Intelligent compensation path planning method for additive manufacturing of five-axis space mold

Abstract

The invention discloses a five-axis space mold additive manufacturing intelligent compensation path planning method, and relates to the technical field of additive manufacturing. By establishing a multi-physics field coupling model, generating an initial five-axis printing path, collecting multi-physics field data in real time, dynamically adjusting the printing path and process parameters, executing a compensation path through five-axis linkage control and verifying the compensation effect through a digital twin platform, the precision and efficiency problems in complex mold manufacturing are solved; through combination of heat-force-flow multi-field coupling prediction and dynamic compensation, limitation of a single physical field is broken through; eliminating a step effect and singular points by adopting a normal alignment strategy and five-axis kinematics optimization; high-frequency sensor closed-loop control and a special G code instruction set are applied to realize high response precision; through intelligent path planning and five-axis cooperative control, an efficient and stable solution is provided for high-precision complex mold additive manufacturing.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing technology, and more particularly to an intelligent compensation path planning method for five-axis spatial mold additive manufacturing. Background Technology

[0002] With the rapid development of high-end industrial fields such as aerospace, precision medical devices, and automobile manufacturing, the manufacturing precision and quality requirements for complex molds are increasing. Traditional mold manufacturing methods (such as CNC machining and EDM) often face problems such as low processing efficiency, serious material waste, and high post-processing costs when dealing with complex geometries and high precision requirements. Additive manufacturing technology, as an emerging manufacturing process, has attracted widespread attention because it can directly and rapidly manufacture complex structures from three-dimensional digital models. However, the complex thermo-mechanical-fluid coupling effect in additive manufacturing leads to problems such as thermal deformation, residual stress, and molten pool instability, which seriously affect the dimensional accuracy and surface quality of the printed parts.

[0003] In additive manufacturing of complex molds, the following issues are particularly prominent: Thermo-mechanical-fluid coupling effect: When lasers melt metal powder, heat conduction, thermal expansion, molten pool flow, and solidification processes are coupled, leading to localized thermal stress concentration and non-uniform deformation, ultimately affecting the geometric accuracy of the printed parts; Real-time monitoring and compensation challenges: Existing additive manufacturing systems lack real-time monitoring and dynamic compensation mechanisms, making it difficult to effectively control thermal deformation, interlayer deformation, and defects during the printing process; Complexity of five-axis linkage control: Printing complex curved surface molds requires five-axis linkage control, but traditional path planning methods easily lead to sudden changes in print head posture and singularity issues, thus affecting print quality and efficiency; High post-processing costs: Due to insufficient printing accuracy, traditional methods typically require a large amount of post-processing work (such as machining and heat treatment), significantly increasing manufacturing costs and delivery cycles. Summary of the Invention

[0004] To address the aforementioned issues, this patent proposes a five-axis additive manufacturing intelligent path planning method based on multiphysics coupling modeling, real-time monitoring, and dynamic compensation. This method predicts deformation behavior through thermo-mechanical-fluid coupling modeling, dynamically adjusts the printing path and process parameters using real-time monitoring data, and achieves high-precision printing through five-axis linkage control. This method not only significantly improves the manufacturing accuracy and surface quality of complex molds but also substantially reduces post-processing costs and shortens delivery cycles, providing a reliable solution for the high-end manufacturing field.

[0005] The present invention provides an intelligent compensation path planning method for five-axis spatial mold additive manufacturing, which is implemented through the following technical solution: An intelligent compensation path planning method for five-axis spatial mold additive manufacturing includes the following steps:

[0006] S1. Establish a multi-physics coupling model that includes thermal field, stress field and molten pool flow field to predict the temperature gradient distribution, thermal stress field and molten pool flow behavior in the additive manufacturing process.

[0007] S2. Based on the geometric features of the mold's 3D model, generate an initial five-axis printing path and optimize the print head posture to avoid kinematic singularities;

[0008] S3. Real-time acquisition of molten pool temperature field, interlayer deformation and acoustic emission signals, and extraction of multi-physics field characteristic parameters;

[0009] S4. Based on the deviation between the prediction model and real-time monitoring data, dynamically adjust the printing path offset, laser power, and scanning speed to generate a compensation path;

[0010] S5. The compensation path is executed through five-axis linkage control, and the NURBS interpolation algorithm is used to ensure the smoothness of the motion trajectory;

[0011] S6. Verify the compensation effect on the digital twin platform and iteratively optimize the compensation parameters using a Bayesian optimization algorithm.

