Intelligent manufacturing method for high-end numerical control machine tool base large piece sand casting core bone

Abstract

The application discloses a high-end numerical control machine tool basic large piece sand mold casting core bone intelligent manufacturing method, and belongs to the technical field of deep integration of high-end equipment manufacturing and intelligent casting. The method balances the weight and strength of the core bone by constructing a sand core digital model and a multi-physical field constraint system, and adopting an improved genetic algorithm and finite element coupling for topology optimization. The method identifies weak areas based on a U-Net deep learning network, and adaptively enhances the bionic structure or heat dissipation channel. The method realizes accurate production through digital twin technology linkage of a robot welding and an additive manufacturing process. Finally, a full life cycle database is constructed, and a reinforcement learning model is used to realize continuous optimization. The application reduces the weight of the core bone, controls the sand core deformation within 0.1mm, shortens the design cycle, improves the casting qualification rate, and can meet the precision casting requirements of high-end numerical control machine tool basic large pieces, and has significant technical and economic value.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of deep integration of high-end equipment manufacturing and intelligent casting, and particularly relates to an intelligent manufacturing method for a core bone of a high-end numerical control machine tool base large piece. BACKGROUND

[0002] As the core bearing component of a machine tool, the performance of a high-end numerical control machine tool base large piece (such as a five-axis linkage machining center and a precision horizontal lathe) directly determines the dynamic precision, structural rigidity and service life of the whole machine. The casting quality of such a base large piece has a decisive influence on the machining precision and long-term stability of the machine tool. The machining precision needs to be strictly controlled at the level of 0.001 mm to meet the micron-level machining requirements of high-end fields such as aerospace and precision molds; at the same time, it is required to have a precision retention of more than 10 years to ensure that the performance attenuation during long-term service is controlled within the minimum range. As the "skeletal system" of the machine tool, the base large piece is characterized by structural complexity, strict performance requirements and super-large size, which together constitute the core challenge of the casting process.

[0003] In terms of structural complexity, the high-end numerical control machine tool base large piece contains multiple complex structural units such as guide rail surfaces, internal cavity rib plates and weight reduction holes. The guide rail surface, as the reference for the motion precision of the machine tool, requires a flatness and straightness of micron level; the internal cavity rib plate needs to be arranged reasonably to achieve rigid reinforcement and weight balance, with a rib plate thickness of usually between 8 mm and 20 mm and a dense distribution; the weight reduction hole needs to achieve lightweight without affecting the structural strength, with the hole diameter and hole spacing accurately matched with the stress distribution. These complex structures require the sand core to accurately match the features of the cavity, and the dimensional error of the sand core needs to be controlled within 0.1 mm, otherwise it will directly lead to uneven machining allowance or size out-of-tolerance of the casting.

[0004] The current core bone design of the high-end numerical control machine tool base large piece still stays at the traditional experience level, and generally adopts the solid round steel welding or cast steel integral casting process. The design process excessively relies on the experience formula of engineers, such as the simple conversion of "core bone diameter = sand core thickness x 0.15", which leads to the widespread existence of "overdesign". The traditional core bone weight of the column sand core of a certain type of machine tool reaches 1.2 tons, and the actual mechanical analysis shows that the material utilization rate is less than 40%, a large amount of redundant materials not only increase the manufacturing cost, but also lead to a local stress concentration of the sand core exceeding 20 MPa, which in turn increases the risk of sand core deformation. This design mode is difficult to balance the contradiction between lightweight and high strength, and has become a bottleneck restricting the improvement of casting quality. SUMMARY

[0005] The application aims at providing a high-end numerical control machine tool basic large piece sand mold casting core bone intelligent manufacturing method to solve the above problems, and improve the problems of difficult balance between core bone weight and support strength, excessive reliance on engineer's experience formula, and large amount of redundant materials and increased manufacturing cost.

[0006] The technical scheme adopted by the application is as follows: a high-end numerical control machine tool basic large piece sand mold casting core bone intelligent manufacturing method, the method comprising the following steps:

[0007] Step S100: constructing a digital basic model and constraint system of core bone design, generating a three-dimensional model of molding sand through reverse engineering software based on a three-dimensional model of a casting, extracting molding sand characteristics and performing partitioning according to an algorithm;

[0008] Step S200: establishing a multi-objective optimization function and constructing a core bone feature database, adopting an improved genetic algorithm fused with a simulated annealing operator, and realizing bidirectional data interaction with a finite element simulation tool;

[0009] Step S300: automatically strengthening a core bone weak area based on a deep learning model;

[0010] Step S400: intelligent manufacturing and precision control based on digital twinning, adaptively selecting a manufacturing process according to core bone structure characteristics, adaptively selecting a robot welding system or an additive manufacturing process, and realizing quality closed loop through digital twinning simulation and online detection;

[0011] Step S500: continuous optimization driven by full life cycle data, constructing a design-manufacturing-service database, and adopting a reinforcement learning model to realize automatic parameter pushing and iteration.

