Sports kettle trepanning method and system based on laser cutting technology

Abstract

The invention relates to the technical field of laser processing, and discloses a sports kettle trepanning method and system based on a laser cutting technology. The method comprises the steps that through three-dimensional point cloud scanning and local curvature calculation, reinforcement learning dynamic path planning, transient thermodynamic coupling parameter optimization, deformable mirror-mechanical arm cooperative control and infrared temperature field feedback closed loop, high-precision cutting and thermal damage suppression of the complex curved surface are achieved, and path re-planning is triggered when deviation exceeds the limit. The system comprises a laser ranging sensor array module, a reinforcement learning decision module, a thermodynamic coupling solver module, a deformable mirror and six-axis mechanical arm cooperative control module and a multi-sensor feedback module. Through curvature self-adaptive laser processing, multi-physical field thermal damage cooperative control, optical-mechanical dynamic focusing and intelligent closed-loop optimization, high-precision cutting of the complex curved surface is achieved, the calculation efficiency and the system stability are both considered, and the processing quality and adaptability are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of laser processing, and specifically to a method and system for opening holes in a sports kettle based on laser cutting technology. Background Art

[0002] As a core component of outdoor sports equipment, the lightweight design and functional hole-opening requirements of sports kettles pose strict requirements on processing technology. Traditional mechanical cutting methods rely on rigid tools and preset paths. When facing the complex double-curved geometric features of kettles, defects such as burrs and cracks often occur at the hole-opening edges due to the mismatch between the tool trajectory and the surface curvature, and it is difficult to achieve the processing of special-shaped holes with micron-level accuracy.

[0003] Although existing laser cutting technologies have the advantage of non-contact processing, there are still significant limitations in complex surface applications: fixed-focus optical systems are difficult to adapt to the defocusing effect caused by the change of surface curvature, resulting in uneven distribution of laser energy density, causing fluctuations in cutting depth or expansion of the heat-affected zone;

[0004] At the same time, traditional processes treat path planning and thermodynamics control separately, lacking the ability to optimize multi-physical field coupling, resulting in fixed processing parameters and being unable to dynamically respond to differences in material heat conduction characteristics or environmental disturbances. In addition, existing systems generally lack a real-time closed-loop feedback mechanism. When temperature field anomalies or geometric deviations occur during the processing, only manual intervention and shutdown adjustment can be relied on, seriously affecting processing efficiency and product consistency. With the development of sports equipment towards personalization and functional integration, developing a laser hole-opening technology with both high-precision geometric adaptability and intelligent thermal damage suppression ability has become a key requirement to break through the industry bottleneck.

[0005] Therefore, the present invention proposes a method and system for opening holes in a sports kettle based on laser cutting technology to solve the deficiencies of the existing technology. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the present invention provides a method and system for opening holes in a sports kettle based on laser cutting technology, which solves the problems of poor curvature adaptability, uncontrollable thermal damage, and insufficient real-time dynamic adjustment ability in the opening of holes in complex surfaces of sports kettles.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for opening holes in a sports kettle based on laser cutting technology includes the following steps:

[0008] S1. Scan the surface of the kettle through a laser ranging sensor array to generate three-dimensional point cloud data, and calculate the local curvature of each point in the three-dimensional point cloud data based on differential geometry formulas;

[0009] S2. According to the local curvature, use a reinforcement learning model to dynamically plan the laser cutting path, where the state space of the reinforcement learning model includes curvature, cutting speed, and focal length, and the action space is the speed increment and the focal length increment;

[0010] S3. Based on the local curvature distribution and the transient thermodynamics coupling model, generate an optimized parameter combination of laser power and cutting speed. The transient thermodynamics coupling model solves the heat conduction equation through finite element discretization and dynamically adjusts the heat source distribution according to the curvature;

[0011] S4. According to the optimized parameter combination, control the deformable mirror to adjust the focusing position of the laser beam, and synchronously correct the end pose of the six-axis robotic arm through Lie group mapping;

[0012] S5. Perform laser cutting, and use an infrared thermal imager to collect the temperature field data of the processing area in real time. If the deviation between the measured temperature field and the model prediction value exceeds the preset threshold, trigger the reinforcement learning model to re-plan the path.

[0013] Preferably, in step S1, after generating three-dimensional point cloud data by scanning the surface of the kettle with a laser ranging sensor array, calculate the local curvature of each point based on differential geometry formulas, specifically:

[0014] For each point, fit a parametric curve in the local neighborhood, and calculate the curvature value through the vector cross product operation of the first derivative and the second derivative of the curve.

[0015] Preferably, in step S2, the reward function of the reinforcement learning model is defined as the weighted difference between the curvature fitting degree evaluation function and the entropy increment of the heat affected zone, where the weight coefficient is used to balance the optimization objectives of path accuracy and heat damage control.