[0012] Preferably, in step S1, based on the three-dimensional geometric model of the mold, a thermo-mechanical-fluid multiphysics coupling prediction model is established, and the temperature field, thermal stress field, and molten pool flow field are calculated using a finite element solver; the thermo-mechanical-fluid coupling modeling includes:

[0013] Thermal field modeling: Establishing a three-dimensional transient heat conduction equation, considering laser power P, scanning speed v, and material thermal diffusivity α:

[0014]

[0015] in (Gaussian heat source model), Q loss For convective / radiative heat dissipation, the heat conduction equation is discretized using an explicit time integration method:

[0016]

[0017] in To ensure numerical stability;

[0018] Stress field modeling: Solving for thermal stress distribution based on thermo-elastic-plastic theory:

[0019] σ ij =C ijkl (∈ kl -α T ΔTδ kl )

[0020] Where ∈ kl For the total strain, α TThe coefficient of thermal expansion;

[0021] Molten pool fluid dynamics modeling: The VOF (Volume of Fluid) method is used to simulate molten pool flow, and the Marangoni effect is coupled to analyze the evolution of molten pool morphology; coupling the Marangoni effect:

[0022]

[0023] (φ is the phase field function, γ is the surface tension coefficient)

[0024] Offline simulation and database construction include: building an offline database. Where P is the laser power, v is the scanning speed, h is the layer thickness, and Δzpred is the predicted deformation; multi-field coupled simulation is performed using the finite element software COMSOL, the three-dimensional transient heat conduction equation is solved using the finite element method, and the temperature field distribution is calculated by combining the Gaussian heat source model;

[0025] Generate a deformation-temperature-stress mapping database for typical working conditions;

[0026] Establish a machine learning model (LSTM) with input features P, v, h. σ von Mises The output is the predicted deformation Δz pred .

[0027] Preferably, in step S2, the deformation Δz is predicted based on the prediction model from step S1. pred Generate the initial five-axis printing path P nominal The printhead orientation is optimized by calculating the surface normal vector n(u,v) using differential geometry, ensuring that the angle between the deposition direction and the normal satisfies θ=arccos(n·t)≤5°, where t is the printhead axis. Simultaneously, the condition number κ(J) is based on the Jacobian matrix condition number κ(J)=||J||·||J -1 Avoid motion singularities.

[0028] In step S2, the preferred five-axis tool attitude optimization includes:

[0029] Normal alignment strategy: Calculate the normal vector n(u,v) of each point based on surface differential geometry, and adjust the printhead attitude by rotating the axis (A / C) to ensure that the angle θ between the deposition direction and the normal is ≤5°;

[0030] The normal alignment strategy is achieved through the coordinated motion of rotation axes A and C, and its inverse kinematics is:

[0031]

[0032] Where n = (n x ,ny ,n z ) represents the surface normal vector;

[0033] Singularity avoidance: Construct a kinematic model of a five-axis machine tool and use Jacobi matrix condition number analysis to automatically avoid singular configurations during path planning;

[0034] Initial path generation includes:

[0035] During initial path generation, layer thickness is dynamically allocated based on curvature κ(u,v):

[0036]

[0037] An equal residual height slicing algorithm is adopted: the layer thickness is adaptively adjusted according to the curvature (h = 0.8R). min R min (where the radius of curvature is the local minimum);

[0038] Generate variable density spiral fill paths: path spacing d = 0.1 mm in high curvature areas and d = 0.3 mm in flat areas.

[0039] In step S3, preferably, an infrared thermal imager, a laser displacement sensor, and an acoustic emission sensor are integrated, with sampling frequencies of not less than 1kHz, 500Hz, and 200kHz, respectively. The absolute value of the temperature gradient, the average value of interlayer deformation, and the abrupt change rate of acoustic signal energy are extracted as characteristic parameters. A dynamic threshold trigger compensation mechanism is set: compensation is activated when the temperature gradient is ≥500K / mm, the deformation is ≥0.1mm, or the acoustic signal energy is ≥3σ of the baseline. During the printing process, the temperature field T of the molten pool is collected in real time by multiple sensors. meas Interlayer deformation Δz meas and acoustic emission signal energy E AE And calculate the characteristic parameters:

[0040] Δz error =Δz meas -Δz pred ,

[0041] Key features are extracted by aligning multi-sensor data using timestamps.

[0042] The preferred key features include: temperature gradient (Threshold: ≥500K / mm triggers early warning); Deformation Δz (Threshold: ≥0.1mm triggers compensation); Sudden change in acoustic signal energy (Threshold: ≥3σ is judged as a defect); When |Δz error |≥0.1mm or|ΔE AE When |≥3, the compensation mechanism is triggered.

