive grain in the basic flow field For the first The contact area between the abrasive grains and the wall surface.

[0209] For the difficult-to-reach bending region, the equivalent abrasive velocity, when substituted into the material removal model, can be expressed as:

[0210] in, Let be the surface roughness at time t under ultrasonic-magnetic coupled abrasive flow. The ratio of the roughness decrease rates for the two types of regions is further given:

[0211] In the formula, The rate of change of surface roughness over time under ultrasonic-magnetic field coupled abrasive flow conditions; This represents the rate of change of surface roughness over time under basic abrasive flow conditions. This relationship is used to illustrate that, under otherwise equal conditions, multi-energy field coupling can theoretically improve the material removal rate in the hard-to-reach bending region, providing a theoretical basis for zoned differentiated polishing.

[0212] S7. Obtain monitoring data, in-situ control results, and polishing results during the SLM printing process to obtain processing data; calculate the prediction deviation based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel; update the model parameters of the surface roughness theoretical prediction model based on the prediction deviation, and update the machine learning correction model based on the processing data to obtain the updated digital twin model of the inner flow channel. Monitoring data includes data collected during the printing process such as the state of the molten pool, layer-by-layer contours, layer numbers, scan paths, and corresponding timestamps.

[0213] Processing data: A data set formed by associating printing process monitoring data, in-situ control results, actual surface roughness after printing, zone polishing parameters and polishing results according to parts, regions and layer numbers.

[0214] Prediction bias: The difference between the actual surface roughness and the corresponding roughness prediction value of the digital twin model under the same region and the same process conditions can be used as the basis for correcting the parameters of the theoretical model and incremental training of the machine learning model.

[0215] Model parameters: Parameters within the theoretical model or material removal model of a digital twin that need to be calibrated using actual data, including scaling factors related to materials and processes. wait.

[0216] Model update: The theoretical model parameters are corrected using new processing data, and new samples are added to the machine learning training set so that the digital twin model's predictions for specific equipment, materials and powder batches gradually approach reality.

[0217] After completing the entire printing and polishing process for each part, the molten pool data, contour layer scan data, in-situ control parameters, and polishing process data monitored during the printing stage are fed back to the digital twin model. A corresponding record is established for each internal flow channel region, recording the predicted roughness, actual roughness, process parameters, and monitoring status.

[0218] The theoretical model parameters can be quantified for uncertainty and updated incrementally using a Bayesian inference framework; for ease of engineering implementation, a first-order Bayesian approximation can also be used for parameter updates.

[0219] While updating the theoretical parameters, new samples formed from the current parts are added to the machine learning training set to perform incremental training on the roughness residual correction model in order to capture influencing factors not covered by the theoretical model, such as the printing difference coefficient between different devices and the powder batch difference coefficient.

[0220] The Bayesian update relation for the theoretical model parameters is as follows:

[0221] in, : To be updated Model parameters, such as scaling factors related to spheroidization, powder adhesion, or material removal. : in this formula This represents the newly acquired set of processing data, which has a different meaning from the flow channel diameter D in the geometric parameters. :parameter Prior distribution before the introduction of this batch of new data. Given model parameters The likelihood function of the processed data D observed under the given conditions. : Marginal probability or normalization term of the processed data D The model parameters are set to values When the conditional probability of D is observed, a first-order approximation can be used for incremental parameter updates to facilitate real-time engineering implementation.

[0222] in, For the first The updated values ​​of the model parameters For the first The values ​​of each model parameter before the update The learning rate controls the magnitude of a single parameter update. For surface roughness, For surface roughness Relative to parameters Sensitivity This represents the deviation between the actual surface roughness / actual polishing effect and the digital twin prediction or virtual test polishing result.

[0223] S8. Obtain the 3D model of the part to be formed, determine the printing parameters and polishing process corresponding to the part to be formed based on the 3D model of the part to be formed and the updated internal flow channel digital twin model, and determine the corresponding local remelting parameters and scanning strategy based on the updated internal flow channel digital twin model during the SLM printing process of the part to be formed.