[0016] Preferably, the entropy increment of the heat affected zone is calculated by the transient thermodynamics coupling model, specifically as the integral of the entropy change of the material density, specific heat capacity at constant pressure, and the predicted temperature field relative to the ambient temperature, and the integration domain covers the spatial range of the heat affected zone.

[0017] Preferably, in step S3, the heat conduction equation of the transient thermodynamics coupling model is discretized into a linear equation system of a heat capacity matrix and a heat conduction matrix by the finite element method, and its heat source term vector is dynamically generated according to the curvature distribution, specifically:

[0018] Superimpose the laser power distribution coefficient at each curvature sampling point with the Gaussian heat source model to generate a spatial energy distribution field related to the curvature, and map it to the finite element grid nodes.

[0019] Preferably, the grid density of the finite element discretization is dynamically adjusted based on the curvature gradient and the cutting speed.

[0020] Specifically: in areas with a large curvature gradient or a cutting speed lower than the minimum threshold, the grid density is proportionally increased, and the adjustment amplitude is determined by the ratio of the curvature gradient vector to the speed parameter.

[0021] Preferably, in step S4, when controlling the deformable mirror to adjust the focusing position of the laser beam, the mirror deformation phase distribution of the deformable mirror is calculated according to the curvature distribution, specifically:

[0022] The curvature data is orthogonally projected with the Zernike basis function within the effective area of the mirror surface to generate Zernike coefficients, and the deformable mirror is driven to generate a wavefront compensation phase matching the surface curvature.

[0023] Preferably, in the Lie group mapping, the end pose correction amount of the six-axis robotic arm is generated through the following steps: calculating the joint angle correction amount based on the linear combination of the Zernike coefficients and the preset weight matrix, and applying the correction amount to the initial pose matrix of the robotic arm through the Lie group exponential mapping to achieve the dynamic coordination of the end pose and the laser focusing position.

[0024] Preferably, in step S5, the deviation determination condition between the measured temperature field and the model prediction value is: when the overall deviation amount between the measured temperature field vector and the predicted temperature field vector exceeds the preset threshold, the reinforcement learning model is triggered to re-plan the laser cutting path.

[0025] The present invention also provides a sports kettle hole-opening system based on laser cutting technology, and the system includes:

[0026] A laser ranging sensor array module for scanning the surface of the kettle and generating three-dimensional point cloud data;

[0027] A reinforcement learning decision module for dynamically planning the laser cutting path according to the curvature data;

[0028] A thermodynamics coupling solver module for solving the transient heat conduction equation and optimizing the laser power and cutting speed parameters;

[0029] A deformable mirror and six-axis robotic arm collaborative control module for adjusting the laser focusing position and correcting the robotic arm pose according to the optimized parameters;

[0030] A multi-sensor feedback module for real-time collecting temperature field data and triggering path re-planning.

[0031] The present invention provides a sports kettle hole-opening method and system based on laser cutting technology. It has the following beneficial effects:

[0032] 1. The present invention uses a laser ranging sensor array and a curvature analysis algorithm to obtain the three-dimensional geometric features of the kettle surface in real time, and combines a reinforcement learning model to dynamically plan the cutting path, effectively solving the problem of path deviation caused by curvature changes in traditional methods, and significantly improving the geometric accuracy and consistency of complex curved surface cutting.

[0033] 2. The present invention optimizes the laser power and cutting speed parameters based on a transient thermodynamics coupling model, and combines an infrared thermal imager to real-time feedback the temperature field data, realizing precise control of the heat-affected zone range and energy distribution, avoiding overheating and melting of materials or thermal stress cracks, and greatly improving the processing quality and product reliability.

[0034] 3. The present invention uses a combined driving mechanism of a deformable mirror to adjust the laser focus position and a six-axis robotic arm pose correction to ensure that the laser beam axis is always aligned with the surface normal, overcoming the defocusing effect of traditional fixed focal length systems in non-planar processing, and enhancing the energy utilization efficiency and process stability of complex curved surface cutting.

[0035] 4. The present invention introduces a closed-loop mechanism of a reinforcement learning model and multi-sensor real-time feedback, which can dynamically detect the temperature field deviation during the processing and trigger path replanning, adaptively compensating for uncertainties such as material property fluctuations and equipment errors, and improving the system's adaptability to complex working conditions.