[0043] Preferably, in step S4, based on the monitoring results of step S3, depending on P in step S2 nominal and Δz in step S3 error Dynamically adjust path and process parameters; dynamic compensation of multi-field coupling: predict deformation through a thermo-mechanical-fluid model, and achieve feedforward-feedback composite control by combining real-time monitoring data, mathematical expression:

[0044]

[0045] Feedforward compensation based on model prediction: This involves calling an offline database to predict the deformation Δz of the next three layers based on current process parameters. pred Generate reverse path compensation amount: P comp =P nominal -k·Δz pred ·n (k = 0.6-0.8 is an empirical coefficient, and n is the normal vector);

[0046] PID parameter adjustment: laser power increment Scan rate increment Δv = f(ΔT, ΔE) AE When a local defect is detected, insert a concentric circle repair path P. repair The line spacing d ≤ 0.05 mm, and the five-axis attitude is adjusted to normal deflection ±10°;

[0047] Concentric circle path P repair The generation method is as follows: taking the defect point (x0, y0, z0) as the center, and according to the radius r j =j·d (j=1,2,...,m) generates concentric circle paths;

[0048] The formula for calculating the coordinates of path points is:

[0049] x j =x0+r j cosθ,y j =y0+r j sinθ(θ∈[0,2π))

[0050] When performing a re-firing, the laser power is set to P. repair =0.6P nominal Frequency f repair =100kHz;

[0051] Feedback-based PID dynamic adjustment: Heat conduction equation (step S1) → Temperature gradient (Step S3) → PID control input (Step S4);

[0052] Design a multivariable PID controller:

[0053]

[0054] Input error e(t): Measured deformation Δz meas vs. predicted value Δz pred ;

[0055] Output adjustment amounts: laser power ΔP (range ±10%), scanning speed Δv (range ±15%); Path replanning strategy: when a porosity ≥2% or a non-fusion defect is detected, the local path replanning module is triggered, and a high-frequency low-power mode (frequency ≥100kHz, power 50-70% of the nominal value) is used for re-printing; when a local defect (such as porosity >2%) is detected: the current printing is paused, and a local area concentric circle filling path (line spacing 0.05mm) is started; the five-axis attitude is adjusted to the optimal repair angle (usually ±10° of the defect point normal), and a high-frequency low-power mode (laser frequency 100kHz, power reduced to 60%) is used for re-printing.

[0056] Preferably, in step S5, NURBS curves are used to interpolate discrete path points to generate a smooth trajectory, ensuring that the acceleration a on each axis is... i (t) is continuous and its maximum value satisfies max(|a i (t)|)≤0.5rad / s 2 ;

[0057] The node vector U and weight factor w of NURBS interpolation are adaptively adjusted according to the path curvature:

[0058] In high curvature regions, the node spacing Δu ≤ 0.1 mm, and the weighting factor w i =1+0.5|κ i |;

[0059] In the low curvature region, the node spacing Δu ≤ 0.5 mm, and the weighting factor w i =1;

[0060] formula:

[0061]

[0062] Where q(t) is the five-axis joint spatial trajectory.

[0063] Preferably, in step S6, the compensation effect is verified in the digital twin platform, requiring the warpage after compensation to be ≤0.05mm; and the compensation parameters k and K are iteratively updated using a Bayesian optimization algorithm. p ,K i ,K d The objective function is:

[0064]

[0065] The Pareto optimal solution is obtained iteratively using the Bayesian optimization algorithm until max(|Δz) is satisfied. meas |)≤0.05mm.

[0066] This invention overcomes the limitations of traditional single-physics-field compensation through a prediction-compensation mechanism driven by multi-field coupling, achieving high-precision deformation prediction through thermo-mechanical-fluid coupling; it solves the problem of abrupt attitude changes in complex surface printing through normal alignment and NURBS interpolation; and it ensures the timeliness of dynamic compensation through millisecond-level closed-loop control. This invention provides reliable precision assurance for additive manufacturing of complex molds in aerospace, precision medical devices and other fields, and can significantly reduce post-processing costs and shorten delivery cycles. Attached Figure Description

[0067] Figure 1 This is a flowchart of the steps of the present invention;

[0068] Figure 2 This is a schematic diagram of the multiphysics coupling model of the present invention;

[0069] Figure 3 This is a schematic diagram of the five-axis linkage printing path planning of the present invention;

[0070] Figure 4 This is a schematic diagram of the dynamic compensation mechanism of the present invention. Detailed Implementation

[0071] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0072] This invention provides an intelligent compensation path planning method for five-axis spatial mold additive manufacturing, comprising the following steps:

[0073] S1. Establish a multiphysics coupled model including the thermal field, stress field, and molten pool flow field to predict the temperature gradient distribution, thermal stress field, and molten pool flow behavior during additive manufacturing; calculate the temperature field, thermal stress field, and molten pool flow field using a finite element solver; thermo-mechanical-fluid coupled modeling includes:

[0074] Thermal field modeling: Establishing a three-dimensional transient heat conduction equation, considering laser power P, scanning speed v, and material thermal diffusivity α:

[0075]

[0076] in (Gaussian heat source model), Qloss For convective / radiative heat dissipation, the heat conduction equation is discretized using an explicit time integration method:

[0077]

[0078] in To ensure numerical stability;

[0079] Stress field modeling: Solving for thermal stress distribution based on thermo-elastic-plastic theory:

[0080] σ ij =C ijkl (∈ kl -α T ΔTδ kl )

