Thermal-sensing autonomous welding path planning method and system and readable storage medium

Abstract

The embodiment of the invention discloses a thermal sensing autonomous welding path planning method and system and a readable storage medium. The method comprises the steps that S1, a four-target welding path optimization model is established; s2, establishing a TD-TSP model and a continuous planning model; s3, solving an optimal welding seam sequence, and inputting the optimal welding seam sequence into the continuous planning model to generate a smooth path; s4, calculating the accurate moving time of the smooth path, and feeding back the accurate moving time to the TD-TSP model to update cost estimation; iteration of the continuous planning model is realized; s5, judging whether the continuous planning model reaches the maximum number of iterations or not; if yes, outputting an optimal welding path; if not, executing S6; s6, judging whether the smooth path meets a constraint condition or not and whether an objective function of the continuous planning model is converged or not; if not, executing S3; and if yes, outputting an optimal welding path. By introducing the heat-induced deformation cost and the heat accumulation constraint, heat deformation and residual stress in the welding process are effectively controlled, weld defects are reduced, and the product percent of pass is increased.

Description

Technical Field

[0001] This invention relates to the field of welding automation technology, and more specifically, to a heat-sensing autonomous welding path planning method. Background Technology

[0002] In the field of welding automation, the core of robot path planning for multi-weld workpieces is to complete the welding operation of all welds by optimizing the welding sequence and robot motion trajectory while meeting process and physical constraints. Traditional path planning methods mainly focus on optimization at the geometric and kinematic levels, and usually construct a multi-objective optimization model that includes path length, welding time, and path smoothness.

[0003] However, existing technologies suffer from the following significant drawbacks: Lack of thermo-temporal dependence: Traditional models assume all welds are independent and unrelated, failing to consider the thermo-coupling effects during welding. In actual welding, the heat input from preceding welds forms a residual temperature field within the workpiece through heat conduction. If the weld sequence is poorly designed, the accumulation of residual heat can lead to an increase in the initial temperature of subsequent welds, causing workpiece deformation, increased residual stress, uneven weld penetration, and forming defects. In severe cases, this can result in structural distortion, cracks, or deterioration of metallurgical properties. NP-hard characteristics conflict with real-time requirements: Multi-weld path planning is an NP-hard problem coupling combinatorial and continuous optimization. Decision variables include the weld sequence and the three-dimensional continuous path. Existing heuristic algorithms have high computational complexity, with overall time consumption far exceeding the real-time planning requirements of industrial scenarios, making it difficult to quickly update planning results. Weak generalization ability in multi-objective trade-offs: Traditional models balance multiple objectives with fixed weights, but these weights rely on manual experience and cannot adapt to changes in process parameters or material changes. Differences in thermal sensitivity between different production scenarios and different base materials require readjustment of weights, increasing the difficulty of engineering applications.

[0004] Therefore, there is an urgent need for a new multi-weld path planning technology that can simultaneously consider thermophysical effects, has real-time solution capabilities, and can adapt to various working conditions. Summary of the Invention

[0005] In view of the above problems, the purpose of this invention is to provide a heat-sensing autonomous welding path planning method, system, and readable storage medium.

[0006] The first aspect of this invention provides a heat-sensing autonomous welding path planning method, the method comprising:

[0007] S1: Based on the total length of the weld connection path, the total welding time, and the path smoothness, a four-objective welding path optimization model is established by introducing the cost of thermally induced deformation.

[0008] S2: Based on the four-objective welding path optimization model, establish the TD-TSP model and the continuous programming model;

[0009] S3: Use the TD-TSP model to solve for the optimal weld sequence, and input the optimal weld sequence into the continuous programming model to generate a smooth path;

[0010] S4: Calculate the precise travel time of the smooth path and feed it back to the TD-TSP model to update the cost estimate; enable iteration of the continuous programming model;

[0011] S5: Determine if the continuous programming model has reached the maximum number of iterations; if yes, output the smooth path corresponding to the last iteration as the optimal welding path; if no, proceed to S6.

[0012] S6: Determine whether the smooth path satisfies the constraints and whether the objective function of the continuous programming model converges;

[0013] If not, then execute S3;

[0014] If so, the corresponding smooth path will be output as the optimal welding path.

[0015] Preferably, the four-target welding path optimization model is as follows:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025] in, These are the weighting coefficients; This is the cost of thermally induced deformation; Indicates the first The weld number of each weld. , This represents the number of welds. Indicates the robot from the weld seam The endpoint moves to the weld. The path from the starting point; Indicates the length of the path; For workpiece temperature; Representing a path curvature on; For collision detection functions, Point Outside the obstacle or at the boundary, Point Intrusion into obstacles;

[0026] A collection of welds with a fixed direction; For path Maximum velocity modulus; For path Maximum acceleration modulus; For path Maximum curvature on;

[0027] For weld exist The initial temperature at that time, The preset heat accumulation threshold; Indicates weld seam The actual length, Indicates weld seam Process speed; Indicates weld seam The speed fluctuation;

[0028] For the cumulative time of the sequence:

[0029]

[0030] For path The idle shift time;

[0031] Cost of thermally induced deformation:

[0032]

[0033] It is a deformation function based on temperature; For weld At any moment Temperature distribution, Described by the Fourier partial differential equation for heat diffusion, which is directly related to the heat input of the weld. The mathematical expression is:

[0034]

[0035] in: For the workpiece's spatial coordinates, It is a three-dimensional physical space; For weld In coordinates ,time Heat input distribution; The thermal diffusivity of the base material; For the workpiece in space ,time Temperature value; for The Laplace operator.