[0012] Further, step S100 comprises the following steps:

[0013] Step S101: defining a multidimensional constraint database of mechanics, thermodynamics and manufacturing process, and establishing a parameterized digital model;

[0014] Step S102: inputting casting three-dimensional model data into the parameterized digital model, inputting temperature field T, velocity field V, temperature gradient , solid phase rate , residual stress at the time of solidification, and inputting pouring gate coordinates G, chill position C and parting surface P according to simulation results of a simulated pouring system;

[0015] Wherein, temperature field T(x, t), velocity field V(x, t), temperature gradient , solid phase rate , residual stress G, C, P are the coordinates of the gating system, the chill position and the parting plane, respectively, which are set when designing the gating system.

[0016] T(x, t) is the temperature value at position x and time t; V(x, t) is the velocity vector at position x and time t;

[0017] Step S103: Align the casting three-dimensional model and the data of the simulated gating system simulation to the same set of voxel grids;

[0018] Uniform voxelization of the casting three-dimensional model with 0.5mm:

[0019] ;

[0020] Sampling the data of the simulated simulation to the same grid:

[0021] , ;

[0022] Generating a sand mold grid:

[0023] ;

[0024] wherein, , , correspond to the first , , voxels in the X, Y, Z directions, respectively; is the inflation radius, is the expansion of the casting boundary by 4mm; is a three-dimensional (0, 1) array, is the voxel inside the casting, is empty; is the sand mold grid.

[0025] Further, according to the algorithm, the sand characteristics are extracted and partitioned, including high stress area, conventional support area and lightweight priority area;

[0026] The high stress area HSZ includes the hot spot area and the cantilever end area ;

[0027] Three-dimensional area extraction is performed on the hot spot area and the cantilever end area in the high stress area HSZ, and the triangulation surface is output after merging processing;

[0028] ;

[0029] ;

[0030] ;

[0031] The conditions for hot spot extraction are local solidification time greater than or equal to 90 seconds, temperature gradient not exceeding 5℃ within 1mm distance, distance from pouring gate less than 25mm; the conditions for cantilever end extraction are cantilever length greater than or equal to 15mm and cantilever angle greater than or equal to 45°, and maximum stress greater than 1.5 times the yield stress.

[0032] wherein t is the local solidification time, is the temperature gradient, is the distance from the pouring gate, L is the cantilever length, is the cantilever angle, is the yield stress, - means voxel mask, which is a 0-1 array with the same size as the three-dimensional grid, 1 means that the voxel is selected (satisfies the condition), 0 means not selected (does not satisfy the condition); means inflation, which refers to expanding the region after merging and outward by 3mm to eliminate small gaps and ensure the continuity and integrity of the high stress area.

[0033] The conventional support zone RSZ includes a planar or shallow cavity region, the planar region is detected and the local minimum thickness of the shallow cavity is calculated;

[0034] Planar detection: the outer surface is calculated by Region Growing and RANSAC algorithm, the normal change and the area 400 of the plane is recorded as ; ;

[0035] wherein Region Growing is region growing, which aggregates adjacent triangular facets with a normal angle of 5° into an initial planar region to obtain a rough patch; RANSAC (Random Sample Consensus) is a random sample consensus that can fit the optimal plane equation and output accurate plane parameters and inlier set; the normal variable refers to the angle between the normal of the triangular facet and the normal of the fitted plane, 5° is considered as obvious warping or circular surface and cannot be regarded as a usable plane, which needs to ensure that the plane is flat enough; the area refers to the actual surface area in three-dimensional space, which needs to ensure that the plane is large enough.

[0036] Shallow cavity detection:

[0037]

[0038] If the cavity is shallow, mark it as shallow cavity region, otherwise not;

[0039] Merge:

[0040] Where, is the local minimum thickness, is the plane detected by the plane detection algorithm;

[0041] Extract data in lightweight priority zone (LPZ):

[0042]

[0043] Where, V is the volume, is the average stress, is the yield stress, is the distance to the high stress region.

[0044] Further, the multi-objective optimization function includes:

[0045] Construct a weight minimization objective function,

[0046] Construct a stiffness maximization objective function,

[0047] Construct a core deformation minimization objective function,

[0048] Where, is the main support beam section width, is the main support beam section height, is the thickness, is the weight reduction hole diameter, is the core bone material density, is the core bone volume, is the material elastic modulus, is the cross-sectional moment of inertia, is the core deformation of each region, is the region weight.