[0224] Parts to be manufactured: New parts that need to be actually manufactured using SLM after the aforementioned training, processing data feedback and digital twin model update are completed.

[0225] Updated internal flow channel digital twin model: The digital twin model has been completed by adjusting the theoretical model parameters and incrementally training the machine learning model using historical and new processing data.

[0226] Actual printing parameters: The updated digital twin model outputs parameters such as laser power, scanning speed, scanning spacing, and layer thickness for each area of ​​the part to be formed and is used for actual SLM printing.

[0227] Actual polishing process: The updated digital twin model determines the polishing route for actual post-processing based on the partitioning of the part to be formed and historical processing data.

[0228] Actual remelting parameters: The remelting control parameters are generated by the model based on similar geometric regions and historical anomaly repair data, and are confirmed or adjusted in conjunction with real-time contour deviations when the part to be formed is actually printed.

[0229] Actual scanning strategy: The model gives a scanning direction strategy based on the features of the area of ​​the part to be formed and the historical learning results, and further determines the direction deflection of adjacent layers by combining the real-time roughness prediction results during the printing process.

[0230] The geometric feature extraction process for the part to be formed is repeated to form a data representation consistent with the training part, which is then mapped to the updated digital twin model. Based on the updated parameters and process database, the model completes flow field calculation, roughness prediction, region division, and process matching, and outputs region-level printing parameters and subsequent polishing routes for actual printing.

[0231] For remelting parameters and scanning strategies, the model can first form initial recommended values ​​based on historical similar regions; after the formed part enters the actual SLM printing, the initial recommended values ​​are then corrected in a closed loop according to the real-time melt pool state, actual contour deviation and real-time roughness prediction results, thereby forming the actual remelting parameters and actual scanning strategies for the part.

[0232] After the parts are processed, the roughness of the inner flow channel is tested and the fluid resistance is tested to verify whether the surface roughness of the entire flow channel meets the design requirements, and the final roughness value of each area is recorded. This quality verification data is then fed back as a sample for the next round of digital twin update.

[0233] A system for coordinated control of surface quality in an internal flow channel, comprising: The geometric feature extraction module is used to obtain the 3D model of the training part and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part. The digital twin construction and prediction module is used to construct a digital twin model of the inner flow channel based on the 3D model of the training part and the geometric feature parameters of the inner flow channel. The digital twin model of the inner flow channel includes a parametric geometric model of the inner flow channel, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. Based on the flow field simulation model, the flow state of the polishing medium in the inner flow channel is simulated to obtain the flow field characteristics. Based on the surface roughness theoretical prediction model, the surface roughness of the inner flow channel is predicted to obtain the roughness prediction result before printing. The pre-printing process planning module is used to partition the inner flow channel according to the flow field characteristics, pre-printing roughness prediction results, and inner flow channel geometric feature parameters to obtain partitioning results; and to match the printing parameters and polishing processes of the corresponding regions according to the partitioning results to obtain the pre-printing process planning results. The printing process roughness prediction module is used to acquire the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area based on the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model to obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the printing process roughness prediction result based on the theoretical roughness prediction value and the residual correction value. The in-situ control module for the printing process is used to acquire the actual contour data of the printed layer, calculate the contour deviation based on the actual contour data and the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel, determine the local remelting parameters based on the contour deviation, determine the scanning strategy based on the roughness prediction results of the printing process, and control the SLM printing process based on the local remelting parameters and the scanning strategy to obtain the in-situ control results of the printing process. The partitioned polishing control module is used to obtain the actual surface roughness of each region of the inner flow channel after SLM printing. Based on the actual surface roughness and the pre-printing process planning results, the corresponding polishing process for each region is determined to obtain the partitioned polishing scheme. The inner flow channel is polished differently according to the partitioned polishing scheme, and the surface roughness detection results during the polishing process are obtained. The polishing parameters are adjusted according to the comparison between the surface roughness detection results and the target roughness threshold of the corresponding region until the corresponding region reaches the target roughness threshold, and the polishing processing result is obtained. The digital twin model update module is used to acquire monitoring data, in-situ control results of the printing process, and polishing results during the SLM printing process to obtain processing data; calculate the prediction deviation based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel; update the model parameters of the surface roughness theoretical prediction model based on the prediction deviation, and update the machine learning correction model based on the processing data to obtain the updated digital twin model of the inner flow channel. The processing parameter determination module is used to acquire the three-dimensional model of the part to be formed, determine the printing parameters and polishing process corresponding to the part to be formed based on the three-dimensional model of the part to be formed and the updated internal flow channel digital twin model, and determine the corresponding local remelting parameters and scanning strategy based on the updated internal flow channel digital twin model during the SLM printing process of the part to be formed.