[0036] 5. The present invention realizes the efficient processing and real-time control of large-scale data while ensuring the accuracy of the transient thermodynamics model through dynamic grid encryption, Lie group mapping pose solution, and parallel computing architecture, meeting the high real-time and high stability requirements of complex curved surface laser processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the flowchart of the method of the present invention;

[0038] Figure 2 is the system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] Please refer to Figure 1 , the embodiments of the present invention provide a method for opening holes in a sports kettle based on laser cutting technology, including the following steps:

[0041] S1. Scan the surface of the kettle through a laser ranging sensor array to generate three-dimensional point cloud data, and calculate the local curvature of each point in the three-dimensional point cloud data based on differential geometry formulas;

[0042] In this embodiment, scanning the surface of the kettle through a laser ranging sensor array to generate three-dimensional point cloud data, and calculating the local curvature of each point based on differential geometry formulas, the specific implementation method is as follows:

[0043] In this embodiment, the laser ranging sensor array is composed of multiple line array laser ranging sensors, and the sensors perform non-contact measurement on the surface of the kettle along a preset scanning trajectory. Preferably, the scanning trajectory is driven by a robotic arm or a rotating platform, so that the sensor array translates and scans along the axis direction of the kettle, and at the same time the kettle rotates uniformly around its axis to ensure that the laser beam covers the entire circumference of the kettle surface. By measuring the flight time (Time-of-Flight, ToF) or phase shift of the laser beam, the three-dimensional spatial coordinates of each measurement point on the kettle surface are obtained, and a point cloud data set P = {p i =(x i , y i , z i )} is generated, where x i , y i , z i respectively represent the coordinate values of point p i in the Cartesian coordinate system, i = 1, 2,..., N is the point cloud data index, and N is the total number of points.

[0044] For each point p i in the point cloud data set, a parametric curve needs to be fitted within its local neighborhood to calculate the curvature. Specifically, with p i as the center point, k adjacent points {p i1 , p i2 ,..., p ik} are selected according to a preset neighborhood radius R or by the k-nearest neighbors (k-NN) algorithm, where k is the preset number of neighborhood points (for example, k = 15). Based on the selected adjacent points, a parametric curve r(s) is fitted using the least squares method, where s is the arc length parameter. Preferably, the parametric curve is a cubic B-spline curve, and its expression is:

[0045]

[0046] Among them, is the cubic B-spline basis function; c j is the control point coordinate; n is the number of control points. The control points {c j} are solved by minimizing the least squares optimization objective function , where s mIs the corresponding value after arc length parameterization.

[0047] Based on differential geometry theory, curvature characterizes the local bending degree of a curve, and its calculation depends on the first derivative and second derivative of the parameterized curve. In this embodiment, the curvature κ i The calculation formula is:

[0048]

[0049] Where r′(s i ) is the first derivative vector (tangent vector) of the parameterized curve r(s) at point p i , representing the tangent direction of the curve at this point, and its modulus ||r′(s i )|| is the instantaneous change rate of the curve at arc length parameter s i ; r″(s i ) is the second derivative vector (normal vector) of the parameterized curve r(s) at point p i , reflecting the change rate of the tangent vector; × is the vector cross product operator, used to calculate the cross product of two vectors, and the modulus of its result is equal to the area of the parallelogram formed by the two vectors; ||·|| is the Euclidean norm (modulus) of the vector.

[0050] Calculation of the first derivative r′(s i ): Obtained by analytically differentiating or numerically differentiating the parameterized curve r(s). For a cubic B-spline curve, its derivative expression is:

[0051]

[0052] Substitute s i at point p i to obtain r′(s i ).

[0053] Calculation of the second derivative r″(s i ): Further differentiate the first derivative r′(s) to get:

[0054]

[0055] Cross product and normalization: Calculate the modulus of the cross product r′(s i )×r″(s i ), and perform normalization processing through the denominator term ||r′(s i )|| 3 to eliminate the influence of parameterization speed on curvature calculation and ensure that the curvature is a geometric invariant quantity.

[0056] S2. According to the local curvature, use a reinforcement learning model to dynamically plan the laser cutting path, where the state space of the reinforcement learning model includes curvature, cutting speed, and focal length, and the action space is the speed increment and the focal length increment;

[0057] In this embodiment, according to the local curvature data, use a reinforcement learning model to dynamically plan the laser cutting path. The specific implementation method is as follows:

[0058] In this embodiment, the reinforcement learning model is constructed based on the Markov Decision Process (MDP) framework. Its core elements include the state space, action space, reward function, and state transition dynamics. The model interacts with the environment (i.e., the laser cutting processing system) in real time to learn the optimal strategy to dynamically adjust the cutting path parameters and achieve the collaborative optimization of path accuracy and thermal damage control.

[0059] The state space S is defined as the set of observable physical quantities during the processing, specifically including:

[0060] Curvature κ: The local curvature value of the current cutting path point, which comes from the curvature distribution data set κ = {κ i} generated in step S1. Its value reflects the local bending degree of the surface, and the unit is mm -1 ;

[0061] Cutting speed v: The instantaneous speed at which the laser head moves along the current path, which is measured in real time by an encoder or a laser Doppler velocimeter, and the unit is millimeters per second (mm / s);

[0062] Focal length f: The vertical distance between the focus of the laser beam and the surface of the kettle, which is feedback-controlled by the laser optical system, and the unit is millimeters (mm). Its change directly affects the laser energy density distribution.