[0049] Further, step S200 includes the following steps:

[0050] Step S201: population initialization, randomly generate 50 groups of core bone structure parameters, including main support beam section width W, main support beam section height H, reinforcement distribution spacing S and weight reduction hole diameter d, to construct the initial population;

[0051] ​​​​​​Step S202: Apply constraints to each individual model, and perform static structural analysis and thermal-structural coupling analysis; output key indicators: maximum stress , maximum deformation , and core bone weight ;

[0052] Step S203: Define fitness function: , where , is the initial design value, =0.3, =0.4, =0.3 are weight coefficients; select individuals with ≥0.8 to enter the next generation, and eliminate individuals that do not meet the constraints;

[0053] Step S204: Genetic operation adopts tournament selection method to select 30% optimal individuals from the parent generation; crossover operator: single-point crossover for main support beam parameters; mutation operator: introduce simulated annealing idea, randomly mutate the spacing between the ribs, and the probability of accepting inferior solutions decreases with the number of iterations;

[0054] Step S205: When the optimal fitness value of the last 5 generations fluctuates less than 1%, output the optimal topology structure.

[0055] Further, step S300 includes the following steps:

[0056] Step S301: Build a database to collect historical core bone failure cases, including stress distribution cloud map, temperature field data, and failure location coordinates, and expand the sample size through data augmentation;

[0057] Step S302: Use an improved U-Net deep learning network to input the stress field / temperature field tensor of finite element analysis, and output a probability map of weak areas;

[0058] Step S303: Use Adam optimizer for training and optimization.

[0059] Further, for the identified weak areas of the core bone, an enhancement scheme is automatically generated, which includes static stress concentration areas, thermal deformation sensitive areas, and lightweight areas, and the enhancement scheme for each weak area of the core bone includes the following:

[0060] Static stress concentration area: use bionic topology structure to increase sectional moment of inertia, adaptively adjust grid size according to stress gradient, and set round corner transition at corners to reduce stress concentration coefficient;

[0061] Thermal deformation sensitive area: design hollow cavity structure with spiral-shaped heat dissipation channels, the channel direction is consistent with the direction of metal liquid flow, and the convection heat dissipation is enhanced;

[0062] Lightweight area, open array type weight reduction hole, hole spacing is 2-3 times of hole diameter.

[0063] Further, step S400 includes the following:

[0064] For the core of the simple structure of the truss, a robot automatic welding system is adopted: the robot automatic welding system generates a welding seam track based on a three-dimensional model, adopts an A* algorithm to optimize the robot path; welding parameters are set, and welding deformation is simulated through digital twinning; a laser tracking sensor is used to correct the welding gun position in real time;

[0065] For complex special-shaped structures, an additive manufacturing process is adopted: the layer thickness is dynamically adjusted according to the curvature change of the core section, and slicing processing is performed; a cladding path is generated, and a spiral line and a parallel line are combined; the temperature field of each layer of cladding is predicted based on the digital twinning model, and the process parameters are dynamically adjusted; laser scanning detection is performed, 2D contour scanning and 3D topography reconstruction are performed on each layer of core section, and a feedback correction strategy is used to correct the contour deviation; the detection data are transmitted to the slicing software in real time to realize quality closed loop, the internal defects of the core are scanned by industrial CT, and the actual size of the core is obtained by three-dimensional laser scanning, which is compared with the digital model to calculate the geometric and position error; if the geometric and position error exceeds the set value, the error area features are automatically extracted and fed back to step S200 for modification of the core structure parameters.

[0066] Further, step S500 includes the following:

[0067] A design-manufacturing-service database is constructed to store data in the design, manufacturing and service stages; a deep deterministic policy gradient (DDPG) algorithm is used to train an intelligent decision model; and the intelligent decision model is used to automatically push optimization parameters for a new design task.

[0068] As described above, due to the adoption of the above technical solutions, the present application has the following beneficial effects:

[0069] 1. The present application balances the weight and strength of the core by constructing a digital core model and a multi-physical field constraint system, and using an improved genetic algorithm and finite element coupling for topology optimization; identifies weak areas based on a U-Net deep learning network, and performs adaptive enhancement of bionic structures or heat dissipation channels; realizes precise production through digital twinning technology linkage of robot welding and additive manufacturing process; and finally constructs a full life cycle database and realizes continuous optimization using a reinforcement learning model.