[0081] Where ∈ kl For the total strain, α T The coefficient of thermal expansion;

[0082] Molten pool fluid dynamics modeling: The VOF (Volume of Fluid) method is used to simulate molten pool flow, and the Marangoni effect is coupled to analyze the evolution of molten pool morphology; coupling the Marangoni effect:

[0083]

[0084] (φ is the phase field function, γ is the surface tension coefficient)

[0085] Offline simulation and database construction include: building an offline database. Where P is the laser power, v is the scanning speed, h is the layer thickness, and Δzpred is the predicted deformation; multi-field coupled simulation is performed using the finite element software COMSOL, the three-dimensional transient heat conduction equation is solved using the finite element method, and the temperature field distribution is calculated by combining the Gaussian heat source model;

[0086] Generate a deformation-temperature-stress mapping database for typical working conditions;

[0087] Establish a machine learning model (LSTM) with input features P, v, h. σ von Mises The output is the predicted deformation Δz pred .

[0088] S2. Based on the geometric features of the mold's 3D model, generate an initial five-axis printing path and optimize the print head posture to avoid kinematic singularities; predict the deformation Δz according to the prediction model from step S1. pred Generate the initial five-axis printing path P nominalThe printhead orientation is optimized by calculating the surface normal vector n(u,v) using differential geometry, ensuring that the angle between the deposition direction and the normal satisfies θ=arccos(n·t)≤5°, where t is the printhead axis. Simultaneously, the condition number κ(J) is based on the Jacobian matrix condition number κ(J)=||J||·||J -1 ||Avoiding motion singularities; 5-axis tool attitude optimization includes:

[0089] Normal alignment strategy: Calculate the normal vector n(u,v) of each point based on surface differential geometry, and adjust the printhead attitude by rotating the axis (A / C) to ensure that the angle θ between the deposition direction and the normal is ≤5°;

[0090] The normal alignment strategy is achieved through the coordinated motion of rotation axes A and C, and its inverse kinematics is:

[0091]

[0092] Where n = (n x ,n y ,n z ) represents the surface normal vector;

[0093] Singularity avoidance: Construct a kinematic model of a five-axis machine tool and use Jacobi matrix condition number analysis to automatically avoid singular configurations during path planning;

[0094] Initial path generation includes:

[0095] During initial path generation, layer thickness is dynamically allocated based on curvature κ(u,v):

[0096]

[0097] An equal residual height slicing algorithm is adopted: the layer thickness is adaptively adjusted according to the curvature (h = 0.8R). min R min (where the radius of curvature is the local minimum);

[0098] An equal residual height slicing algorithm is adopted: the layer thickness is adaptively adjusted according to the curvature (h = 0.8R). min R min (where the radius of curvature is the local minimum);

[0099] Generate variable density spiral fill paths: path spacing d = 0.1 mm in high curvature areas and d = 0.3 mm in flat areas;

[0100] S3. Real-time acquisition of molten pool temperature field, interlayer deformation, and acoustic emission signals; extraction of multi-physics characteristic parameters; integration of infrared thermal imager, laser displacement sensor, and acoustic emission sensor, with sampling frequencies no less than 1kHz, 500Hz, and 200kHz respectively; extraction of absolute temperature gradient, mean interlayer deformation, and acoustic signal energy mutation rate as characteristic parameters; setting a dynamic threshold to trigger compensation mechanism: compensation is activated when temperature gradient ≥ 500K / mm, deformation ≥ 0.1mm, or acoustic signal energy ≥ baseline 3σ; during printing, the molten pool temperature field T is acquired in real-time through multiple sensors. meas Interlayer deformation Δz meas and acoustic emission signal energy E AE And calculate the characteristic parameters:

[0101] Δz error =Δz meas -Δz pred ,

[0102] Key features are extracted by aligning multi-sensor data using timestamps; key features include: temperature gradient. (Threshold: ≥500K / mm triggers early warning); Deformation Δz (Threshold: ≥0.1mm triggers compensation); Sudden change in acoustic signal energy (Threshold: ≥3σ is judged as a defect); When |Δz error |≥0.1mm or|ΔE AE When |≥3, the compensation mechanism is triggered.

[0103] S4. Based on the deviation between the prediction model and real-time monitoring data, dynamically adjust the printing path offset, laser power, and scanning speed to generate a compensation path; based on the monitoring results of step S3, and depending on the P of step S2... nominal and Δz in step S3 error Dynamically adjust path and process parameters; dynamic compensation of multi-field coupling: predict deformation through a thermo-mechanical-fluid model, and achieve feedforward-feedback composite control by combining real-time monitoring data, mathematical expression:

[0104]

[0105] Feedforward compensation based on model prediction: This involves calling an offline database to predict the deformation Δz of the next three layers based on current process parameters. pred Generate reverse path compensation amount: P comp =P nominal -k·Δz pred ·n

[0106] (k = 0.6-0.8 is an empirical coefficient, and n is the normal vector);

[0107] PID parameter adjustment: Laser power increment ΔP = K p ·Δz error +K i ∫Δz error dt+ Scan rate increment Δv = f(ΔT, ΔE) AE When a local defect is detected, insert a concentric circle repair path P. repair The line spacing d ≤ 0.05mm, and the five-axis attitude is adjusted to normal deflection.