[0036] Preferably, S3 specifically includes:

[0037] A PINN model is constructed to predict the temperature field distribution and thermally induced deformation during the welding process in real time, and to estimate the thermal cost of the weld sequence.

[0038] Based on thermal cost estimation, the near-end strategy optimization (PPO) algorithm is used to solve the TD-TSP model and output an approximately optimal weld sequence.

[0039] The optimal weld sequence is input into a continuous programming model, and a smooth path that satisfies obstacle avoidance and kinematic constraints is generated based on the CHOMP algorithm.

[0040] Preferably, the network parameters θ of the PINN model are trained by minimizing the following composite loss function:

[0041]

[0042] in: For data fitting loss based on sensor observations; This represents the residual loss in the thermal diffusion equation; These are weighting coefficients; The number of samples in the observation data. For sample index, ; Predicting spatial coordinates for the temperature field At any moment Temperature value; For the first The actual temperature values ​​at each observation point; For sampling point index, , This represents the number of sampling points used to verify the physical equations. The thermal diffusivity of the base material. To predict temperature The Laplace operator, In spatial coordinates ,time The welding heat input density.

[0043] Preferably, the formula for calculating the thermal cost estimation of the weld sequence is:

[0044]

[0045] Indicates weld sequence The Middle When welding a strip weld, the spatial coordinates of the workpiece are: Position at time The predicted temperature value.

[0046] Preferably, the TD-TSP model is:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] in, From the weld End point to weld Estimated travel time from the starting point; For weld Welding time; Represents the time-dependent heat cost; node set Where node 0 represents the initial position, node Indicates weld seam Directed edge set , indicating from the weld To the weld The transfer; Indicates weld seam The initial temperature.

[0056] Preferably, the reward function of the PPO algorithm is designed as follows:

[0057]

[0058] in: For the cost of travel time; This represents the current welding time of the weld seam. This represents an increase in heat cost; For value function estimation, This represents the discount factor, used to measure the weight of future rewards on current decisions;

[0059] The PPO algorithm uses a policy network. Generate an action probability distribution and perform stable training by pruning the proxy objective function:

[0060]

[0061] in For probability ratios, For the dominant function, It is the clipping coefficient for controlling the update magnitude of the policy in the PPO algorithm.

[0062] Preferably, the continuous programming model is:

[0063]

[0064]

[0065]

[0066]

[0067] in, Indicates the first Parametric curves of the path, for The first derivative of represents the second derivative. Indicates the first Path in parameters curvature at that point Indicates the maximum permissible curvature of the path.

[0068] Preferably, the step of inputting the optimal weld sequence into a continuous programming model and generating a smooth path that satisfies obstacle avoidance and kinematic constraints based on the CHOMP algorithm specifically involves:

[0069] First, the optimal weld sequence is parameterized as follows: And construct the objective function for the CHOMP algorithm optimization:

[0070]

[0071] Where: the first term is the smoothing term, which minimizes the trajectory kinetic energy; the second term... The potential field function includes: obstacle potential field: Encourage staying away from obstacles; thermal field: Avoid high-temperature areas.

[0072] Then, the weld path is updated using the covariant gradient descent method. The update formula for the covariant gradient descent method is as follows:

[0073]

[0074] in, For the new weld path, This is the quality matrix.

[0075] A second aspect of the present invention provides a heat-sensing autonomous welding path planning system, including a memory and a processor. The memory includes a heat-sensing autonomous welding path planning method program, which, when executed by the processor, implements the steps of a heat-sensing autonomous welding path planning method.

[0076] A third aspect of the present invention provides a computer-readable storage medium including a heat-sensing autonomous welding path planning method program, wherein when the heat-sensing autonomous welding path planning method program is executed by a processor, it implements the steps of a heat-sensing autonomous welding path planning method.

[0077] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0078] 1. Significantly improved welding quality: By introducing thermally induced deformation costs and thermal accumulation constraints, the thermal deformation and residual stress during the welding process are effectively controlled, weld defects are reduced, and the product qualification rate is improved.

[0079] 2. Significantly improved computational efficiency: By adopting a hierarchical solution architecture and the PINN proxy model, the calculation time for thermal costs is reduced from seconds to milliseconds, meeting the needs of real-time industrial planning.

[0080] 3. Strong adaptability: Through an adaptive weighting mechanism, the target weights can be automatically adjusted according to the characteristics of the parent material and production needs, reducing reliance on human experience and enhancing the system's generalization ability.

