Laser metal curved surface automatic focus processing system and method

Abstract

The present invention discloses a laser metal curved surface automatic focus processing system and method thereof, including: a multimodal data fusion device, a curved surface detection module with a secondary tracking structure, integrated 3D structured light vision and vibration sensing equipment; a dynamic focus correction module, configured with an adaptive energy control unit and a beam shaping component; a nonlinear motion control console, including a six-degree-of-freedom parallel mechanism motion platform, and its servo control system integrates a time-space phase compensation algorithm; a core control unit, the core control unit has an embedded fusion algorithm model, and establishes a nonlinear mapping between material removal efficiency η and multivariable dynamic parameters. The processing quality level of the present invention is improved through the multi-source data fusion of 3D structured light and vibration sensing, the curvature recognition accuracy is improved, and the focus offset is reduced in combination with the axial vibration compensation mechanism, thereby effectively improving the processing quality.

Description

Technical Field

[0001] The present invention belongs to the field of biotechnology, and more specifically, it particularly relates to a laser metal curved surface automatic focus-following processing system. Meanwhile, the present invention also relates to a laser metal curved surface automatic focus-following processing method. Background Art

[0002] Traditional laser surface processing technology has three technical bottlenecks: First, in terms of online detection, single-source sensing systems (such as pure visual detection) are easily disturbed by environmental vibration on complex curvature surfaces, resulting in a curvature recognition error rate of more than 12%, causing focus drift defects; Second, the existing temperature control models mostly use fixed parameter PID control, which is difficult to adapt to the μs-level changes in the molten pool temperature field during high-speed processing, resulting in overheating ablation or poor fusion ratio of more than 7.5%; Third, the six-axis linkage platform has a phenomenon of asynchronous branch movement during dynamic compensation, and the positioning accuracy decay rate under high-frequency conditions is as high as 0.8μm / s;

[0003] In the existing technology, most of them use a single laser displacement sensor for focus tracking. Its insufficient sampling frequency makes it impossible to compensate for vibrations above 300 Hz. Although there is a processing system that introduces temperature closed-loop control, a coupled prediction model for radial and axial temperature gradients has not been established. The known technical inspiration only stays in single-dimensional parameter adjustment, and there is still a lack of effective solutions for multi-physical field collaborative optimization. Therefore, we propose a laser metal surface automatic focus processing system and method to effectively improve the above-mentioned deficiencies. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art, and to propose a laser metal surface automatic follow-focus processing system and method. At the processing quality level, the curvature recognition accuracy is improved through the fusion of multi-source data of 3D structured light and vibration sensing, and the focus offset is reduced by combining the axial vibration compensation mechanism, thereby effectively improving the processing quality.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] Laser metal surface automatic focus processing system, including:

[0007] A multimodal data fusion device, provided with a curved surface detection module of a secondary tracking structure, integrating 3D structured light vision and vibration sensing equipment, wherein the secondary tracking structure comprises a first-level global posture detection array and a second-level local curvature compensation ring; the second-level local curvature compensation ring adopts a rotating off-axis scanning mode;

[0008] A dynamic focus correction module is configured with an adaptive energy control unit and a beam shaping component, wherein the energy control unit uses a progressive temperature field prediction model to control the laser output power;

[0009] A nonlinear motion control console, comprising a six-degree-of-freedom parallel mechanism motion platform, whose servo control system integrates a spatiotemporal phase compensation algorithm, and the algorithm introduces a discrete cosine transform to optimize the posture decoupling process;

[0010] A core control unit, wherein the core control unit has an embedded fusion algorithm model to establish a nonlinear mapping between the material removal efficiency η and the multivariable dynamic parameters:

[0011] ;

[0012] In the formula, is the real-time curvature change rate, δK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th is the processing energy density threshold, and α, β, and γ are multimodal weight factors.

[0013] Preferably, the second-stage local curvature compensation ring is provided with an axial vibration compensation mechanism, which is connected to a vibration sensing device via a microsecond data bus to establish a curvature change rate. The coupling relationship between the compensation adjustment coefficient δK(t) is expressed as:

[0014] ;

[0015] In the formula, μ is the compensation gain coefficient, κ is the phase correction factor, The characteristic frequency range of high-frequency vibration is 300-500Hz.