[0063] The state vector at time step t is represented as s t =(κ t , v t , f t ), and its dynamic update is driven by action execution and environmental feedback.

[0064] The action space A is the set of control instructions that the agent can output, specifically including:

[0065] Speed increment Δv: The adjustment amount for the current cutting speed, and its value range is determined by the physical performance limitations of the laser and the robotic arm. For example, Δv ∈ [-v max , +v max , and the unit is millimeters per second (mm / s);

[0066] Focal length increment Δf: The adjustment amount for the current laser focal length, limited by the deformation range of the deformable mirror and the focal depth of the optical system. For example, Δf ∈ [-f max , +f max , with the unit of millimeter (mm).

[0067] The action vector is represented as a t = (Δv t , Δf t ) at time step t. After its execution, it drives the laser and the robotic arm through the control system interface, causing the system state to transfer from s t to s t+1 .

[0068] The reward function R of the reinforcement learning model is designed in the form of multi-objective optimization, and its mathematical expression is:

[0069] R = α·Re(κ, κ target ) - β·ΔS HAZ ;

[0070] where Re(κ, κ target ) is the path fitting degree evaluation function; it quantifies the matching degree between the current curvature κ and the target curvature κ target . Preferably, this function uses cosine similarity calculation:

[0071]

[0072] where κ target is the curvature value of the target path, generated according to the preset processing accuracy requirements; ||·|| represents the Euclidean norm of the vector;

[0073] ΔS HAZ is the entropy increment of the heat affected zone, calculated by the transient thermodynamics coupling model in step S3, with the unit of joule per kelvin (J / K);

[0074] α and β are weight coefficients, which are dimensionless positive real numbers used to balance the optimization objectives of path fitting degree and heat damage control. Preferably, their initial values are set by domain experience and dynamically adjusted during the training process.

[0075] The entropy increment ΔS HAZ,t is based on the second law of thermodynamics and characterizes the energy dissipation caused by the irreversible change of the temperature field during the processing. Its calculation formula is:

[0076]

[0077] where ρ is the density of the kettle material, with the unit of kilogram per cubic meter (kg / m 3 ), obtained from the material property database; c pCp is the specific heat capacity at constant pressure of the material, with the unit of joule per kilogram per kelvin (J / (kg·K)), and is measured by differential scanning calorimetry (DSC).

[0078] T1 is the temperature field distribution predicted in the thermodynamic model, representing the temperature value at position x at time t, with the unit of kelvin (K).

[0079] T0 is the ambient temperature, which is collected in real time by a temperature and humidity sensor, with the unit of kelvin (K).

[0080] Ω1 is the spatial integration domain of the heat affected zone, which is dynamically determined by the laser action area and the heat diffusion range, and its boundary is delimited by the temperature threshold T threshold = T0 + ΔT, where ΔT is the preset temperature rise tolerance.

[0081] Q-learning update rule: The Q-value function is updated using the value iteration algorithm based on the Bellman equation, and its mathematical expression is:

[0082]

[0083] where η is the learning rate, with the value range of 0 < η ≤ 1, which controls the influence degree of new information on the Q value; γ is the discount factor, with the value range of 0 ≤ γ < 1, which weighs the importance of the current reward and the future reward.

[0084] Deep Q-network (DQN) extension: When the state space dimension is high, a deep neural network is used to approximate the Q-value function, and its loss function is defined as:

[0085]

[0086] where θ is the neural network weight parameter; θ - is the target network weight parameter, which is copied from the online network regularly. is the experience replay pool, which stores historical transition samples (s, a, r, s′).

[0087] Exploration strategy design: The ε-greedy strategy is used to balance exploration and exploitation, and its action selection probability is:

[0088]

[0089] where ∈ decays with the number of training steps, for example, ∈ = ∈ 初始 ·e -kt , and k is the decay coefficient.

[0090] S3. Generate an optimized parameter combination of laser power and cutting speed based on the local curvature distribution and transient thermodynamics coupling model. The transient thermodynamics coupling model solves the heat conduction equation through finite element discretization and dynamically adjusts the heat source distribution according to the curvature.