[0234] This application also discloses a terminal device, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it employs an internal flow channel surface quality collaborative control method.

[0235] The terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server. The terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0236] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0237] The memory can be an internal storage unit of the terminal device, such as a hard disk or RAM of the terminal device, or an external storage device of the terminal device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the terminal device. Furthermore, the memory can be a combination of internal storage units and external storage devices of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0238] In this terminal device, the method for coordinated control of the surface quality of the internal flow channel in the above embodiments is stored in the memory of the terminal device and is loaded and executed on the processor of the terminal device for convenient use.

[0239] This application also discloses a computer-readable storage medium, which stores a computer program, wherein when the computer program is executed by a processor, it employs the internal flow channel surface quality collaborative control method described in the above embodiments.

[0240] The computer program can be stored in a computer-readable medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or certain middleware. The computer-readable medium includes any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above-mentioned components.

[0241] The method for coordinated control of internal flow channel surface quality in the above embodiments is stored in the computer-readable storage medium and loaded and executed on the processor to facilitate the storage and application of the above method.

[0242] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0243] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for synergistic control of the surface quality of an internal flow channel, characterized in that, include: Obtain a 3D model of the training part, and extract the geometric feature parameters of the inner flow channel from the 3D model of the training part; A digital twin model of the internal flow channel is constructed based on the three-dimensional model of the training part and the geometric feature parameters of the internal flow channel. The digital twin model of the internal flow channel includes a parametric geometric model of the internal flow channel, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. Based on the aforementioned flow field simulation model, the flow state of the polishing medium in the inner flow channel is simulated to obtain the flow field characteristics. The surface roughness of the inner channel is predicted based on the surface roughness theoretical prediction model, and the roughness prediction result before printing is obtained. Based on the flow field characteristics, the roughness prediction results before printing, and the geometric characteristic parameters of the inner flow channel, the inner flow channel is divided into partitions to obtain the partitioning results. Based on the partitioning results, the printing parameters and polishing processes of the corresponding regions are matched to obtain the pre-printing process planning results; Acquire the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area according to the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model to obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the roughness prediction result of the printing process according to the theoretical roughness prediction value and the residual correction value. The actual contour data of the printed layer is obtained, and the contour deviation is calculated based on the actual contour data and the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel. The local remelting parameter is determined based on the contour deviation, the scanning strategy is determined based on the roughness prediction result of the printing process, and the SLM printing process is controlled based on the local remelting parameter and the scanning strategy to obtain the in-situ control result of the printing process. After SLM printing is completed, the actual surface roughness of each region of the inner flow channel is obtained. Based on the actual surface roughness and the pre-printing process planning results, the polishing process corresponding to each region is determined to obtain a partitioned polishing scheme. The inner flow channel is differentially polished according to the partitioned polishing scheme, and the surface roughness detection results during the polishing process are obtained; the polishing parameters are adjusted according to the comparison results between the surface roughness detection results and the target roughness threshold of the corresponding area until the corresponding area reaches the target roughness threshold, and the polishing processing result is obtained. The process involves acquiring monitoring data during SLM printing, in-situ control results of the printing process, and polishing results to obtain processing data; calculating the prediction deviation based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel; updating the model parameters of the theoretical surface roughness prediction model based on the prediction deviation, and updating the machine learning correction model based on the processing data to obtain the updated digital twin model of the inner flow channel. A three-dimensional model of the part to be formed is obtained. Based on the three-dimensional model of the part to be formed and the updated internal flow channel digital twin model, the printing parameters and polishing process corresponding to the part to be formed are determined. During the SLM printing process of the part to be formed, the corresponding local remelting parameters and scanning strategy are determined based on the updated internal flow channel digital twin model.