[0081] 4. Motion stability optimization: By using the path smoothness cost function and kinematic constraints, we ensure smooth and stable robot motion, reduce vibration, and extend equipment life. Attached Figure Description

[0082] Figure 1 This is a flowchart of a heat-sensing autonomous welding path planning method as described in Example 1. Detailed Implementation

[0083] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0084] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0085] Example 1

[0086] like Figure 1 As shown, this embodiment discloses a heat-sensing autonomous welding path planning method, the method comprising:

[0087] S1: Based on the total length of the weld connection path, the total welding time, and the path smoothness, a four-objective welding path optimization model is established by introducing the cost of thermally induced deformation.

[0088] In this embodiment, the four-target welding path optimization model is as follows:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] in, These are the weighting coefficients; This is the cost of thermally induced deformation; Indicates the first The weld number of each weld. , This represents the number of welds. Indicates the robot from the weld seam The endpoint moves to the weld. The path from the starting point; Indicates the length of the path; For workpiece temperature; Representing a path curvature on; For collision detection functions, Point Outside the obstacle or at the boundary, Point Intrusion into obstacles;

[0099] A collection of welds with a fixed direction; For path Maximum velocity modulus; For path Maximum acceleration modulus; For path Maximum curvature on;

[0100] For weld exist The initial temperature at that time, The preset heat accumulation threshold; Indicates weld seam The actual length, Indicates weld seam Process speed; Indicates weld seam The speed fluctuation;

[0101] For the cumulative time of the sequence:

[0102]

[0103] For path The idle shift time;

[0104] Cost of thermally induced deformation:

[0105]

[0106] It is a deformation function based on temperature; For weld At any moment Temperature distribution, Described by the Fourier partial differential equation for heat diffusion, which is directly related to the heat input of the weld. The mathematical expression is:

[0107]

[0108] in: For the workpiece's spatial coordinates, It is a three-dimensional physical space; For weld In coordinates ,time Heat input distribution; The thermal diffusivity of the base material; For the workpiece in space ,time Temperature value; for The Laplace operator.

[0109] It should be noted that in the field of welding automation, path planning for multi-weld workpieces aims to complete the welding operation of all welds through the orderly movement of robots or automated equipment. The core is to optimize key production indicators while meeting process and physical constraints. To achieve this goal, it is necessary to first clarify the basic elements and decision variables of the problem, then construct an objective function based on actual needs, and establish constraints in conjunction with scenario boundary conditions, ultimately forming a complete mathematical optimization model.

[0110] Let the workpiece to be welded include A single independent weld, denoted as the weld set. For any weld The starting point of its welding path is The endpoint is If the weld direction is determined by the process, such as in a full penetration weld where the penetration depth needs to be ensured along a fixed direction, then... This is the only irreversible welding direction; otherwise... The direction is selectable; the actual weld length is That is, the Euclidean distance between the start and end points; furthermore, for any weld... The welding process must follow a fixed process speed. The welding heat input is determined by factors such as the type of welding wire and the material of the base metal. There are fixed obstacles around the workpiece, and its geometric boundary is defined by a set of... Description: Robotic motion has physical limits, including maximum speed. Maximum acceleration and the upper limit of path curvature .

[0111] The decision-making process for multi-weld path planning requires simultaneously determining the welding sequence and the robot's motion trajectory; therefore, two types of core decision variables are defined.

[0112] 1) Weld sequence variables

[0113]

[0114] in It is a set An arrangement, Indicates the first The weld number of each weld. That is, the robot presses The welding was completed in the following order.

[0115] 2) Weld connection path variables

[0116]

[0117] in Indicates that the robot starts from the first End of the weld Move to the Starting point of a strip weld The trajectory of motion is defined. This indicates returning to the standby starting point. A three-dimensional continuous curve can be represented in parametric form. ,in For path parameters, For path endpoint constraints, These are the first and second derivatives of the path, respectively.

[0118] The core objective of path planning is to balance production efficiency and motion stability, which requires simultaneously minimizing the total path length, total welding time, and path smoothness. Therefore, a multi-objective minimization function is constructed as follows:

[0119]

[0120] in, The total length of the connection path between welds reflects the time consumed by the robot's idle movement; the calculation formula is as follows:

[0121]

[0122] In the formula For path The velocity vector magnitude is the integral result of which is the arc length of a single connecting path. The summation is the total length of all connecting paths. The total welding time for all welds is obtained by summing the welding times of individual welds, and the calculation formula is as follows:

[0123]

[0124] In the formula For the first Welding time for each weld seam These represent the length of the weld and the welding speed, respectively, which conform to the physical relationship of time = length / speed. The overall path smoothness index is quantified by the path curvature integral, and the calculation formula is as follows:

[0125]

[0126] In the formula For path In parameters curvature at that point Represents the cross product of vectors. Arc length element The integral result is the cumulative curvature of a single path, and the summation is the total smoothness cost of all connected paths. Let the target weight coefficient satisfy... It is used to balance the priorities of the three objectives based on actual production needs.