[0016] Preferably, the progressive temperature field prediction model integrates a distributed Kalman filter, which is dynamically updated every 50 ms, and its state includes the radial temperature gradient Δ and axial temperature gradient Δ The dual variable cooperative control expression is:

[0017] ;

[0018] Where ρ is the laser absorption correction coefficient, σ is the material thermal diffusion response function, and ξ is the phase delay threshold.

[0019] Preferably, the spatiotemporal phase compensation algorithm is specifically:

[0020] 1) Perform cubic spline interpolation processing on the displacement data of each branch chain of the parallel mechanism;

[0021] 2) Use 16th-order discrete cosine transform to extract characteristic frequency components;

[0022] 3) Establish the phase compensation relationship of the vibration sensor data, the expression is:

[0023] ;

[0024] In the formula, F k is the vibration amplitude spectrum of the kth frequency band, Δτ is the lag time compensation, T is the processing cycle, arg(F k ) is F k The phase information of The overall represents the phase rotation factor, which is used to adjust the phase of the signal in the frequency domain.

[0025] Preferably, the multimodal weight factor is dynamically adjusted through a deep reinforcement learning model, specifically:

[0026] Construct a state space S including laser power fluctuation value, spot defocus value and surface fluctuation entropy value;

[0027] Define the action space A as a three-dimensional adjustment vector of (α, β, γ);

[0028] The PPO algorithm is used to update the strategy every 100ms;

[0029] Among them, in the PPO algorithm, the reward function R is expressed as:

[0030] ;

[0031] In the formula, σ r is an empirical constant, ω is the process efficiency weight, and ω is set to 0.7-0.9 during training.

[0032] A laser metal curved surface automatic focus processing method, the method is implemented based on the above system, and comprises the following steps:

[0033] S1. Through the multimodal data acquisition of integrated 3D structured light vision and vibration sensing, a dual-layer dynamic tracking including global posture detection and local curvature compensation is implemented, where the local curvature compensation adopts a rotational off-axis scanning mode;

[0034] S2. Implement dynamic control of laser energy based on the progressive temperature field prediction model, and adjust the beam shaping parameters in combination with the temperature gradient change of the molten pool;

[0035] S3, adopts the motion control strategy with integrated spatiotemporal phase compensation algorithm to optimize the multi-axis coordinated control of the six-DOF parallel mechanism through discrete cosine transform;

[0036] S4. Establish a nonlinear mapping model for material removal efficiency. The dynamic parameters based on real-time updates include:

[0037] ;

[0038] Among them, ρ(t) is the real-time curvature change rate, dK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th (t) is the transient energy density threshold, dz(t) is the normal compensation displacement;

[0039] S5. Dynamically optimize the process parameter weights through the deep reinforcement learning model and build a closed-loop feedback adjustment system.

[0040] Preferably, the local curvature compensation in step S1 establishes an axial vibration coupling model, which satisfies:

[0041] ;

[0042] Among them, μ is the compensation gain coefficient, κ is the phase correction factor, the integration range is 300-500Hz high-frequency vibration frequency band, α(t) is the vibration mode characteristic function, sinc(πκfv) is the sinc function, where κ and is a parameter;

[0043] The dynamic update of the progressive temperature field prediction model in step S2 satisfies:

[0044] ;

[0045] Among them, T r is the radial temperature gradient of the molten pool, T z is the axial temperature gradient, ⊕ represents the tensor convolution operation, τ is the delay time constant, t c is the critical processing time node, Δ sys (t) represents the change at time t.

[0046] Preferably, the spatiotemporal phase compensation algorithm in step S3 comprises:

[0047] S31, performing cubic spline interpolation on the displacement data of each branch of the parallel mechanism to generate a continuous trajectory function Φ(t);

[0048] S32, establishing a frequency domain compensation operator;

[0049] Among them, the expression of the time-space phase compensation algorithm is:

[0050] ;

[0051] Where Ψ(t) is a parabolic time window function, and ⊗ represents a mixed convolution operation in the time and frequency domains.

[0052] Preferably, the nonlinear mapping model adopts a hierarchical fuzzy reasoning strategy, wherein the mapping surface shape function is defined as:

[0053] ;

[0054] In the formula, c i is the cluster center of process parameters, σ i is the width factor of the Gaussian distribution, and the weight coefficient α i satisfy .