2. The method for coordinated control of internal flow channel surface quality according to claim 1, characterized in that, The process of obtaining a 3D model of the training part and extracting the geometric feature parameters of the inner flow channel from the 3D model of the training part includes: Obtain the three-dimensional model of the training part, and identify the spatial orientation and cross-sectional topology of the internal flow channel in the three-dimensional model of the training part; Based on the spatial orientation and cross-sectional topology of the inner flow channel, the inner flow channel diameter, radius of curvature, bending angle, depth-to-diameter ratio, and cross-sectional change rate are extracted to obtain the geometric characteristic parameters of the inner flow channel.

3. The method for coordinated control of internal flow channel surface quality according to claim 1, characterized in that, The process of constructing a digital twin model of the inner flow channel based on the three-dimensional model of the training part and the geometric feature parameters of the inner flow channel, obtaining flow field features based on the flow field simulation model, and obtaining the roughness prediction result before printing based on the surface roughness theoretical prediction model includes: Based on the three-dimensional model of the training part and the geometric feature parameters of the inner flow channel, a parametric geometric model of the inner flow channel is established. The parametric geometric model of the inner flow channel is used to characterize the spatial orientation, cross-sectional topology and theoretical contour data corresponding to each position of the inner flow channel. A flow field simulation model is constructed based on the parametric geometric model of the inner flow channel. The flow field simulation model is used to perform computational fluid dynamics simulation on the flow state of the polishing medium in the inner flow channel to obtain the flow velocity, turbulent kinetic energy and wall shear stress in each region of the inner flow channel, and thus obtain the flow field characteristics. The surface roughness theoretical prediction model is established based on the step effect, spheroidization effect and powder adhesion effect that generate the surface roughness of the inner flow channel during the SLM forming process. The theoretical roughness of each region of the inner flow channel is determined according to the surface roughness theoretical prediction model, and the roughness prediction result before printing is obtained. A machine learning correction model is set up to perform residual correction on the theoretical roughness prediction value output by the surface roughness theoretical prediction model based on the melt pool state data during the SLM printing process. The parameterized geometric model of the inner flow channel, the flow field simulation model, the surface roughness theoretical prediction model, and the machine learning correction model are associated, and historical printing and polishing data are imported to construct the digital twin model of the inner flow channel.

4. The method for coordinated control of internal flow channel surface quality according to claim 1, characterized in that, The inner flow channel is partitioned according to the flow field characteristics, the roughness prediction results before printing, and the geometric characteristic parameters of the inner flow channel to obtain the partitioning results; Based on the partitioning results, the printing parameters and polishing processes of the corresponding regions are matched to obtain the pre-printing process planning results, including: Based on the flow field characteristics, the roughness prediction results before printing, and the geometric characteristic parameters of the inner flow channel, the inner flow channel is divided into a straight and easily accessible area, a curved and difficult-to-access area, and a micro-blind end area, thus obtaining the partitioning results; Among them, the straight-through easily accessible zone is determined according to the flow state of the polishing medium in the inner channel, the curved difficult-to-access zone is determined according to the relationship between the inner channel curvature radius and the inner channel diameter, and the micro-blind end zone is determined according to the inner channel diameter and the inner channel end structure. For different regions, with the goal of reducing the surface roughness of the corresponding region after printing, at least one corresponding printing parameter is determined from laser power, scanning speed, scanning spacing and layer thickness; Based on the geometric characteristics of the internal flow channels and the accessibility of the polishing medium in different regions, the corresponding polishing process is matched, and the printing parameters, polishing process and region information corresponding to each region are associated to obtain the pre-printing process planning result.