[0091] In this embodiment, an optimized parameter combination of laser power and cutting speed is generated based on the local curvature distribution and transient thermodynamics coupling model. The specific implementation method is as follows:

[0092] In this embodiment, the transient thermodynamics coupling model is constructed based on the law of conservation of energy. Its control equation is a three-dimensional unsteady heat conduction equation, which completely describes the spatio-temporal evolution of the temperature field on the surface of the kettle during the laser cutting process. The differential form of the equation in the Cartesian coordinate system is:

[0093]

[0094] where ρ is the density of the kettle material, obtained from the material property database, with the unit of kilograms per cubic meter (kg / m 3 ); c p is the specific heat capacity at constant pressure of the material, calibrated by differential scanning calorimetry (DSC), with the unit of joules per (kilogram·kelvin) (J / (kg·K)); λ is the thermal conductivity of the material, measured by the steady-state heat flow method, with the unit of watts per (meter·kelvin) (W / (m·K)); T1 is the temperature field, which is a function of the spatial coordinates x = (x, y, z) and time t, with the unit of kelvin (K); Q laser is the laser heat source term, describing the spatial distribution of laser energy input; h c is the convective heat transfer coefficient, calibrated by fluid dynamics simulation or experiment, with the unit of watts per (square meter·kelvin) (W / (m 2 ·K)); T0: the ambient temperature, collected in real time by a temperature and humidity sensor, with the unit of kelvin (K); ∈ is the emissivity of the material surface, measured by an infrared spectrometer; σ SB is the Stefan-Boltzmann constant, with a value of 5.67×10 -8 .

[0095] Spatially discretize the above equation by the finite element method. The specific steps include:

[0096] Geometric discretization: Discretize the kettle wall into a dynamic mesh composed of tetrahedral or hexahedral elements, and record the coordinates of the element nodes as x j =(x j , y j , z j ).

[0097] Shape function selection: Adopt the linear Lagrangian shape function N j(x)Describe the approximate solution of the temperature distribution within the element:

[0098]

[0099] where N node is the number of element nodes; T j (t) is the temperature value of node j at time t.

[0100] Weak form derivation: The governing equation is transformed into the weak form by the Galerkin weighted residual method and integrated over the element to obtain the global matrix equation:

[0101]

[0102] where C is the heat capacity matrix, calculated by the element integral C ij = ∫ Ω ρc p N i N j dV; K = K + K conv + K rad , K is the heat conduction matrix, calculated by the element integral ; K conv is the convective heat dissipation matrix, calculated by the boundary integral K conv,ij = ∫ Γ h c N i N j dS; K rad is the radiative heat dissipation matrix, and the non - linear term is linearized by Newton - Raphson iteration; T2 is the node temperature vector; Q is the heat source term vector, generated by coupling the laser energy input and the curvature distribution.

[0103] Each element Q of the heat source term Q j represents the equivalent laser power input at node j, which is defined as the superposition of multiple moving Gaussian heat sources:

[0104]

[0105] where P i is the laser power distribution coefficient associated with the curvature κ i , and the calculation formula is P i = k p ·κ i + P base , where k p is the power - curvature proportionality coefficient; P base is the reference power, determined by calibration experiments; x j is the spatial coordinate of node j, in meters (m); p i is the coordinate of the curvature sampling point i, from the curvature data set P = {p of step S1i} is provided; σ is the Gaussian heat source radius, and the calculation formula is σ = σ0 / κ i , where σ0 is the reference radius to ensure that the heat source is more concentrated in the high-curvature region; M is the number of curvature sampling points, which is consistent with the scale of the point cloud data in step S1.

[0106] To balance the calculation efficiency and accuracy, the finite element mesh density is dynamically adjusted according to the curvature gradient and cutting speed. The specific rules are as follows:

[0107]

[0108] where Δd j is the change in mesh density at node j, with the unit of cells / m. A positive value indicates densification, and a negative value indicates sparsification; is the curvature gradient vector at node j, calculated by the central difference method where Δx is the node spacing; v is the current cutting speed, output by the reinforcement learning model in step S2, with the unit of m / s; v min is the minimum speed threshold to avoid a zero denominator, and the value is v min = 0.1 mm / s; k mesh is the mesh adjustment coefficient, determined by mesh convergence analysis.

[0109] S4. According to the optimized parameter combination, control the deformable mirror to adjust the focusing position of the laser beam, and synchronously correct the end pose of the six-axis robotic arm through Lie group mapping;

[0110] In this embodiment, according to the optimized parameter combination, control the deformable mirror to adjust the focusing position of the laser beam, and synchronously correct the end pose of the six-axis robotic arm through Lie group mapping. The specific implementation method is as follows:

[0111] Curvature-driven modeling of the phase distribution of the deformable mirror

[0112] In this embodiment, the phase distribution φ(x,y) of the mirror surface deformation of the deformable mirror is generated by driving with curvature distribution data, and its mathematical expression is a linear combination of Zernike polynomial basis functions:

[0113]

[0114] where Z n (x,y) is the nth Zernike basis function, defined as an orthogonal polynomial in polar coordinates (r,θ), and its specific form is:

[0115]

[0116] where is the radial polynomial, satisfying the orthogonality condition:

[0117] ∫∫ Ω Z n (x,y)Z m (x,y)dxdy = δ nm ;

[0118] Here, n is the Zernike mode order; m is the angular frequency, is the normalized radial coordinate; θ = arctan(y / x) is the angular coordinate, δ nm is the Kronecker function.