[0070] 2. The present application realizes lightweight, high strength and efficient manufacturing of the core by constructing an intelligent design-simulation optimization-digital manufacturing-data closed loop full process technical system, and meets the precise casting needs of high-end CNC machine tool basic large parts. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 Flow chart of the method of the present application;

[0072] Figure 2 Schematic diagram of the full life cycle data closed loop of the present application. DETAILED DESCRIPTION

[0073] The present application will be described in detail below with reference to the drawings.

[0074] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described below with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0075] Performance requirements are another core feature of the foundation heavy piece. In the service process, the foundation heavy piece needs to bear the combined action of multiple loads: periodic alternating load (up to 10-50 kN) generated during cutting processing, self-weight load of workpieces and moving parts (large workbench bearing exceeds 10 tons), and thermal deformation stress during machine tool operation. In order to cope with these complex working conditions, the casting material is usually selected from high-strength cast iron (such as HT300, QT500) or low-alloy cast steel, which has a tensile strength of ≥300 MPa, a hardness controlled in the range of 180-250 HBW, and a mechanical property fluctuation strictly controlled within 5%, to ensure the stability of the whole machine performance.

[0076] The size super-large feature further aggravates the casting difficulty. The bed length of the large CNC gantry milling machine can reach more than 10 m, the width exceeds 3 m, and the weight of a single casting reaches 20-50 tons; the corresponding sand core weight exceeds 5 tons, and the volume can reach dozens of cubic meters. This super large size puts high requirements on the support strength of the core bone, which not only needs to bear the bending moment generated by the self-weight of the sand core, but also needs to resist the impact vibration during the sand core handling process. Any support failure may cause the sand core to collapse, resulting in direct losses of tens of thousands of yuan.

[0077] Sand casting, as the mainstream forming process of the foundation heavy piece, relies on the stable forming of the sand core in its core link. The sand core, as the key tooling for forming the inner cavity of the casting, its structural stability directly determines the dimensional accuracy and internal quality of the casting. The core bone, as the "skeleton" of the sand core, bears three core functions: first, the support function, which needs to bear the self-weight load of the sand core in the whole process of sand core modeling, handling, and clamping, to ensure that the sand core does not deform during the transfer process; second, the impact resistance function, which needs to resist the impact pressure of 0.5-2 MPa and the high temperature thermal stress of 1300-1600 ℃ during the metal liquid pouring stage, to prevent the sand core from collapsing under the impact of the metal liquid; and third, the isolation function, which needs to avoid direct contact with the metal liquid through reasonable structural design, to prevent the "sand sticking" defect from affecting the surface quality of the casting.

[0078] The application realizes the lightweight, high-strength and efficient manufacturing of the core bone by constructing an intelligent design-simulation optimization-digital manufacturing-data closed-loop whole-process technical system, and meets the precision casting demand of high-end CNC machine tool basic large parts.

[0079] As shown in Figure 1 , the application proposes a high-end CNC machine tool basic large part sand mold casting core bone intelligent manufacturing method, including the following steps:

[0080] Step S100: constructing a digital basic model and constraint system of core bone design;

[0081] Sand core and casting feature extraction: based on the three-dimensional design model (such as STL format) of high-end CNC machine tool large parts, the topological features of the inner cavity of the casting are extracted through reverse engineering software (such as UG), and the shape contour, key size (length, wall thickness, cavity number) and spatial position relationship of the sand core are determined;

[0082] The sand core is divided into different functional areas by using partition algorithm, which is divided into high stress zone HSZ, regular support zone RSZ and lightweight priority zone LPZ. The multidimensional constraint parameters of mechanics, thermology and manufacturing process are defined, and the parameterized digital model is established. The specific implementation is as follows:

[0083] (1) Input the three-dimensional model data of the casting into the parameterized digital model, according to the simulation results of the simulated gating system, input the temperature field T(x, t), velocity field V(x, t), temperature gradient , solid phase rate , residual stress during solidification, and input the gating coordinate G, chill position C and parting surface P;

[0084] (2) Unified meshing, aligning the model and the simulated data of the simulated gating system to the same set of voxel grid.

[0085] 0.5mm uniform voxelization is performed on : ;

[0086] Among them, is a three-dimensional (0, 1) array, is the voxel inside the casting, is empty;

[0087] Sample the simulated data to the same grid: , ;

[0088] Generate sand mold grid: .