[0108] ±10°;

[0109] Concentric circle path P repair The generation method is as follows: taking the defect point (x0, y0, z0) as the center, and according to the radius r j =j·d (j=1,2,...,m) generates concentric circle paths;

[0110] The formula for calculating the coordinates of path points is:

[0111] x j =x0+r j cosθ,y j =y0+r j sinθ(θ∈[0,2π))

[0112] When performing a re-firing, the laser power is set to P. repair =0.6P nominal Frequency f repair =100kHz;

[0113] Feedback-based dynamic PID adjustment:

[0114] Design a multivariable PID controller:

[0115]

[0116] Input error e(t): Measured deformation Δz meas vs. predicted value Δz pred ;

[0117] Output adjustment parameters: laser power ΔP (range ±10%), scanning speed Δv (range ±15%);

[0118] Path replanning strategy: When a porosity ≥2% or a non-fusion defect is detected, the local path replanning module is triggered, and a high-frequency low-power mode (frequency ≥100kHz, power 50-70% of the nominal value) is used for re-printing; when a local defect (such as porosity >2%) is detected: the current printing is paused, and a local area concentric circle filling path (line spacing 0.05mm) is started; the five-axis attitude is adjusted to the optimal repair angle (usually ±10° of the defect point normal), and a high-frequency low-power mode (laser frequency 100kHz, power reduced to 60%) is used for re-printing.

[0119] S5. The compensation path is executed through five-axis linkage control, and the NURBS interpolation algorithm is used to ensure the smoothness of the motion trajectory; NURBS curves are used to interpolate discrete path points to generate a smooth trajectory, ensuring the acceleration a of each axis. i (t) is continuous and its maximum value satisfies max(|a i (t)|)≤0.5rad / s 2 ;

[0120] The node vector U and weight factor w of NURBS interpolation are adaptively adjusted according to the path curvature:

[0121] In high curvature regions, the node spacing Δu ≤ 0.1 mm, and the weighting factor w i =1+0.5|κ i |;

[0122] In the low curvature region, the node spacing Δu ≤ 0.5 mm, and the weighting factor w i =1;

[0123] formula:

[0124]

[0125] Where q(t) is the five-axis joint spatial trajectory.

[0126] S6. Verify the compensation effect on the digital twin platform, and iteratively optimize the compensation parameters using a Bayesian optimization algorithm, requiring the warpage after compensation to be ≤0.05mm; and iteratively update the compensation parameters k and K using a Bayesian optimization algorithm. p ,K i ,K d The objective function is:

[0127]

[0128] The Pareto optimal solution is obtained iteratively using the Bayesian optimization algorithm until max(|Δz) is satisfied. meas |)≤0.05mm.

[0129] Example 1:

[0130] 1. Multiphysics Modeling and Simulation Prediction

[0131] (1) Thermal-mechanical-fluid coupling modeling

[0132] Thermal field modeling:

[0133] Establish a three-dimensional transient heat conduction equation, considering laser power P, scanning speed v, and material thermal diffusivity α:

[0134]

[0135] in (Gaussian heat source model), Q loss For convective / radiative heat dissipation. Stress field modeling:

[0136] Based on the thermo-elastic-plastic theory, the thermal stress distribution is solved:

[0137] σ ij =C ijkl (∈ kl -α T ΔTδ kl )

[0138] Where ∈ kl For the total strain, α T is the coefficient of thermal expansion.

[0139] Molten pool hydrodynamic modeling:

[0140] The VOF (Volume of Fluid) method was used to simulate molten pool flow, coupled with the Marangoni effect:

[0141]

[0142] (φ is the phase field function, γ is the surface tension coefficient)

[0143] (2) Offline simulation and database construction

[0144] Multi-field coupled simulations are performed using finite element software (such as COMSOL) to generate a deformation-temperature-stress mapping database under typical working conditions.

[0145] Establish a machine learning model (such as random forest or LSTM), input parameters (laser power, velocity, layer thickness), and output predicted deformation (accuracy up to ±5μm).

[0146] 2. Five-axis path planning and initial path generation

[0147] (1) Five-axis tool attitude optimization

[0148] Normal alignment strategy: The normal vector n(u,v) of each point is calculated based on the surface differential geometry. The printhead posture is adjusted by rotating the axis (A / C) to ensure that the angle θ between the deposition direction and the normal is ≤5°.

[0149] Singularity avoidance: Construct a kinematic model of a five-axis machine tool and use Jacobi matrix condition number analysis to automatically avoid singular configurations during path planning.