[0119] a n is the n-th Zernike coefficient, calculated by the projection integral of the curvature distribution κ(x,y) over the effective area Ω2 of the mirror surface:

[0120]

[0121] where, Ω2 is the effective optical area of the deformable mirror, whose shape matches the laser spot size, usually a circular area, and the radius R is determined by the laser beam waist radius. For example, Ω2 = {(x,y)|x 2 + y 2 ≤ R 2}.

[0122] Through the above mapping, the curvature distribution κ(x,y) is converted into the mirror surface deformation phase φ(x,y), enabling the laser focus position to dynamically adapt to the surface curvature change. The phase gradient corresponding to the high curvature region (such as the surface edge) is larger, thus driving the deformable mirror to generate a larger local deformation to compensate for the laser defocus caused by the surface geometric distortion.

[0123] The end pose correction of the six-axis robotic arm is based on the exponential mapping theory of the Lie group SE(3), converting the deformable mirror phase distribution parameters into the robotic arm joint angle adjustment amounts. The specific process is as follows:

[0124] Definition of the Lie algebra generator: Each degree of freedom of the robotic arm corresponds to a Lie algebra generator G k ∈ se(3), and its matrix form is:

[0125]

[0126] where, is the skew-symmetric matrix of the rotation component, determined by the rotation direction of the robotic arm joint axis. For example, for the joint rotating around the x-axis; ω k = [1,0,0] T , and the corresponding skew-symmetric matrix is:

[0127]

[0128] is the translational component vector, which is calculated from the lengths of the robotic arm links and joint positions. The subscript k = 1, 2, …, 6 corresponds to the six joints of the robotic arm.

[0129] Calculation of joint angle correction: The joint angle correction θ of each joint k is linearly weighted by the Zernike coefficients:

[0130]

[0131] where w kn is a preset weight coefficient matrix, which is determined through a calibration experiment. Specifically, by applying a unit Zernike mode excitation, measuring the pose offset of the robotic arm end, and fitting the weight coefficients using the least squares method to ensure an accurate match of the motion coupling relationship of the optical-mechanical system. N is the total number of Zernike modes, and usually the first 36 terms are selected to cover the main aberration modes.

[0132] End pose update: The joint angle correction is converted into the pose matrix of the end effector through the Lie group exponential map:

[0133]

[0134] where T3 is the initial pose matrix of the robotic arm, which is calculated from the forward kinematics model and is expressed as:

[0135]

[0136] where R0 ∈ SO(3) is the initial rotation matrix; is the initial translation vector.

[0137] exp(·) is the Lie group exponential map operator, which converts the Lie algebra element into a rigid body transformation matrix in the SE(3) group. The specific calculation uses the Padé approximation or Taylor series expansion:

[0138]

[0139] where is the matrix representation of the Lie algebra element.

[0140] S5. Perform laser cutting, and collect the temperature field data of the processing area in real time through an infrared thermal imager. If the deviation between the measured temperature field and the model prediction value exceeds the preset threshold, the reinforcement learning model is triggered to re-plan the path;

[0141] In this embodiment, laser cutting is performed and the temperature field of the processing area is monitored in real time through an infrared thermal imager. If the deviation between the measured data and the model prediction exceeds the limit, path replanning is triggered. The specific implementation method is as follows:

[0142] In this embodiment, during the laser cutting process, a non-contact infrared thermal imager is used to monitor the real-time temperature field of the processing area. The infrared thermal imager captures the thermal radiation signal at a preset sampling frequency (e.g., 30 Hz), and converts it into temperature field data T through the Stefan-Boltzmann law. real ={Treal,j}. Preferably, the spectral response range of the thermal imager covers the 8-14 μm band, and the spatial resolution matches the laser spot diameter to ensure that the temperature field data is aligned with the positions of the finite element mesh nodes in step S3.

[0143] Define the measured temperature field T real and the predicted temperature field T model ={T model,j} in the transient thermodynamics model in step S3, and the deviation between them is the Euclidean norm:

[0144]

[0145] where N node is the number of finite element mesh nodes; T real,j and T model,j are the measured and predicted temperature values of node j, respectively, with the unit of Kelvin (K).