[0055] Technical effects and advantages of the present invention: Compared with the prior art, the laser metal surface automatic follow-focus processing system provided by the present invention forms a feedback cascade system through the synergistic mechanism of the secondary tracking structure, and the global posture detection array and the local curvature compensation ring form a feedback cascade system. The coarse adjustment output of the first-stage array provides a spatial reference coordinate system for the second-stage compensation ring, which uses rotary off-axis scanning to generate a dynamic compensation vector field. This mechanism effectively eliminates the error cascade transmission of a single detection source and realizes the decoupling identification of the surface morphology and vibration characteristics; the progressive temperature field prediction model establishes an implicit mapping relationship between the temperature diffusion process and the laser energy density through the covariance analysis of the radial and axial gradients of the molten pool. This mechanism breaks through the traditional single-direction temperature compensation mode and introduces the third-order partial differential coupling equation of the heat flow vector, so that the instantaneous temperature field control has multi-dimensional collaborative constraint capabilities;

[0056] In the control of six-degree-of-freedom platforms, the space-time phase compensation algorithm creatively introduces discrete cosine transform into the multi-axis decoupling process, and forms an analytical manifold space of kinematic parameters by combining frequency domain component extraction with cubic spline interpolation. This mechanism breaks through the singularity limitation of the traditional Euler angle decoupling method and provides a posture update method without cumulative error. The dynamic weight adjustment framework constructed by the deep reinforcement learning model transforms the high-order coupling relationship between process parameters into a topological optimization problem in the state space. This mechanism generates the optimal path for parameter adjustment under multi-dimensional constraints and realizes adaptive gradient tracking of the material removal efficiency surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of the laser metal curved surface automatic focus processing system of the present invention;

[0058] Figure 2 This is a flow chart of the laser metal curved surface automatic follow-focus processing method of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] The present invention provides a laser metal surface automatic focus processing system architecture and method thereof. In terms of processing quality: through the fusion of multi-source data of 3D structured light and vibration sensing, the curvature recognition accuracy is improved, and the focus offset is reduced by combining the axial vibration compensation mechanism; in terms of process stability: the progressive temperature field prediction model introduces a bivariate Kalman filter, and the radial temperature gradient fluctuation range of the molten pool is compressed to ±15°C / mm. No heat accumulation will occur during continuous processing for 15 minutes at a laser power of 1000W; in terms of dynamic response: the phase compensation algorithm based on the 16th-order DCT widens the vibration suppression bandwidth of the six-degree-of-freedom platform to 550Hz, and the time domain step response time is shortened to 5.7ms, which is an improvement compared with traditional PID control;

[0061] Specifically, if Figure 1 , Laser metal surface automatic focus processing system, including:

[0062] The multimodal data fusion device is equipped with a surface detection module with a secondary tracking structure, integrating 3D structured light vision and vibration sensing equipment. The secondary tracking structure includes a first-level global posture detection array and a second-level local curvature compensation ring; the second-level local curvature compensation ring adopts a rotating off-axis scanning mode; the second-level local curvature compensation ring is equipped with an axial vibration compensation mechanism, which is connected to the vibration sensing equipment through a microsecond data bus to establish a curvature change rate The coupling relationship between the compensation adjustment coefficient δK(t) is expressed as:

[0063] ;

[0064] In the formula, μ is the compensation gain coefficient, κ is the phase correction factor, The characteristic frequency range of high-frequency vibration is 300-500Hz;

[0065] A dynamic focus correction module is configured with an adaptive energy control unit and a beam shaping component, wherein the energy control unit uses a progressive temperature field prediction model to control the laser output power;

[0066] The nonlinear motion control console includes a six-degree-of-freedom parallel mechanism motion platform, and its servo control system integrates a time-space phase compensation algorithm. The algorithm introduces discrete cosine transform to optimize the posture decoupling process; further, the time-space phase compensation algorithm is specifically as follows:

[0067] 1) Perform cubic spline interpolation processing on the displacement data of each branch chain of the parallel mechanism;

[0068] 2) Use 16th-order discrete cosine transform to extract characteristic frequency components;

[0069] 3) Establish the phase compensation relationship of the vibration sensor data, the expression is:

[0070] ;

[0071] In the formula, F k is the vibration amplitude spectrum of the kth frequency band, Δτ is the lag time compensation, T is the processing cycle, arg(F k ) is F k The phase information of The overall represents the phase rotation factor, which is used to adjust the phase of the signal in the frequency domain.