[0089] (3) High stress zone HSZ extracts three-dimensional area that meets the following conditions:

[0090] hot spot zone extracted (where t is the local solidification time, is the temperature gradient, is the distance from the gate):

[0091]

[0092] cantilever tip zone extracted (where L is the overhang length, is the overhang angle, is the yield stress):

[0093]

[0094] merge and post-processing, output the triangulated surface , with the high stress zone scalar field:

[0095] ;

[0096] (4) regular support zone RSZ extracts the three-dimensional region that meets the conditions, RSZ is defined as a plane or shallow cavity region, that is, neither high stress nor large cavity, the main task is to "support" and "heat transfer";

[0097] plane detection method: the outer surface is calculated by Region Growing and RANSAC algorithm, the normal changes and the area 400 plane is recorded as .

[0098] shallow cavity detection method: calculate the local minimum thickness:

[0099] ;

[0100] If , it is marked as a shallow cavity zone; otherwise, it is not a shallow cavity zone;

[0101] merge: ;

[0102] wherein, is the local minimum thickness, is the plane detected by the plane;

[0103] (5) lightweight priority zone LPZ extracts the three-dimensional region that meets the conditions: (where V is the volume, is the average stress, is the yield stress, is the distance from the high stress zone).

[0104] LPZ is defined as the internal space of a sand core that is large in volume, low in stress, and has no critical functional requirements.

[0105] ;

[0106] (6) Establish a multi-dimensional constraint database that includes mechanics, thermodynamics, and manufacturing processes:

[0107] Mechanical constraints: Input core weight ( ), hydrostatic pressure of molten metal ( ,in The density of the molten metal, It is the acceleration due to gravity. (Height of molten metal), compressive strength of core material ( ), yield strength of core material ( (e.g., Q355B steel with a strength of 355MPa).

[0108] Thermal constraints: casting temperature (T casting, approximately 1350℃ for cast iron, approximately 1550℃ for cast steel), thermal conductivity of core material ( Steel with a strength of 45W / (m) K) and allowable thermal deformation (ΔL≤0.1mm / m);

[0109] Manufacturing constraints: Welding process limitations (minimum weld spacing ≥30mm, weld leg height ≥5mm), additive manufacturing parameters (layer thickness 2-5mm, forming angle ≥45°), and processing equipment stroke (e.g., robot welding working radius ≥3m).

[0110] (7) Digital Model Construction

[0111] Construct an initial digital model of the core beam using parametric modeling software (such as SolidWorks), and define the following design variables: ① Main support beam cross-sectional dimensions (width W, height H); ② Stiffener arrangement parameters (spacing S, thickness). ); ③Diameter (d) and distribution density of weight reduction holes.

[0112] Step S200: Establish a multi-objective optimization function and construct a core feature database. Use an improved genetic algorithm that integrates simulated annealing operators to achieve bidirectional data interaction with finite element simulation tools.

[0113] (1) Taking lightweight, high strength, and low deformation as the core, i.e., minimizing weight, maximizing stiffness, and minimizing sand core deformation as the objectives, a multi-objective optimization function is established:

[0114] Construct a weight minimization objective function. ;

[0115] Objective function of maximum stiffness construction, ;

[0116] Objective function of minimum core deformation construction, ;

[0117] Multi-objective optimization function: high stress area = 0.4, regular support area = 0.3, lightweight priority area = 0.3;

[0118] (2) An improved genetic algorithm with a simulated annealing operator is adopted to realize bidirectional data interaction with a finite element simulation tool (such as ANSYS Workbench), and the specific process includes the following steps:

[0119] Step S201: population initialization, 50 groups of core structure parameters are randomly generated, including the main support beam section width W, the main support beam section height H, the reinforcing rib distribution spacing S and the weight reduction hole diameter d, and the initial population is constructed;

[0120] Step S202: constraint conditions (core self-weight load, metal liquid impact pressure, temperature field distribution) are applied to each individual model, and static structure analysis (stress , strain ) and thermal-structure coupling analysis (thermal deformation amount ) are performed; the key indicators: maximum stress , maximum deformation and core weight are outputted;

[0121] Step S203: define the fitness function: , wherein , are the initial design values, = 0.3, = 0.4, = 0.3 are the weight coefficients; select the individuals with ≥ 0.8 to enter the next generation, and eliminate the individuals that do not meet the constraint conditions;

[0122] Step S204: genetic operation adopts the tournament selection method, and 30% of the best individuals are selected from the parent generation; the crossover operator: single-point crossover is performed on the main support beam parameters (crossover probability 0.7); the mutation operator: the simulated annealing idea is introduced, and random mutation is performed on the rib spacing (mutation probability 0.1), and the probability of accepting inferior solutions decreases with the number of iterations;

[0123] Step S205: when the optimal fitness value fluctuates less than 1% for 5 consecutive generations, the optimal topology structure is outputted.