[0150] (2) Initial path generation

[0151] An equal residual height slicing algorithm is adopted: the layer thickness is adaptively adjusted according to the curvature (h = 0.8R). min R min (where the radius of curvature is the local minimum).

[0152] Generate variable density spiral fill paths: path spacing d = 0.1 mm in high curvature areas and d = 0.3 mm in flat areas.

[0153] 3. Real-time monitoring and data fusion

[0154] (1) Multi-sensor integration

[0155] Infrared thermal imager: monitors the temperature field of the molten pool (sampling rate 1kHz, spatial resolution 0.1mm).

[0156] Laser displacement sensor: measures interlayer deformation (accuracy ±2μm, Z-axis repeatability 0.5μm). Acoustic emission sensor: captures solidification crack signals in the molten pool (frequency band 50-200kHz).

[0157] (2) Data synchronization and feature extraction

[0158] Key features extracted by aligning multi-sensor data using timestamps: temperature gradient (Threshold: ≥500K / mm triggers early warning); Deformation Δz (threshold: ≥0.1mm triggers compensation); Sudden change in acoustic signal energy (threshold: ≥3σ is judged as a defect).

[0159] 4. Implementation of intelligent compensation algorithm

[0160] Dynamic compensation with multi-field coupling: Deformation is predicted by a thermo-mechanical-fluid model, and feedforward-feedback composite control is achieved by combining real-time monitoring data.

[0161] Mathematical expression:

[0162] (1) Feedforward compensation based on model prediction

[0163] By accessing an offline database, the deformation Δz of the next three layers can be predicted based on current process parameters. pred ;

[0164] Generate reverse path compensation amount: P comp =P nominal -k·Δz pred ·n (k = 0.6-0.8 is an empirical coefficient, and n is the normal vector);

[0165] PID parameter adjustment: laser power increment Scan rate increment Δv = f(ΔT, ΔE) AE );

[0166] When a local defect is detected, insert a concentric circle repair path P. repair The line spacing d ≤ 0.05 mm, and the five-axis attitude is adjusted to normal deflection ±10°;

[0167] Concentric circle path P repair The generation method is as follows:

[0168] Centered on the defect point (x0, y0, z0), and with radius r j =j·d (j=1,2,...,m) generates a concentric circle path; the formula for calculating the coordinates of the path points is:

[0169] x j =x0+r j cosθ,y j =y0+r j sinθ(θ∈[0,2π))

[0170] When performing a re-firing, the laser power is set to P. repair =0.6P nominal Frequency f repair =100kHz.

[0171] (2) Feedback-based PID dynamic adjustment

[0172]

[0173] Design a multivariable PID controller:

[0174]

[0175] Input error e(t): Measured deformation Δz meas vs. predicted value Δz pred .

[0176] Output adjustment parameters: laser power ΔP (range ±10%), scanning speed Δv (range ±15%).

[0177] (3) Path replanning strategy

[0178] When a porosity of ≥2% or a non-fusion defect is detected, the local path replanning module is triggered, and a high-frequency low-power mode (frequency ≥100kHz, power 50-70% of the nominal value) is used for re-mapping.

[0179] When local defects (such as porosity > 2%) are detected:

[0180] Pause the current print job and start a local area concentric circle fill path (line spacing 0.05mm);

[0181] Adjust the five-axis attitude to the optimal repair angle (usually ±10° of the defect point normal);

[0182] The high-frequency, low-power mode (laser frequency 100kHz, power reduced to 60%) was used for the re-firing.

[0183] 5. Five-axis motion control and execution

[0184] (1) Kinematic interpolation optimization

[0185] NURBS curve interpolation is used to smooth discrete path points to ensure continuous acceleration along each axis (C). 2 continuous).

[0186] The joint angular acceleration α is limited to ≤0.5 rad / s². 2 Avoid mechanical vibration.

[0187] (2) Real-time control architecture

[0188] Based on an RTX64 real-time system, the control cycle T c =1ms.

[0189] Five-axis linkage control command generation process:

[0190] Sensor data → Data fusion → Compensation algorithm → Path update → Inverse kinematics solution → G-code generation → Servo drive

[0191] 6. Verification and Optimization

[0192] (1) Offline verification

[0193] Import the mold CAD model into the digital twin platform and compare the deformation before and after compensation:

[0194] Uncompensated: Warpage 0.12-0.25mm.

[0195] After compensation: warpage ≤ 0.05mm.

[0196] (2) Online optimization

[0197] The compensation coefficient k and PID parameters are automatically adjusted using a Bayesian optimization algorithm.

[0198] Objective function: Minimize ∑(Δz) meas ) 2 +λ·t print .

[0199] After 10-20 iterations, the Pareto optimal solution is found.