[0072] The core control unit has an embedded fusion algorithm model to establish a nonlinear mapping between material removal efficiency η and multivariable dynamic parameters:

[0073] ;

[0074] In the formula, is the real-time curvature change rate, δK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th is the processing energy density threshold, α, β, γ are multimodal weight factors, where the multimodal weight factors are dynamically adjusted through the deep reinforcement learning model, specifically:

[0075] Construct a state space S including laser power fluctuation value, spot defocus value and surface fluctuation entropy value;

[0076] Define the action space A as a three-dimensional adjustment vector of (α, β, γ);

[0077] The PPO algorithm is used to update the strategy every 100ms;

[0078] Among them, in the PPO algorithm, the reward function R is expressed as:

[0079] ;

[0080] In the formula, σ r is an empirical constant, ω is the process efficiency weight, and ω is set to 0.7-0.9 during training.

[0081] It should be noted that the distributed Kalman filter is integrated in the progressive temperature field prediction model, and a dynamic update is performed every 50ms. Its state includes the radial temperature gradient Δ and axial temperature gradient Δ The dual variable cooperative control expression is:

[0082] ;

[0083] Where ρ is the laser absorption correction coefficient, σ is the material thermal diffusion response function, and ξ is the phase delay threshold.

[0084] This embodiment also proposes a laser metal surface automatic focus processing method, which is based on the above system implementation and includes the following steps:

[0085] S1. Through the multi-modal data acquisition of integrated 3D structured light vision and vibration sensing, a double-layer dynamic tracking including global posture detection and local curvature compensation is implemented, wherein the compensation of local curvature adopts a rotating off-axis scanning mode; in step S1, the local curvature compensation establishes an axial vibration coupling model to meet the following requirements:

[0086] ;

[0087] Among them, μ is the compensation gain coefficient, κ is the phase correction factor, the integration range is 300-500Hz high-frequency vibration frequency band, α(t) is the vibration mode characteristic function, sinc(πκfv) is the sinc function, where κ and is a parameter.

[0088] S2. Implement dynamic control of laser energy based on the progressive temperature field prediction model, and adjust the beam shaping parameters in combination with the change of the molten pool temperature gradient; the dynamic update of the progressive temperature field prediction model in step S2 satisfies:

[0089] ;

[0090] Among them, T r is the radial temperature gradient of the molten pool, T z is the axial temperature gradient, ⊕ represents the tensor convolution operation, τ is the delay time constant, t c is the critical processing time node, Δ sys (t) represents the change at time t.

[0091] S3, adopting a motion control strategy integrating a spatiotemporal phase compensation algorithm, and optimizing the multi-axis coordinated control of a six-degree-of-freedom parallel mechanism through discrete cosine transform; the spatiotemporal phase compensation algorithm in step S3 includes:

[0092] S31, performing cubic spline interpolation on the displacement data of each branch of the parallel mechanism to generate a continuous trajectory function Φ(t);

[0093] S32, establishing a frequency domain compensation operator;

[0094] Among them, the expression of the time-space phase compensation algorithm is:

[0095] ;

[0096] Where Ψ(t) is a parabolic time window function, and ⊗ represents a mixed convolution operation in the time and frequency domains.

[0097] S4. Establish a nonlinear mapping model of material removal efficiency, and update the dynamic parameters based on real-time updates including:

[0098] ;

[0099] Among them, ρ(t) is the real-time curvature change rate, dK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th (t) is the transient energy density threshold, dz(t) is the normal compensation displacement; in addition, the nonlinear mapping model adopts a hierarchical fuzzy reasoning strategy, in which the mapping surface shape function is defined as:

[0100] ;

[0101] In the formula, c i is the cluster center of process parameters, σ i is the width factor of the Gaussian distribution, and the weight coefficient α i satisfy .

[0102] S5. Dynamically optimize the process parameter weights through a deep reinforcement learning model to build a closed-loop feedback adjustment system; in step S5, the deep reinforcement learning model adopts a multi-parameter interactive reward mechanism:

[0103] ;

[0104] Among them, ΔP is the process parameter deviation penalty term, satisfaction is the indicator function, tanh is the hyperbolic tangent activation function, R(s,a) represents the reward value for performing action a in state s, and σr is the scaling factor used to adjust the range of the reward value. It represents the logarithmic ratio of the new and old parameters, which is used to quantify the change of parameters. ω represents the weight parameter, which is used to balance the impact of the logarithmic ratio of the new and old parameters and the change of satisfaction.

[0105] Compared with the existing technology, the above has the advantage of entering the school.