[0124] Step S300: automatically enhancing the core bone weak area based on the deep learning model;

[0125] Step S301: constructing a database, collecting 1000+ historical core bone failure cases, the cases containing stress distribution cloud map, temperature field data, failure position coordinates, expanding the sample number to 5000 through data enhancement (rotation, scaling, noise addition);

[0126] Step S302: using an improved U-Net deep learning network, inputting the stress field / temperature field tensor of finite element analysis, outputting a weak area probability map (pixel value 0-1, greater than 0.8 is determined as a weak area);

[0127] Step S303: using Adam optimizer for training optimization, learning rate 0.001, iteration 100 rounds, verification set accuracy rate reaching 92%, ensuring accurate identification of stress concentration area ( ) and thermal sensitive area ( > 0.05mm).

[0128] For the identified core bone weak area, an automatic enhancement scheme is generated, the core bone weak area including static stress concentration area, thermal deformation sensitive area and lightweight area, and the enhancement scheme for each core bone weak area including the following:

[0129] Static stress concentration area (such as the end of cantilever beam), using bionic topological structure (imitating honeycomb hexagonal grid) to increase cross-sectional moment of inertia, grid size being adaptively adjusted according to stress gradient (grid density increasing by 30% in high stress area), and setting round corner transition (radius R=10-20mm) at corner to reduce stress concentration coefficient (from 1.8 to below 1.2);

[0130] Thermal deformation sensitive area (such as the area close to the inner gate), designing hollow cavity structure, built-in spiral-shaped heat dissipation channel (diameter 8-15mm), channel direction being consistent with metal liquid flow direction to enhance convective heat dissipation; using variable thickness design (thickness gradually changing from 20mm at center to 10mm at edge) to reduce thermal expansion and contraction difference.

[0131] Lightweight area, opening array type weight reduction hole (diameter 15-30mm), hole spacing being 2-3 times of hole diameter to ensure structural stability; using lattice structure (such as body-centered cubic) to realize additive manufacturing.

[0132] Step S400: intelligent manufacturing and precision control based on digital twinning, adaptively selecting manufacturing process according to core bone structure characteristics, adaptively selecting robot welding system or additive manufacturing process, realizing quality closed loop through digital twinning simulation and online detection;

[0133] The core bone structure is roughly divided into a simple structure of a truss type and a complex special-shaped structure. For the core bone of the simple structure of the truss type, a robot automatic welding system is adopted. The robot automatic welding system generates a welding seam track based on a three-dimensional model, and adopts an A* algorithm to optimize a robot path. Welding parameters (current 200 A-250 A, voltage 28 V-32 V, and speed 300 mm / min-500 mm / min) are set, and welding deformation is simulated through digital twinning. A laser tracking sensor is used to correct the position of a welding torch in real time.

[0134] For the complex special-shaped structure, an additive manufacturing process is adopted. The layer thickness (2 mm-5 mm) is dynamically adjusted according to the curvature change of the core bone cross section, and slicing processing is performed. In the process, a 2-mm-thin layer is used in a high-curvature area to reduce the step effect, and a 5-mm-thick layer is used in a low-curvature area (such as a flat plate section) to improve efficiency. A cladding path is generated, and a spiral line and a parallel line are combined. The spiral line is continuously scanned in one direction, the number of start-stop times is reduced, and the spiral line is suitable for large flat areas. The parallel line is alternated at 0° / 90°, anisotropic shrinkage is eliminated, and the parallel line is used for thin-walled features. In the transition area, a 5-mm-Bézier curve is used to transition at the spiral-parallel intersection to avoid material accumulation caused by path mutation. The cladding temperature field of each layer is predicted based on the digital twinning model, and the process parameters are dynamically adjusted. Laser scanning detection is performed, 2D contour scanning and 3D topography reconstruction are performed on each core bone cross section, and a feedback correction strategy is used to correct the contour deviation. The detection data are transmitted to the slicing software in real time to realize quality closed loop. The internal defects of the core bone are scanned by industrial CT, and the actual size of the core bone is obtained by three-dimensional laser scanning. The actual size is compared with the digital model, and the geometric and position errors are calculated. If the geometric and position errors exceed the set value, the error area features are automatically extracted and fed back to step S200 for modification of the core bone structure parameters.

[0135] As shown in Figure 2 step S500: full life cycle data driven continuous optimization, a design-manufacturing-service database is constructed, and a reinforcement learning model is used to realize automatic pushing and iteration of parameters.

[0136] The design-manufacturing-service database stores data in the design, manufacturing, and service stages. A deep deterministic policy gradient (DDPG) algorithm is used to train an intelligent decision-making model. Based on the intelligent decision-making model, optimized parameters are automatically pushed for a new design task.