5. The method for coordinated control of internal flow channel surface quality according to claim 1, characterized in that, The process of acquiring molten pool state data during SLM printing and obtaining printing process roughness prediction results based on the surface roughness theoretical prediction model and machine learning correction model includes: High-speed cameras or photoelectric sensors are used to acquire the molten pool radiation signals corresponding to each scanning path during the SLM printing process; The molten pool temperature and size are determined based on the molten pool radiation signal, and the molten pool state data is obtained. Based on the printing parameters and internal flow channel geometry of the current printing area, the theoretical roughness prediction value of the current printing area is determined by the surface roughness theoretical prediction model. The molten pool state data is input into the machine learning correction model, and the machine learning correction model determines the residual correction value corresponding to the theoretical roughness prediction value based on the molten pool state data. The theoretical roughness prediction value is fused with the residual correction value to determine the surface roughness of the inner flow channel in the current printing area, thus obtaining the roughness prediction result of the printing process.

6. The method for coordinated control of internal flow channel surface quality according to claim 1, characterized in that, The process involves acquiring the actual contour data of the printed layer, calculating the contour deviation based on the actual contour data and the theoretical contour data, and adjusting the SLM printing process based on the contour deviation and the roughness prediction results of the printing process to obtain in-situ control results for the printing process, including: After the printing layer is completed, the inner flow channel contour corresponding to the printing layer is scanned in situ to obtain the actual contour data. Obtain the theoretical contour data corresponding to the printing layer in the parameterized geometric model of the inner flow channel, and determine the contour deviation corresponding to each contour position based on the position difference between the actual contour data and the theoretical contour data; The contour deviation threshold is determined based on the layer thickness of the printed layer. The contour deviation is compared with the contour deviation threshold to determine the abnormal position where the contour deviation exceeds the contour deviation threshold. Based on the contour deviation corresponding to the abnormal location, the local remelting power and the number of local remelting cycles are determined to obtain the local remelting parameters. The roughness prediction result of the printing process is compared with the target roughness threshold. When the roughness prediction result of the printing process does not meet the target roughness threshold, the scanning direction deflection angle of the adjacent printing layer is determined according to the surface roughness theoretical prediction model to obtain the scanning strategy. The abnormal location is locally remelted according to the local remelting parameters, and the scanning direction of adjacent printing layers is adjusted according to the scanning strategy to obtain the in-situ control result of the printing process.

7. The method for coordinated control of internal flow channel surface quality according to claim 4, characterized in that, The process involves obtaining the actual surface roughness of each region of the inner flow channel after SLM printing, and performing differentiated polishing based on the pre-printing process planning results to obtain the polishing treatment results, including: After SLM printing is completed, the actual surface roughness of each region of the inner channel is obtained. The actual surface roughness is compared with the target roughness threshold of the corresponding region. The processing priority and corresponding polishing process of each region are determined in combination with the process planning results before printing. For the directly accessible region, a liquid metal-driven abrasive flow polishing process is determined; for the curved and difficult-to-reach region, an ultrasonic and magnetic field coupled abrasive flow polishing process is determined; and for the micro-blind end region, an electrochemical polishing process is determined, thus obtaining the partitioned polishing scheme. Differential polishing is performed on the corresponding area of ​​the inner flow channel according to the partitioned polishing scheme, and the surface roughness detection results of the corresponding area are obtained during the polishing process; The surface roughness detection result is compared with the target roughness threshold of the corresponding area. When the surface roughness detection result does not reach the target roughness threshold, at least one of the polishing parameters, polishing time, polishing pressure and polishing power, is adjusted, and the corresponding area is polished. When the surface roughness detection result reaches the target roughness threshold, polishing of the corresponding area is stopped, and the polishing result is obtained.