[0106] Optimization of multi-physics data fusion mechanism: Through the synergistic mechanism of the secondary tracking structure, the global attitude detection array and the local curvature compensation loop form a feedback cascade system. The coarse adjustment output of the first-stage array provides a spatial reference coordinate system for the second-stage compensation loop, which uses rotary off-axis scanning to generate a dynamic compensation vector field. This mechanism effectively eliminates the error cascade transmission of a single detection source and realizes the decoupled identification of surface morphology and vibration characteristics.

[0107] Molten pool thermodynamic dynamic equilibrium mechanism: The progressive temperature field prediction model establishes an implicit mapping relationship between the temperature diffusion process and the laser energy density through the covariance analysis of the radial and axial gradients of the molten pool. This mechanism breaks through the traditional unidirectional temperature compensation mode and introduces the third-order partial differential coupling equation of the heat flux vector, so that the instantaneous temperature field control has multi-dimensional collaborative constraint capabilities.

[0108] Time-frequency domain hybrid motion compensation mechanism: In the six-degree-of-freedom platform control, the time-space phase compensation algorithm creatively introduces discrete cosine transform into the multi-axis decoupling process. By combining frequency domain component extraction with cubic spline interpolation, an analytical manifold space of kinematic parameters is formed. This mechanism breaks through the singularity limitation of the traditional Euler angle decoupling method and provides a posture update method without cumulative error.

[0109] Nonlinear process parameter matching mechanism: The dynamic weight adjustment framework constructed by the deep reinforcement learning model transforms the high-order coupling relationship between process parameters into a topological optimization problem in the state space. This mechanism generates the optimal path for parameter adjustment under multi-dimensional constraints and realizes adaptive gradient tracking of the material removal efficiency surface.

[0110] Cross-scale vibration suppression propagation mechanism: High-frequency vibration compensation establishes a wave propagation attenuation chain from the sensing layer to the execution layer through the fast response characteristics of the microsecond bus. This mechanism uses the phase intermodulation effect of axial vibration and off-axis scanning to generate active cancellation suppression peaks in the wave vector superposition.

[0111] Multimodal information entropy balance mechanism: The multimodal weight factor dynamic adjustment system is based on the information entropy principle to build a mixed entropy pool model for laser power and defocus parameters. This mechanism forms an adaptive damping effect of parameter fluctuations within the characteristic frequency band through a composite evaluation of Shannon entropy and Kolmogorov complexity.

[0112] In summary, a closed-loop functional chain of "sensing-prediction-execution-feedback" is formed among the various mechanism elements to achieve quantum control-level processing accuracy; a differential homeomorphism mapping of the non-steady-state processing process is established to make the process parameter matching have topological invariance; the variational principle of thermal-mechanical-vibration multi-field coupling is applied to overcome the singularity instability problem of ultrafast laser processing; the cross-layer architecture design of manifold learning and dynamic optimization breaks through the dimensional limitations of traditional linear compensation strategies.

[0113] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Laser metal surface automatic focus processing system, characterized by: include: A multimodal data fusion device, provided with a curved surface detection module of a secondary tracking structure, integrating 3D structured light vision and vibration sensing equipment, wherein the secondary tracking structure comprises a first-level global posture detection array and a second-level local curvature compensation ring; the second-level local curvature compensation ring adopts a rotating off-axis scanning mode; A dynamic focus correction module is configured with an adaptive energy control unit and a beam shaping component, wherein the energy control unit uses a progressive temperature field prediction model to control the laser output power; A nonlinear motion control console, comprising a six-degree-of-freedom parallel mechanism motion platform, whose servo control system integrates a spatiotemporal phase compensation algorithm, and the algorithm introduces a discrete cosine transform to optimize the posture decoupling process; A core control unit, wherein the core control unit has an embedded fusion algorithm model to establish a nonlinear mapping between the material removal efficiency η and the multivariable dynamic parameters: ； In the formula, is the real-time curvature change rate, δK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th is the processing energy density threshold, and α, β, and γ are multimodal weight factors.

2. The laser metal curved surface automatic focus processing system according to claim 1, characterized in that: The second-level local curvature compensation ring is equipped with an axial vibration compensation mechanism, which is connected to the vibration sensing device through a microsecond data bus to establish a curvature change rate. The coupling relationship between the compensation adjustment coefficient δK(t) is expressed as: ； In the formula, μ is the compensation gain coefficient, κ is the phase correction factor, The characteristic frequency range of high-frequency vibration is 300-500Hz.