[0137] The following data are stored in the design-manufacturing-service database:

[0138] Design stage: optimization algorithm parameters, finite element simulation results, and AI model output;

[0139] Manufacturing stage: process parameters (welding current, additive speed, etc.), and detection data (size error, defect distribution, etc.);

[0140] Service stage: feedback of casting machining precision (such as guide rail surface flatness), sand core failure cases.

[0141] The reinforcement learning model is trained by using a deep deterministic policy gradient (DDPG) algorithm, a parameter-manufacturing process-casting quality is taken as a state space, and a casting qualified rate maximization is taken as a reward function, so as to train an intelligent decision model: a standard qualified rate is 95%, a reward value is +10 when the casting qualified rate is greater than the standard qualified rate 95%, and a penalty value is -20 when the sand core fails; the model decision accuracy is improved through 1000+ production batch data training.

[0142] The above merely describes the preferred embodiments of the present application and is not intended to limit the present application, and any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A high-end numerical control machine tool base large piece sand mold casting core bone intelligent manufacturing method, characterized in that, The method comprises the following steps: Step S100: Constructing a digitalized basic model of the core design and a constraint system, generating a three-dimensional model of the sand mold based on the three-dimensional model of the casting through reverse engineering software, and extracting sand features and performing partitioning according to an algorithm; Step S100 comprises the following steps: Step S101: Defining a multi-dimensional constraint database of mechanics, thermodynamics and manufacturing processes, and establishing a parameterized digital model; Step S102: input the casting three-dimensional model data into the parameterized digital model, input the temperature field T, the velocity field V, the temperature gradient during solidification , the solid phase rate , the residual stress , and the gating system coordinate G, the chill position C, and the parting surface P according to the simulation results of the simulated gating system. Step S103: Aligning the three-dimensional model of the casting and the data of the simulated pouring system simulation to the same set of voxel grids; Uniform voxelization of the three-dimensional model of the casting by 0.5 mm: ； Sampling the data of the simulated simulation to the same grid: ， ; Generating a sand mold grid: ； wherein, , , correspond to the first, second, third , , voxels in the X, Y, Z directions respectively; is the inflated radius; is a three-dimensional (0, 1) array, is the voxel inside the casting, is empty; is the sand mold mesh; Extracting sand features and performing partitioning according to an algorithm, and the partitioning includes a high stress zone, a conventional support zone and a lightweight priority zone; The high stress zone HSZ includes the hot spot zone and the cantilever end zone ; extracting a three-dimensional region of the hot spot zone and the cantilever end zone in the high stress zone HSZ, outputting a triangulated surface after merging processing ; ； ； ； where t is the local solidification time, is the temperature gradient, is the distance from the gate, L is the overhang length, is the overhang angle, is the yield stress, - is the voxel mask, which is a 0-1 array with the same size as the three-dimensional grid, 1 means the voxel is selected, 0 means not selected; is the voxel mask, which is a 0-1 array with the same size as the three-dimensional grid, 1 means the voxel is selected, 0 means not selected; is the voxel mask, which is a 0-1 array with the same size as the three-dimensional grid, 1 means the voxel is selected, 0 means not selected; is the voxel mask, which is a 0-1 array with the same size as the three-dimensional grid, 1 means the voxel is selected, 0 means not selected; The conventional support zone RSZ includes a planar or shallow cavity area, the planar area is detected, and the local minimum thickness of the shallow cavity is calculated; Plane detection: The outer surface is calculated by Region Growing and RANSAC algorithm, the normal variation and the area 400 of the plane is recorded as ; Shallow cavity detection: ； determining whether it is a shallow cavity, if then marking it as a shallow cavity region, otherwise it is not a shallow cavity region; Merging: ; wherein, is a local minimum thickness, is a plane detected by the plane detection; Extracting data in the lightweight priority zone LPZ: ； where V is the volume, is the average stress, is the yield stress, is the distance from the high stress zone; Step S200: Establishing a multi-objective optimization function and constructing a core feature database, and adopting an improved genetic algorithm with a fusion simulated annealing operator to realize bidirectional data interaction with a finite element simulation tool; Step S300: Automatically strengthening the weak area of the core based on a deep learning model; Step S300 comprises the following steps: Step S301: Constructing a database, collecting historical core failure cases, the cases containing stress distribution cloud maps, temperature field data and failure position coordinates, and expanding the sample quantity through data enhancement; Step S302: Adopting an improved U-Net deep learning network, inputting a stress field / temperature field tensor of finite element analysis, and outputting a weak area probability map; Step S303: Training and optimizing using an Adam optimizer; Step S400: Intelligent manufacturing and precision control based on digital twinning, adaptively selecting a manufacturing process according to the structural characteristics of the core, adaptively selecting a robot welding system or an additive manufacturing process, and realizing quality closed loop through digital twinning simulation and online detection; Step S500: Continuous optimization driven by full life cycle data, constructing a design-manufacturing-service database, and realizing automatic pushing and iteration of parameters by adopting a reinforcement learning model.