8. A system for coordinated control of surface quality in an internal flow channel, characterized in that, include: The geometric feature extraction module is used to acquire a three-dimensional model of the training part and extract the geometric feature parameters of the inner flow channel from the three-dimensional model of the training part. The digital twin construction and prediction module is used to construct an internal flow channel digital twin model based on the three-dimensional model of the training part and the geometric feature parameters of the internal flow channel. The internal flow channel digital twin model includes an internal flow channel parametric geometric model, a flow field simulation model, a surface roughness theoretical prediction model, and a machine learning correction model. Based on the aforementioned flow field simulation model, the flow state of the polishing medium in the inner flow channel is simulated to obtain the flow field characteristics. The surface roughness of the inner channel is predicted based on the surface roughness theoretical prediction model, and the roughness prediction result before printing is obtained. The pre-printing process planning module is used to partition the inner flow channel according to the flow field characteristics, the pre-printing roughness prediction results, and the geometric characteristic parameters of the inner flow channel, and obtain the partitioning results. Based on the partitioning results, the printing parameters and polishing processes of the corresponding regions are matched to obtain the pre-printing process planning results; The printing process roughness prediction module is used to acquire the melt pool state data during the SLM printing process, determine the theoretical roughness prediction value of the corresponding printing area according to the surface roughness theoretical prediction model, input the melt pool state data into the machine learning correction model to obtain the residual correction value corresponding to the theoretical roughness prediction value, and determine the printing process roughness prediction result according to the theoretical roughness prediction value and the residual correction value. The in-situ control module for the printing process is used to acquire the actual contour data of the printing layer, calculate the contour deviation based on the actual contour data and the theoretical contour data corresponding to the parameterized geometric model of the inner flow channel, determine the local remelting parameters based on the contour deviation, determine the scanning strategy based on the roughness prediction result of the printing process, and control the SLM printing process based on the local remelting parameters and the scanning strategy to obtain the in-situ control result of the printing process. The partitioned polishing control module is used to obtain the actual surface roughness of each region of the inner flow channel after SLM printing, and to determine the polishing process corresponding to each region based on the actual surface roughness and the process planning results before printing, so as to obtain the partitioned polishing scheme. The inner flow channel is differentially polished according to the partitioned polishing scheme, and the surface roughness detection results during the polishing process are obtained; the polishing parameters are adjusted according to the comparison results between the surface roughness detection results and the target roughness threshold of the corresponding area until the corresponding area reaches the target roughness threshold, and the polishing processing result is obtained. The digital twin model update module is used to acquire monitoring data, in-situ control results of the printing process, and polishing results during the SLM printing process to obtain processing data; calculate the prediction deviation based on the actual surface roughness in the processing data and the roughness prediction results corresponding to the digital twin model of the inner flow channel; update the model parameters of the theoretical surface roughness prediction model based on the prediction deviation, and update the machine learning correction model based on the processing data to obtain the updated digital twin model of the inner flow channel. The processing parameter determination module is used to acquire the three-dimensional model of the part to be formed, determine the printing parameters and polishing process corresponding to the part to be formed based on the three-dimensional model of the part to be formed and the updated internal flow channel digital twin model, and determine the corresponding local remelting parameters and scanning strategy based on the updated internal flow channel digital twin model during the SLM printing process of the part to be formed.

9. A terminal device, comprising a memory and a processor, characterized in that, The memory stores a computer program that can run on a processor, and when the processor loads and executes the computer program, it employs the method described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is loaded and executed by the processor, it employs the method described in any one of claims 1 to 7.

Images