[0146] Set the temperature field deviation threshold ε T as the trigger condition, and its value is determined based on the material thermal damage tolerance and the model confidence. When the following formula is satisfied, the reinforcement learning model in step S2 is triggered to re-plan the path:

[0147] ||T real -T model ||2>∈ T ;

[0148] Preferably, ε T is dynamically adjusted through a calibration experiment: in the initial processing stage, a small batch of trial cutting experiments are carried out, and the root mean square (RMS) value of the model prediction error is statistically calculated, and ε T =k safe ·RMS, where k safe ≥1 is a safety factor.

[0149] Please refer to Figure 2 , the present invention also provides a sports kettle hole-opening system based on laser cutting technology, and the system includes:

[0150] Laser ranging sensor array module. This module is composed of multiple linear array laser ranging sensors, and obtains the three-dimensional geometric information of the kettle surface through non-contact scanning. The sensor array translates and scans along the axis of the kettle in a preset trajectory, and at the same time combines with a rotating platform to drive the kettle to rotate uniformly around the axis to ensure that the laser beam covers the entire circumference of the surface. During the scanning process, the spatial coordinates of each sampling point are calculated in real time based on the time-of-flight measurement principle, generating a high-precision three-dimensional point cloud dataset, providing a geometric reference for subsequent curvature analysis and path planning. Preferably, the sensor array is equipped with an adaptive exposure adjustment function, which can dynamically optimize the measurement signal-to-noise ratio according to the surface reflectivity, avoiding data distortion in high-reflective or dark areas.

[0151] Reinforcement learning decision-making module. This module adopts a deep reinforcement learning framework, constructs a state space with curvature, cutting speed, and focal length as core parameters, and outputs speed increment and focal length adjustment amount through the action space. During the training process, the path fitting degree and thermal damage control index are synchronously optimized through a multi-objective reward function, and the model generalization ability is improved by combining the experience replay mechanism and the dynamic exploration strategy. The decision-making module receives the curvature distribution data and environmental feedback signals in real time, and online adjusts the cutting path parameters to ensure the dynamic adaptation of the laser energy distribution and the motion trajectory on complex curved surfaces. Preferably, the module integrates a transfer learning mechanism, which can quickly adapt to kettle workpieces with different materials or geometric features based on historical processing data.

[0152] Thermodynamic coupling solver module. This module constructs a transient heat conduction numerical model based on the finite element method to simulate the temperature field evolution and heat affected zone diffusion behavior during the laser cutting process. Through the dynamic mesh encryption technology, the mesh resolution is automatically increased in areas with large curvature gradients or low cutting speeds, balancing the calculation efficiency and prediction accuracy. The solver receives the laser power, cutting speed, and curvature distribution parameters in real time, iteratively optimizes the heat source distribution pattern, and outputs an energy-speed parameter combination that meets the thermal damage constraints. Preferably, the module integrates a material property adaptive calibration function, which can reverse-correct key parameters such as thermal conductivity and specific heat capacity according to the infrared temperature measurement data.

[0153] Deformable mirror and six-axis robotic arm collaborative control module. This module realizes the synchronous adjustment of the laser focusing position and the processing posture through an optical-mechanical joint control strategy. The deformable mirror drives the mirror surface deformation based on the curvature distribution data, uses the Zernike polynomial decomposition technology to generate a phase distribution that matches the surface geometry, and dynamically adjusts the laser focus position to compensate for the defocus effect. The six-axis robotic arm converts the phase parameters into joint angle correction amounts through the Lie group mapping theory, and combines the inverse kinematics solution to adjust the pose of the end effector in real time to ensure that the axis of the laser beam is always aligned with the surface normal. Preferably, the module is built-in with a closed-loop feedback unit, which monitors the focus offset in real time through a laser interferometer and dynamically calibrates the collaborative control parameters of the deformable mirror and the robotic arm.

[0154] Multi-sensor feedback module. This module integrates an infrared thermal imager, a position encoder, and an optical detection unit to build a multi-dimensional data acquisition network. The infrared thermal imager captures the temperature field distribution of the processing area at a high frame rate and transmits the data to the control center in real time through a wireless communication protocol. When it is detected that the deviation between the measured temperature field and the model prediction value exceeds the limit, an exception handling protocol is triggered to pause the current processing flow and start the reinforcement learning model re-planning mechanism. Preferably, the module is equipped with a data fusion engine to synchronously analyze multi-modal signals such as temperature, vibration, and light intensity, realize the comprehensive evaluation of the processing state and fault pre-diagnosis, and provide a decision basis for system self-healing.