[0127] In addition, the model needs to satisfy three types of boundary conditions: geometric, technological, and equipment kinematics, which can be converted into the following mathematical constraints:

[0128] 1) Obstacle avoidance constraints: the robot's connection path No to obstacles A collision occurs when all points on the path are outside the obstacle. This is determined by the collision detection function. Quantification: If , indicating a point Outside the obstacle or at the boundary; if , indicating a point Intrusion into the obstacle. Therefore, the constraint is:

[0129]

[0130] 2) Weld direction constraint: For welds with a fixed direction, the set of such welds is denoted as... Its welding direction must be Irreversible, i.e., sequence Intermediate weld The start and end points of the welding must be strictly matched. and The constraints are:

[0131]

[0132] 3) Robot kinematic constraints: Robot motion must conform to physical limits to avoid damage or vibration caused by excessive speed, acceleration, or curvature. Constraints include: Velocity constraint: The velocity modulus at any point on the connection path must not exceed... :

[0133]

[0134] Acceleration constraint: The acceleration magnitude at any point on the connecting path does not exceed :

[0135]

[0136] Curvature constraint: The curvature of any point on the connecting path does not exceed :

[0137]

[0138] Combining the above objective function and constraints, the mathematical optimization problem of multi-weld welding path planning can be uniformly described as:

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145]

[0146]

[0147]

[0148]

[0149] This model fully integrates the objectives, boundary conditions, and constraints of multi-weld seam planning, providing a clear mathematical framework for the design of subsequent solution algorithms.

[0150] However, although current multi-weld path planning models have established an optimization framework at the geometric and kinematic levels, their design assumptions and technical characteristics have significant flaws in actual engineering scenarios, making it difficult to coordinate welding quality and production efficiency. Specifically, this is reflected in the following three aspects:

[0151] 1) Lack of thermo-temporal dependence leads to significant welding quality risks.

[0152] The model assumes all welds are independent and unrelated, neglecting the thermo-coupling effects during welding. In actual welding, the heat input of each weld... A residual temperature field will be formed inside the workpiece through heat conduction, and this temperature field will change over time according to the Fourier thermal diffusion equation. The thermal diffusivity of the base material. The temperature of the workpiece diffuses outwards. If the weld sequence... Improper design can lead to residual heat from preceding welds accumulating inside the workpiece, causing an increase in the initial temperature of subsequent welds and directly resulting in workpiece deformation. Increased residual stress can lead not only to defects such as uneven weld penetration and poor weld formation, but also, in severe cases, to workpiece structural distortion, cracking, or decreased metallurgical properties. Such designs that ignore thermo-temporal dependencies may require additional post-planning costs and could even result in product scrapping.

[0153] 2) The contradiction between NP-hard characteristics and real-time requirements

[0154] The model's decision variables include the weld sequence. With connection path Two categories: the former is a collection of weld seams. The arrangement of the two curves, the latter being a three-dimensional continuous curve. This problem is essentially an NP-hard problem coupling combinatorial optimization and continuous optimization: existing heuristic algorithms, when solving it, already have a computational complexity of up to [missing information - likely a number] for the sequence optimization stage alone. After optimizing the non-convex path generation, the overall time consumption far exceeds the real-time planning requirements of industrial scenarios. The computational efficiency of traditional algorithms is insufficient to meet the needs of real-time planning, and their adaptability is limited.

[0155] 3) The generalization ability of multi-objective trade-offs is weak.

[0156] The model uses fixed weights The model aims to balance three key objectives: path length, welding time, and path smoothness. However, the weights of these objectives rely on manual experience for setting. For example, mass production scenarios prioritize shortening welding time, while high-precision scenarios prioritize path smoothness. Furthermore, the varying thermal sensitivity of different base materials necessitates readjusting the weights. This reliance on manual intervention prevents the model from adapting to changes in process parameters or material replacements, resulting in weak generalization ability and increased difficulty in engineering applications.

[0157] To address the aforementioned shortcomings, the existing static geometric planning model needs to be transformed into a variant of the Time-Dependent Traveling Salesman Problem (TD-TSP). The core logic is to quantify the thermal time-series dependency into a computable cost term, achieving deep coupling between geometric path planning and physical quality control. Simultaneously, the objective function structure and constraint system are optimized to improve the model's adaptability to different scenarios. The key to this transformation lies in overcoming the limitations of static optimization that only considers spatial geometry and fixed process parameters, introducing dynamic characteristics of the time dimension—including not only weld welding time and robot idle time, but also the diffusion effect of preceding weld heat input over time. This ensures that the planning results simultaneously meet geometric feasibility, motion stability, and physical quality requirements.

[0158] The transformed model aims at the coordinated optimization of quality, efficiency, and stability. It adds the cost of thermally induced deformation to the existing three objectives, forming a four-objective weighted minimization model, the mathematical expression of which is as follows:

[0159]

[0160] It has a decision variable: weld sequence Weld assembly The arrangement, Indicates the first The weld number of each weld, and Corresponding robot standby position (Return to initial position after welding); Connection path : Set of three-dimensional continuous curves ,in Indicates the robot from the weld seam The End Move to weld The starting point The trajectory.