[0200] Technical advantages and measured data:

[0201] index Traditional methods This plan Increase Dimensional accuracy (mm) ±0.15 ±0.05 66.7％ Surface roughness Ra (μm) 3.2 1.6 50％ Printing time (h) 38 32 15.8％ Material waste rate 12％ 5％ 58.3％

[0202] Example 2:

[0203] Application Case: Challenge of aircraft engine turbine disk mold: 800mm in diameter, containing 72 blade slots, traditional methods can cause slot deformation of up to 0.2mm.

[0204] Take aero-engine turbine blade molds as an example:

[0205] Step S1: Predict the asymmetric thermal deformation Δz caused by blade torsion pred =0.18mm;

[0206] Step S2: Generate normal alignment path, with a layer thickness of h = 30 μm in the leaf tip region and h = 120 μm in the leaf root region;

[0207] Step S3: Real-time detection of the blade tip temperature gradient Trigger compensation;

[0208] Step S4: Calculate the compensation path offset ΔP = 0.7 × 0.18 mm = 0.126 mm, and reduce the laser power ΔP = -8%;

[0209] Step S5: Execute the compensation path via G832 command, with acceleration limited to 0.4 rad / s². 2 ;

[0210] Step S6: Verify the maximum deformation Δz meas =0.04mm, Bayesian optimization adjusts k from 0.7 to 0.68.

[0211] Effect comparison:

[0212]

[0213]

[0214] This solution involves predicting the thermal-mechanical coupling and compensating for the slot offset by 0.15mm in advance, while adjusting the laser power in real time (dynamically from 300W to 270-330W).

[0215] Results: The slot accuracy reached ±0.04mm, the assembly clearance qualification rate increased from 65% to 98%, the subsequent EDM machining allowance was reduced by 70%, and the total manufacturing cost was reduced by 25%.

Claims

1. A five-axis spatial mold additive manufacturing intelligent compensation path planning method, characterized in that, Includes the following steps: S1. Establish a multi-physics coupling model that includes thermal field, stress field and molten pool flow field to predict the temperature gradient distribution, thermal stress field and molten pool flow behavior in the additive manufacturing process. S2. Based on the geometric features of the mold's 3D model, generate an initial five-axis printing path and optimize the print head posture to avoid kinematic singularities; S3. Real-time acquisition of molten pool temperature field, interlayer deformation and acoustic emission signals, and extraction of multi-physics field characteristic parameters; S4. Based on the deviation between the prediction model and real-time monitoring data, dynamically adjust the printing path offset, laser power, and scanning speed to generate a compensation path; S5. The compensation path is executed through five-axis linkage control, and the NURBS interpolation algorithm is used to ensure the smoothness of the motion trajectory; S6. Verify the compensation effect on the digital twin platform and iteratively optimize the compensation parameters using a Bayesian optimization algorithm.

2. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S1, based on the three-dimensional geometric model of the mold, a thermo-mechanical-fluid multiphysics coupling prediction model is established, and the temperature field, thermal stress field and molten pool flow field are calculated by the finite element solver. Thermal-mechanical-fluid coupling modeling includes: Thermal field modeling: Establishing a three-dimensional transient heat conduction equation, considering laser power P, scanning speed v, and material thermal diffusivity α: in Q loss For convective / radiative heat dissipation; When solving the heat conduction equation, the equation is discretized using the explicit time integration method: in To ensure numerical stability; Stress field modeling: Solving for thermal stress distribution based on thermo-elastic-plastic theory: s ij =C ijkl (∈ kl -a T ΔTδ kl ) Where ∈ kl For the total strain, α T The coefficient of thermal expansion; Molten pool fluid dynamics modeling: The VOF (Volume of Fluid) method is used to simulate molten pool flow, and the Marangoni effect is coupled to analyze the evolution of molten pool morphology; coupling the Marangoni effect: Offline simulation and database construction include: building an offline database. Where P is the laser power, v is the scanning speed, h is the layer thickness, and Δzpred is the predicted deformation; multi-field coupled simulation is performed using the finite element software COMSOL, the three-dimensional transient heat conduction equation is solved using the finite element method, and the temperature field distribution is calculated by combining the Gaussian heat source model; Generate a deformation-temperature-stress mapping database for typical working conditions; Establish a machine learning model with input features P, v, h. σ vonMises The output is the predicted deformation Δz pred .

3. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S2, the deformation Δz is predicted based on the prediction model from step S1. pred Generate the initial five-axis printing path P nominal The printhead orientation is optimized by calculating the surface normal vector n(u,v) using differential geometry, ensuring that the angle between the deposition direction and the normal satisfies θ=arccos(n·t)≤5°, where t is the printhead axis. Simultaneously, the condition number κ(J) is based on the Jacobian matrix condition number κ(J)=||J||·||J -1 Avoid motion singularities.

4. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S2, the five-axis tool attitude optimization includes: Normal alignment strategy: Calculate the normal vector n(u,v) of each point based on surface differential geometry, and adjust the printhead posture by rotating the axis to ensure that the angle θ between the deposition direction and the normal is ≤5°; The normal alignment strategy is achieved through the coordinated motion of rotation axes A and C, and its inverse kinematics is: Where n = (n x ,n y ,n z ) represents the surface normal vector; Singularity avoidance: Construct a kinematic model of a five-axis machine tool and use Jacobi matrix condition number analysis to automatically avoid singular configurations during path planning; Initial path generation includes: During initial path generation, layer thickness is dynamically allocated based on curvature κ(u,v): An equal residual height slicing algorithm is adopted: the layer thickness is adaptively adjusted according to the curvature (h = 0.8R). min R min (where the radius of curvature is the local minimum); Generate variable density spiral fill paths: path spacing d = 0.1 mm in high curvature areas and d = 0.3 mm in flat areas.

5. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S3, an infrared thermal imager, a laser displacement sensor, and an acoustic emission sensor are integrated, with sampling frequencies of not less than 1kHz, 500Hz, and 200kHz, respectively; the absolute value of the temperature gradient, the mean value of interlayer deformation, and the abrupt change rate of acoustic signal energy are extracted as characteristic parameters. A dynamic threshold-triggered compensation mechanism is set: compensation is activated when the temperature gradient is ≥500K / mm, the deformation is ≥0.1mm, or the acoustic signal energy is ≥3σ of the baseline; during the printing process, the molten pool temperature field T is collected in real time by multiple sensors. meas Interlayer deformation Δz meas and acoustic emission signal energy E AE And calculate the characteristic parameters: Δz error =Δz meas -Δz pred , Key features are extracted by aligning multi-sensor data using timestamps.

6. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 5, characterized in that: The key features include: temperature gradient Deformation Δz; sudden change in acoustic signal energy; when |Δz error |≥0.1mm or|ΔE AE When |≥3, the compensation mechanism is triggered.

7. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S4, based on the monitoring results of step S3, and depending on P from step S2... nominal and Δz in step S3 error Dynamically adjust the path and process parameters; Dynamic compensation involving multiple fields: Deformation is predicted using a thermo-mechanical-fluidic model, and feedforward-feedback composite control is achieved by combining real-time monitoring data. Mathematical expression: Feedforward compensation based on model prediction: This involves calling an offline database to predict the deformation Δz of the next three layers based on current process parameters. pred Generate reverse path compensation amount: P comp =P nominal -k·Δz pred ·n; PID parameter adjustment: laser power increment Scan rate increment Δv = f(ΔT, ΔE) AE When a local defect is detected, insert a concentric circle repair path P. repair The line spacing d ≤ 0.05 mm, and the five-axis attitude is adjusted to normal deflection ±10°; Concentric circle path P repair The generation method is as follows: taking the defect point (x0, y0, z0) as the center, and according to the radius r j =j·d (j=1,2,...,m) generates concentric circle paths; The formula for calculating the coordinates of path points is: x j =x0+r j cosθ,y j =y0+r j sinθ(θ∈[0,2π)) When performing a re-firing, the laser power is set to P. repair =0.6P nominal Frequency f repair =100kHz; Feedback-based PID dynamic adjustment: Heat conduction equation → Temperature gradient Control input; Design a multivariable PID controller: Input error e(t): Measured deformation Δz meas vs. predicted value Δz pred ; Output adjustment parameters: laser power ΔP, scanning speed Δv; Path replanning strategy: When a porosity ≥2% or a non-fusion defect is detected, the local path replanning module is triggered, and a high-frequency, low-power mode is used for re-printing; when a local defect is detected: the current printing is paused, and a concentric circle filling path for the local area is started; the five-axis attitude is adjusted to the optimal repair angle, and a high-frequency, low-power mode is used for re-printing.

8. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S5, NURBS curves are used to interpolate discrete path points to generate a smooth trajectory, ensuring that the acceleration a on each axis is... i (t) is continuous and its maximum value satisfies max(|a i (t)|)≤0.5rad / s 2 The node vector U and weight factor w of NURBS interpolation are adaptively adjusted according to the path curvature. In high curvature regions, the node spacing Δu ≤ 0.1 mm, and the weighting factor w i =1+0.5|κ i |; In the low curvature region, the node spacing Δu ≤ 0.5 mm, and the weighting factor w i =1; formula: Where q(t) is the five-axis joint spatial trajectory.

9. The intelligent compensation path planning method for five-axis spatial mold additive manufacturing according to claim 1, characterized in that: In step S6, the compensation effect is verified in the digital twin platform, requiring the warpage after compensation to be ≤0.05mm; and the compensation parameters k and K are iteratively updated using a Bayesian optimization algorithm. p ,K i ,K d The objective function is: The Pareto optimal solution is obtained iteratively using the Bayesian optimization algorithm until max(|Δz) is satisfied. meas |)≤0.05mm.

Images