[0155] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for opening holes in a sports kettle based on laser cutting technology, characterized in that, Including the following steps: S1. Scanning the surface of the kettle through a laser ranging sensor array to generate three-dimensional point cloud data, and calculating the local curvature of each point in the three-dimensional point cloud data based on differential geometry formulas; S2. According to the local curvature, using a reinforcement learning model to dynamically plan the laser cutting path, where the state space of the reinforcement learning model includes curvature, cutting speed, and focal length, and the action space is the speed increment and the focal length increment; S3. Generating an optimized parameter combination of laser power and cutting speed based on the local curvature distribution and the transient thermodynamics coupling model, where the transient thermodynamics coupling model solves the heat conduction equation through finite element discretization and dynamically adjusts the heat source distribution according to the curvature; S4. According to the optimized parameter combination, controlling the deformable mirror to adjust the focusing position of the laser beam, and synchronously correcting the end pose of the six-axis robotic arm through Lie group mapping; S5. Performing laser cutting, and collecting the temperature field data of the processing area in real time through an infrared thermal imager. If the deviation between the measured temperature field and the model prediction value exceeds the preset threshold, triggering the reinforcement learning model to re-plan the path.

2. The method for opening holes in a sports kettle based on laser cutting technology according to claim 1, characterized in that In the step S1, after generating the three-dimensional point cloud data by scanning the surface of the kettle through the laser ranging sensor array, calculating the local curvature of each point based on differential geometry formulas, specifically: For each point, fitting a parametric curve in the local neighborhood, and calculating the curvature value through the vector cross product operation of the first derivative and the second derivative of the curve.

3. The method for opening holes in a sports kettle based on laser cutting technology according to claim 1, characterized in that, In the step S2, the reward function of the reinforcement learning model is defined as the weighted difference between the curvature fitting degree evaluation function and the entropy increment of the heat affected zone, where the weight coefficient is used to balance the optimization objectives of path accuracy and heat damage control.

4. The method for opening holes in a sports kettle based on laser cutting technology according to claim 3, wherein, The entropy increment of the heat affected zone is calculated through the transient thermodynamics coupling model, specifically: based on the material density, specific heat capacity at constant pressure, and the entropy change integral of the predicted temperature field relative to the ambient temperature, and the integration domain covers the spatial range of the heat affected zone.

5. The method for opening holes in a sports kettle based on laser cutting technology according to claim 1, wherein, In the step S3, the heat conduction equation of the transient thermodynamics coupling model is discretized into a linear equation system of a heat capacity matrix and a heat conduction matrix through the finite element method, and its heat source term vector is dynamically generated according to the curvature distribution, specifically: Superposing the laser power distribution coefficient at each curvature sampling point with the Gaussian heat source model to generate a spatial energy distribution field related to the curvature, and mapping it to the finite element grid nodes.

6. The method for opening a hole in a sports kettle based on laser cutting technology according to claim 5, wherein, The grid density of the finite element discretization is dynamically adjusted based on the curvature gradient and the cutting speed, Specifically: in the area where the curvature gradient is large or the cutting speed is lower than the minimum threshold, the grid density is increased proportionally, and the adjustment amplitude is determined by the ratio of the curvature gradient vector to the speed parameter.

7. The method for opening a hole in a sports kettle based on laser cutting technology according to claim 1, wherein In the step S4, when controlling the deformable mirror to adjust the focusing position of the laser beam, calculating the mirror deformation phase distribution of the deformable mirror according to the curvature distribution, specifically: Performing orthogonal projection of the curvature data and the Zernike basis function within the effective area of the mirror to generate Zernike coefficients, and driving the deformable mirror to generate a wavefront compensation phase matching the surface curvature.

8. The method for opening holes in a sports kettle based on laser cutting technology according to claim 7, wherein, In the Lie group mapping, the end - pose correction amount of the six - axis robotic arm is generated through the following steps: Calculate the joint - angle correction amount based on the linear combination of Zernike coefficients and a preset weight matrix, and apply the correction amount to the initial pose matrix of the robotic arm through the Lie group exponential mapping to achieve the dynamic coordination of the end - pose and the laser focusing position.

9. The method for opening holes in a sports kettle based on laser cutting technology according to claim 1, characterized in that, In step S5, the deviation determination condition between the measured temperature field and the model prediction value is: When the overall deviation amount between the measured temperature - field vector and the predicted temperature - field vector exceeds a preset threshold, trigger the reinforcement - learning model to re - plan the laser cutting path.

10. A motion kettle hole opening system based on laser cutting technology, applied to the method described in any one of claims 1-9, characterized in that, The system includes: A laser - ranging sensor array module, which is used to scan the surface of the kettle and generate three - dimensional point - cloud data; A reinforcement - learning decision - making module, which is used to dynamically plan the laser cutting path according to the curvature data; A thermodynamics - coupling solver module, which is used to solve the transient heat - conduction equation and optimize the laser power and cutting - speed parameters; A deformable - mirror and six - axis robotic - arm collaborative control module, which is used to adjust the laser focusing position and correct the robotic - arm pose according to the optimized parameters; A multi - sensor feedback module, which is used to collect temperature - field data in real time and trigger path re - planning.

Images