[0161] Total path length in the objective function Total welding time and path smoothness cost The definition and calculation method are the same as the original model, and the weight coefficients are: And satisfy An adaptive adjustment mechanism is adopted: automatically adjusting based on the thermal sensitivity of the base material. Adjust according to production rhythm This addresses the generalization problem of traditional artificial weights. The focus here is on the newly added thermally induced deformation cost. and its calculation method

[0162] thermally induced deformation cost The deformation risk caused by the quantification of thermodynamic time-series dependence is calculated using the following formula:

[0163]

[0164] in: For the cumulative time of the sequence:

[0165]

[0166] For path The idle time is determined by the path length and calculate; For weld At any moment Temperature distribution, affected by the preceding weld seam The influence of heat input; It is a deformation function based on temperature; the higher the temperature, the greater the deformation.

[0167] Fusion of thermal cost calculation and PINN proxy model: thermally induced deformation cost The core is to accurately calculate the temperature distribution. It is necessary to establish a computational model based on the physical process of thermal diffusion, while also taking into account the real-time requirements.

[0168] Fundamentals of the thermal diffusion equation: temperature distribution Described by the Fourier Partial Differential Equation (PDE), which directly relates to the heat input of the weld. The mathematical expression is:

[0169]

[0170] in: The coordinates of the workpiece in space; Weld In coordinates ,time The heat input distribution is determined by the heat input distribution. With weld length Derivation using Gaussian distribution For a moment The welding position, The radius of thermal influence; The thermal diffusivity of the base material determines the rate of heat transfer within the workpiece.

[0171] Directly solving the thermal diffusion PDE requires high computational cost and cannot meet the needs of real-time planning. Therefore, a Physics-Informed Neural Network (PINN) is introduced as a surrogate model to accelerate the processing. To address the complex temperature field calculations and thermally induced deformation, this embodiment uses PINN for rapid prediction, and then calculates the cost based on the prediction results.

[0172] PINN's core advantage lies in embedding PDE constraints into network training to achieve rapid prediction of temperature field and deformation. PINN's inputs include the spatial location of the preceding weld seam. Heat input Its welding completion time, and the current spatial coordinates to be predicted. and cumulative time Output includes: weld seam At any moment Temperature distribution With corresponding deformation Training loss includes data loss and PDE loss, ensuring that the prediction results conform to physical laws. Through the PINN surrogate model, thermal costs are minimized. The computation time has been reduced from seconds to milliseconds, effectively resolving the contradiction between NP-hard problems and real-time performance.

[0173] Based on existing geometric, technological, and kinematic constraints, thermodynamic temporal constraints and uncertainty constraints are added to form a complete constraint system, ensuring the physical feasibility of the planning results:

[0174] 1) Obstacle avoidance constraint: All points on the connecting path must be located outside the obstacles to avoid collisions. The mathematical expression is:

[0175]

[0176] in For collision detection functions, Point Outside the obstacle or at the boundary, Point Intrusion into obstacles.

[0177] 2) Weld direction constraint: Welds with a fixed direction must strictly follow the welding direction specified in the process. The mathematical expression is:

[0178]

[0179] 3) Robot kinematic constraints

[0180] A. Speed ​​Constraint: The velocity modulus of any point on the connection path must not exceed the robot's maximum speed to prevent damage from excessive speed.

[0181]

[0182] B. Acceleration Constraint: The acceleration modulus at any point on the connection path shall not exceed the robot's maximum acceleration to avoid vibration affecting welding accuracy.

[0183]

[0184] 4) Curvature constraint: The curvature of any point on the connecting path does not exceed the upper limit of curvature to ensure smooth motion.

[0185]

[0186] 5) Thermodynamic time series and uncertainty constraints

[0187] A. Thermal accumulation constraint: To prevent residual heat from preceding welds from causing excessively high initial temperatures in subsequent welds, thus controlling the risk of deformation.

[0188]

[0189] in For weld When welding begins (time) The initial temperature of ) The preset heat accumulation threshold.

[0190] B. Uncertainty Constraints: Addressing disturbances in process parameters to ensure the robustness of planning results.

[0191]

[0192] This constraint allows for a maximum welding time deviation of 10%, adapting to parameter fluctuations in real-world scenarios.

[0193] In summary, the multi-weld path planning problem, which transforms the existing static geometric programming model into a variant of the Time-Dependent Traveling Salesman Problem (TD-TSP), falls under Mixed-Integer Nonlinear Programming (MINLP). Its structure directly corresponds to the decision variables: Outer layer: Combinatorial optimization problem; Corresponding weld sequence. The choice needs to be made from Selecting the optimal sequence that satisfies the heat accumulation constraint from a given permutation is an NP-hard problem; Inner layer: continuous non-convex optimization problem: corresponding connection path The generation of the curve requires the combined effect of obstacle avoidance constraints, kinematic constraints, and thermal potential fields to produce a smooth and feasible three-dimensional curve.

[0194] This classification clarifies the design direction of subsequent solution algorithms: a hierarchical solution strategy should be adopted, while the computational complexity of thermal costs should be reduced through the PINN proxy model, ultimately achieving synergistic optimization of welding quality, production efficiency (and motion stability).

[0195] S2: Based on the four-objective welding path optimization model, establish the TD-TSP model and the continuous programming model.