3. The laser metal curved surface automatic focus processing system according to claim 1, characterized in that: The progressive temperature field prediction model integrates a distributed Kalman filter, which is dynamically updated every 50 ms. Its state includes the radial temperature gradient Δ and axial temperature gradient Δ The dual variable cooperative control expression is: ； Where ρ is the laser absorption correction coefficient, σ is the material thermal diffusion response function, and ξ is the phase delay threshold.

4. The laser metal curved surface automatic focus processing system according to claim 1, characterized in that: The spatiotemporal phase compensation algorithm is specifically: 1) Perform cubic spline interpolation processing on the displacement data of each branch chain of the parallel mechanism; 2) Use 16th-order discrete cosine transform to extract characteristic frequency components; 3) Establish the phase compensation relationship of the vibration sensor data, the expression is: ； In the formula, F k is the vibration amplitude spectrum of the kth frequency band, Δτ is the lag time compensation, T is the processing cycle, arg(F k ) is F k The phase information of The overall represents the phase rotation factor, which is used to adjust the phase of the signal in the frequency domain.

5. The laser metal curved surface automatic focus processing system according to claim 1, characterized in that: The multimodal weight factor is dynamically adjusted through a deep reinforcement learning model, specifically: Construct a state space S including laser power fluctuation value, spot defocus value and surface fluctuation entropy value; Define the action space A as a three-dimensional adjustment vector of (α, β, γ); The PPO algorithm is used to update the strategy every 100ms; Among them, in the PPO algorithm, the reward function R is expressed as: ； In the formula, σ r is an empirical constant, ω is the process efficiency weight, and ω is set to 0.7-0.9 during training.

6. A laser metal surface automatic focus processing method, the method is implemented based on the system according to any one of claims 1 to 5, characterized in that: The steps include: S1. Through the multimodal data acquisition of integrated 3D structured light vision and vibration sensing, a dual-layer dynamic tracking including global posture detection and local curvature compensation is implemented, where the local curvature compensation adopts a rotational off-axis scanning mode; S2. Implement dynamic control of laser energy based on the progressive temperature field prediction model, and adjust the beam shaping parameters in combination with the temperature gradient change of the molten pool; S3, adopts the motion control strategy with integrated spatiotemporal phase compensation algorithm to optimize the multi-axis coordinated control of the six-DOF parallel mechanism through discrete cosine transform; S4. Establish a nonlinear mapping model for material removal efficiency. The dynamic parameters based on real-time updates include: ； Among them, ρ(t) is the real-time curvature change rate, dK(t) is the axial compensation adjustment coefficient, V adj is the vibration compensation strength, E th (t) is the transient energy density threshold, dz(t) is the normal compensation displacement; S5. Dynamically optimize the process parameter weights through the deep reinforcement learning model and build a closed-loop feedback adjustment system.

7. The method for automatically focusing laser metal curved surface according to claim 6, characterized in that: The local curvature compensation in step S1 establishes an axial vibration coupling model that satisfies: ； Wherein, μ is the compensation gain coefficient, κ is the phase correction factor, the integration range is 300-500Hz high frequency vibration band, α(t) is the vibration mode characteristic function, sinc(πκfv) is the sinc function, where κ and is a parameter; The dynamic update of the progressive temperature field prediction model in step S2 satisfies: ； Among them, T r is the radial temperature gradient of the molten pool, T z is the axial temperature gradient, ⊕ represents the tensor convolution operation, τ is the delay time constant, t c is the critical processing time node, Δ sys (t) represents the change at time t.

8. The method for automatically focusing laser metal curved surface according to claim 6, characterized in that: The spatiotemporal phase compensation algorithm in step S3 includes: S31, performing cubic spline interpolation on the displacement data of each branch of the parallel mechanism to generate a continuous trajectory function Φ(t); S32, establishing a frequency domain compensation operator; Among them, the expression of the time-space phase compensation algorithm is: ； Where Ψ(t) is a parabolic time window function, and ⊗ represents a mixed convolution operation in the time and frequency domains.

9. The method for automatically focusing laser metal curved surface according to claim 6, characterized in that: The nonlinear mapping model adopts a hierarchical fuzzy reasoning strategy, where the mapping surface shape function is defined as: ； In the formula, c i is the cluster center of process parameters, σ i is the width factor of the Gaussian distribution, and the weight coefficient α i satisfy .

Images