2. The intelligent manufacturing method of high-end CNC machine tool base large piece sand mold core bone according to claim 1, characterized in that, The multi-objective optimization function comprises: constructing a weight minimization objective function, ; construct a stiffness maximization objective function, ; constructing a sand core deformation minimization objective function, ; wherein, is the main support beam cross-sectional width, is the main support beam cross-sectional height, is the thickness, is the weight reduction hole diameter, is the core bone material density, is the core bone volume, is the material elastic modulus, is the cross-sectional moment of inertia, is the amount of deformation of each area sand core, is the area weight.

3. The intelligent manufacturing method of high-end CNC machine tool base large piece sand mold core bone according to claim 1, characterized in that, Step S200 comprises the following steps: Step S201: Population initialization, randomly generating 50 groups of core structure parameters, including the main support beam section width W, the main support beam section height H, the reinforcing rib distribution spacing S and the weight reduction hole diameter d, and constructing an initial population; Step S202: apply constraints to each individual model, and perform static structural analysis and thermal-structural coupling analysis; output key indicators: maximum stress , maximum deformation , and core bone weight ; Step S203: define fitness function: wherein , , is the initial design value, = 0.3, = 0.4, = 0.3 is the weight coefficient; screen individuals with ≥ 0.8 into the next generation, eliminate individuals that do not meet the constraint conditions; Step S204: Genetic operation adopts a tournament selection method, 30% of the best individuals are selected from the parent generation; a single-point crossover is performed on the main support beam parameters; a simulated annealing idea is introduced to randomly mutate the rib spacing, and the probability of accepting a poor solution decreases with the number of iterations; Step S205: When the optimal fitness value fluctuates less than 1% for 5 consecutive generations, the optimal topological structure is output.

4. The intelligent manufacturing method of high-end CNC machine tool base large piece sand mold core bone according to claim 1, characterized in that, For the identified weak area of the core, an enhancement scheme is automatically generated, the weak area of the core includes a static stress concentration area, a thermal deformation sensitive area and a lightweight area, and the enhancement scheme for each weak area of the core comprises the following: Static stress concentration area, using bionic topology to increase the cross-sectional moment of inertia, the grid size is adaptively adjusted according to the stress gradient, and a round corner transition is set at the corner to reduce the stress concentration coefficient; Thermal deformation sensitive area, design hollow cavity structure, built-in spiral heat dissipation channel, channel direction consistent with the direction of metal liquid flow, enhance the convection heat dissipation; Lightweight area, open array type lightening holes, hole spacing is 2-3 times of the hole diameter.

5. The intelligent manufacturing method of high-end CNC machine tool base heavy sand type core bone according to claim 1, characterized in that, Step S400 includes the following: For the core bone of truss type simple structure, a robot automatic welding system is used: the robot automatic welding system generates a weld seam trajectory based on a three-dimensional model, and uses A* algorithm to optimize the robot path; set the welding parameters, and simulate the welding deformation through digital twinning; use laser tracking sensor to correct the welding gun position in real time; For complex special-shaped structure, using additive manufacturing process: according to the change of core bone section curvature, dynamically adjust the layer thickness, carry on the slicing processing; Generate cladding path, spiral line and parallel line composite path; Based on the digital twin model, predict the temperature field of each layer of cladding, dynamically adjust the process parameters; laser scanning detection, 2D contour scanning and 3D topography reconstruction are carried out on each layer of core bone section, and feedback correction strategy is used to correct the contour deviation; the detection data is transmitted to the slicing software in real time to realize the quality closed loop, the internal defects of the core bone are detected by industrial CT scanning, and the actual size of the core bone is obtained by three-dimensional laser scanning, compared with the digital model, the form error is calculated; if the form error exceeds the set value, the error area features are automatically extracted and fed back to step S200 for correction of the core bone structure parameters.

6. The intelligent manufacturing method of high-end CNC machine tool base large piece sand mold core bone according to claim 1, characterized in that, Step S500 includes the following: Constructing a design-manufacturing-service database to store the data of design, manufacturing and service stages; using deep deterministic policy gradient (DDPG) algorithm to train intelligent decision model; based on the intelligent decision model, automatically push the optimization parameters for new design tasks.

Images