[0196] It should be noted that after constructing the four-objective welding path optimization model in step S1, this model, as a highly coupled mixed-integer nonlinear programming problem, faces the challenge of excessively high computational complexity when solved directly. To achieve efficient solution, this application decomposes the original problem into two relatively independent and sequentially solvable subproblems: weld sequence planning based on TD-TSP, and continuous path planning based on constraint optimization. The original MINLP problem is decoupled into two subproblems: weld sequence optimization and path trajectory optimization.

[0197] (1) Weld sequence optimization:

[0198] Arrange the weld seams in sequence As a decision variable, it is a TD-TSP variant that minimizes the total welding time and the cost of thermally induced deformation, while being subject to time dependence of weld direction constraints and thermal timing constraints.

[0199] (2) Path trajectory optimization:

[0200] Has connection path The decision variables are a constrained continuous path optimization problem with the objective of minimizing path length and curvature integral, subject to obstacle avoidance and kinematic constraints.

[0201] The two sub-problems are coupled through interface variables: weld sequence optimization provides weld sequence σ for path trajectory optimization, and path trajectory optimization provides accurate movement time cost for weld sequence optimization.

[0202] In this embodiment, the TD-TSP model is:

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210]

[0211] in, From the weld End point to weld Estimated travel time from the starting point; For weld Welding time; Represents the time-dependent heat cost; node set Where node 0 represents the initial position, node Indicates weld seam Directed edge set , indicating from the weld To the weld The transfer; Indicates weld seam The initial temperature.

[0212] It should be noted that this embodiment models the weld sequence planning as a time-dependent traveling salesman problem: node set Where node 0 represents the initial position, node Indicates weld seam Directed edge set , indicating from the weld To the weld The transfer.

[0213] side The cost consists of three parts:

[0214]

[0215] in: From the weld End point to weld Estimated travel time from the starting point; For weld Welding time; This represents the time-dependent thermal cost, based on the arrival time at the weld. The moment Calculation. Decision variables are defined as follows:

[0216]

[0217] Therefore, the TD-TSP model can be expressed as follows:

[0218]

[0219]

[0220]

[0221]

[0222]

[0223]

[0224] In this embodiment, the continuous programming model is:

[0225]

[0226]

[0227]

[0228]

[0229] in, Indicates the first Parametric curves of the path, for The first derivative of represents the second derivative. Indicates the first Path in parameters curvature at that point Indicates the maximum permissible curvature of the path.

[0230] S3: Use the TD-TSP model to solve for the optimal weld sequence, and input the optimal weld sequence into the continuous programming model to generate a smooth path.

[0231] In this embodiment, S3 includes:

[0232] S3.1: Construct the PINN model, use the PINN model to predict the temperature field distribution and thermally induced deformation in real time during the welding process, and calculate the thermal cost estimate of the weld sequence;

[0233] S3.2: Based on thermal cost estimation, the Proximal Policy Optimization (PPO) algorithm is used to solve the TD-TSP model and output an approximately optimal weld sequence;

[0234] S3.3: Input the optimal weld sequence into the continuous programming model and generate a smooth path that satisfies obstacle avoidance and kinematic constraints based on the Covariant Hamiltonian Optimization for Motion Planning (CHOMP) algorithm.

[0235] In this embodiment, the calculation formula for estimating the thermal cost of the weld sequence is as follows:

[0236]

[0237] Indicates weld sequence The Middle When welding a strip weld, the spatial coordinates of the workpiece are: Position at time The predicted temperature value.

[0238] In this embodiment, the reward function of the PPO algorithm is designed as follows:

[0239]

[0240] in: For the cost of travel time; This represents the current welding time of the weld seam. This represents an increase in heat cost; For value function estimation, This represents the discount factor, used to measure the weight of future rewards on current decisions;

[0241] The PPO algorithm utilizes a Deep Reinforcement Learning (DRL) policy network. Generate an action probability distribution and perform stable training by pruning the proxy objective function:

[0242]

[0243] in For probability ratios, For the dominant function, It is the clipping coefficient for controlling the update magnitude of the policy in the PPO algorithm.

[0244] In this embodiment, the step of inputting the optimal weld sequence into the continuous programming model and generating a smooth path that satisfies obstacle avoidance and kinematic constraints based on the CHOMP algorithm is as follows:

[0245] First, the optimal weld sequence is parameterized as follows: And construct the objective function for the CHOMP algorithm optimization:

[0246]

[0247] Where: the first term is the smoothing term, which minimizes the trajectory kinetic energy; the second term... The potential field function includes: obstacle potential field: Encourage staying away from obstacles; thermal field: Avoid high-temperature areas.

[0248] Then, the weld path is updated using the covariant gradient descent method. The update formula for the covariant gradient descent method is as follows:

[0249]

[0250] in, For the new weld path, This is the quality matrix.

[0251] S4: Calculate the precise travel time of the smooth path and feed it back to the TD-TSP model to update the cost estimate; enable iteration of the continuous programming model;

[0252] S5: Determine if the continuous programming model has reached the maximum number of iterations; if yes, output the smooth path corresponding to the last iteration as the optimal welding path; if no, proceed to S6.

[0253] S6: Determine whether the smooth path satisfies the constraints and whether the objective function of the continuous programming model converges;

[0254] If not, then execute S3;

[0255] If so, the corresponding smooth path will be output as the optimal welding path.

[0256] It should be noted that the precise travel time of the smooth path described in S4 is the result obtained after optimization by the CHOMP algorithm, and is part of the cost, namely the time cost. This is an important basis for decision-making.

[0257] It should be noted that, to support real-time solving of TD-TSP, this embodiment uses PINN as a surrogate model for the thermal diffusion equation. The inputs to the PINN model are weld sequence characteristics, time information, and process parameters. The outputs of the PINN model are temperature field prediction and thermal deformation estimation. The loss function includes the partial differential equation (PDE) residuals plus data fitting terms.

[0258] In this embodiment, during the TD-TSP solution process, PINN is used to quickly evaluate the thermal cost of different sequences, thereby achieving real-time integration of physical constraints.

[0259] It should be noted that, as a specific embodiment, this embodiment provides a Thermal-Aware Autonomous Welding Path Planning Framework (TAWPP). This framework decomposes the original mixed-integer nonlinear programming problem into two relatively independent sub-problems: sequence optimization and path planning, and employs an alternating optimization strategy for efficient solution. The overall framework integrates physical modeling, reinforcement learning, and trajectory optimization techniques to form a closed-loop decision system, enabling the coordinated optimization of welding quality and efficiency while satisfying thermal constraints.

[0260] The TAWPP framework consists of three core modules: a thermal sensing module, a sequence optimization module, and a path planning module.

[0261] 1) Thermal sensing module: PINN is used to predict the temperature field distribution and thermally induced deformation during the welding process in real time, providing thermal cost estimation for sequence optimization.

[0262] 2) Sequence optimization module: The weld sorting problem is modeled as a TD-TSP problem and solved using the Proximal Policy Optimization (PPO) algorithm to output an approximately optimal weld sequence.

[0263] 3) Path planning module: Receives the sequence optimization results and generates a smooth path that satisfies obstacle avoidance and kinematic constraints based on the CHOMP algorithm.

[0264] The three modules form a closed-loop optimization by executing alternately: the sequence optimization provides the welding order for path planning, the path planning returns the precise movement time to update the cost estimate of the sequence optimization, and the thermal sensing module provides dynamic physical constraints for both.

[0265] The core of the thermal sensing module is to construct a surrogate model of the thermal diffusion equation, enabling rapid prediction of temperature field and deformation. The PINN structure has the advantage of embedding physical laws, reducing reliance on large amounts of labeled data.

[0266] PINN's input is spatial coordinates ,time Heat input history of preceding welds The output is a prediction of the current temperature field. and deformation field Network parameters Training is performed by minimizing the following composite loss function:

[0267]

[0268] in: For data fitting loss based on sensor observations; This represents the residual loss in the thermal diffusion equation; These are weighting coefficients used to balance the two losses. After training, this module can compute any sequence in real time. In time Estimation of heat cost:

[0269]

[0270] The sequence optimization module models the weld sorting problem as a Markov decision process and uses the PPO algorithm to solve it.

[0271] State space: includes the current temperature field, the set of welded seams, the robot's current position, and the geometric information of the remaining seams.

[0272] Action Space: Select the next weld number to be welded from the remaining welds.

[0273] The reward function is designed as follows:

[0274]

[0275] in: The cost of travel time (feedback from the path planning module); This represents the current welding time of the weld seam. This represents an increase in heat cost; This is a value function estimate used to encourage long-term optimization.

[0276] The PPO algorithm utilizes a Deep Reinforcement Learning (DRL) policy network. Generate action probability distributions and perform stable training using a clipped surrogate objective:

[0277]

[0278] in For probability ratios, This is the dominant function.

[0279] The weld sequence is determined using the sequence optimization module. Then, the path planning module uses the CHOMP algorithm to optimize the connection trajectory between welds. The path is parameterized as follows: And construct the following optimization objective:

[0280]

[0281] Where: the first term is the smoothing term, which minimizes the trajectory kinetic energy; the second term... The potential field function includes: obstacle potential field: Encourage staying away from obstacles; thermal field: To avoid high-temperature areas, CHOMP iteratively updates the path using covariant gradient descent.

[0282]

[0283] in This is the mass matrix, used to introduce dynamic constraints. Iterate until the path satisfies the constraints and the objective function converges.

[0284] In summary, the initialization of the TAWPP framework involves loading environment parameters, warming up the PINN model, and initializing the Deep Reinforcement Learning (DRL) policy network. Sequence generation includes the DRL agent generating candidate weld sequences based on the current thermal field and geometric information. Path refinement means that for each candidate sequence, the CHOMP algorithm is called to calculate the exact path. The system calculates the exact move time and hot cost, feeding these into the DRL reward function. Policy updates are performed on the DRL policy network based on this feedback, generating an improved sequence. If the sequence performance no longer improves or reaches the maximum number of iterations, the final solution is output.

[0285] It should be noted that, for sequence optimization, this embodiment uses a DRL policy network for sequence optimization, but it is not limited to using a DRL policy network for sequence optimization. For example, a genetic algorithm (GA) can also be used instead of DRL, with the individual encoded as the weld sequence and the fitness function being the negative of the total cost.

[0286] For path planning, this embodiment uses the CHOMP algorithm to optimize the connection trajectory between welds, but it is not limited to using the CHOMP algorithm to optimize the connection trajectory between welds. The RRT* algorithm can be used instead of CHOMP, which is especially suitable for high-dimensional complex obstacle environments.

[0287] For thermal field prediction, when training data is lacking, interpolation estimation can be performed using a pre-calculated thermal field database based on the finite element method.

[0288] Example 2

[0289] This embodiment discloses a heat-sensing autonomous welding path planning system, including a memory and a processor. The memory includes a heat-sensing autonomous welding path planning method program. When the heat-sensing autonomous welding path planning method program is executed by the processor, it implements the steps of a heat-sensing autonomous welding path planning method as described in Embodiment 1.

[0290] Example 3

[0291] This embodiment discloses a computer-readable storage medium, which includes a heat-sensing autonomous welding path planning method program. When the heat-sensing autonomous welding path planning method program is executed by a processor, it implements the steps of a heat-sensing autonomous welding path planning method as described in Embodiment 1.

[0292] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0293] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0294] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0295] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0296] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A heat-sensing autonomous welding path planning method, characterized in that, The method includes: S1: Based on the total length of the weld connection path, the total welding time, and the path smoothness, a four-objective welding path optimization model is established by introducing the cost of thermally induced deformation. S2: Based on the four-objective welding path optimization model, establish the TD-TSP model and the continuous programming model; S3: Use the TD-TSP model to solve for the optimal weld sequence, and input the optimal weld sequence into the continuous programming model to generate a smooth path; S4: Calculate the precise travel time of the smooth path and feed it back to the TD-TSP model to update the cost estimate; enable iteration of the continuous programming model; S5: Determine if the continuous programming model has reached the maximum number of iterations; if yes, output the smooth path corresponding to the last iteration as the optimal welding path; if no, proceed to S6. S6: Determine whether the smooth path satisfies the constraints and whether the objective function of the continuous programming model converges; If not, then execute S3; If so, the corresponding smooth path will be output as the optimal welding path.

2. The heat-sensing autonomous welding path planning method according to claim 1, characterized in that, The four-objective welding path optimization model is as follows: ； ； ； ； ； ； ； ； in, These are the weighting coefficients; This is the cost of thermally induced deformation; Indicates the first The weld number of each weld. , This represents the number of welds. Indicates the robot from the weld seam The endpoint moves to the weld. The path from the starting point; Indicates the length of the path; For workpiece temperature; Representing a path curvature on; For collision detection functions, Point Outside the obstacle or at the boundary, Point Intrusion into obstacles; A collection of welds with a fixed direction; For path Maximum velocity modulus; For path Maximum acceleration modulus; For path Maximum curvature on; For weld exist The initial temperature at that time, The preset heat accumulation threshold; Indicates weld seam The actual length, Indicates weld seam Process speed; Indicates weld seam The speed fluctuation; For the cumulative time of the sequence: ; For path The idle shift time; Cost of thermally induced deformation: ; It is a deformation function based on temperature; For weld At any moment Temperature distribution, Described by the Fourier partial differential equation for heat diffusion, which is directly related to the heat input of the weld. The mathematical expression is: ; in: For the workpiece's spatial coordinates, It is a three-dimensional physical space; For weld In coordinates ,time Heat input distribution; The thermal diffusivity of the base material; For the workpiece in space ,time Temperature value; for The Laplace operator.

3. The heat-sensing autonomous welding path planning method according to claim 2, characterized in that, S3 specifically includes: A PINN model is constructed to predict the temperature field distribution and thermally induced deformation during the welding process in real time, and to estimate the thermal cost of the weld sequence. Based on thermal cost estimation, the TD-TSP model is solved using a near-end strategy optimization algorithm, and the approximately optimal weld sequence is output. The optimal weld sequence is input into a continuous programming model, and a smooth path that satisfies obstacle avoidance and kinematic constraints is generated based on the CHOMP algorithm.

4. The heat-sensing autonomous welding path planning method according to claim 3, characterized in that, The network parameters θ of the PINN model are trained by minimizing the following composite loss function: ； in: For data fitting loss based on sensor observations; This represents the residual loss in the thermal diffusion equation; These are weighting coefficients; The number of samples in the observation data. For sample index, ; Predicting spatial coordinates for the temperature field At any moment Temperature value; For the first The actual temperature values ​​at each observation point; For sampling point index, , This represents the number of sampling points used to verify the physical equations. The thermal diffusivity of the base material. For the Laplace operator to predict temperature T^, In spatial coordinates ,time The welding heat input density; The formula for estimating the thermal cost of the weld sequence is as follows: ; Indicates weld sequence The Middle When welding a strip weld, the spatial coordinates of the workpiece are: Position at time The predicted temperature value.

5. The heat-sensing autonomous welding path planning method according to claim 4, characterized in that, The TD-TSP model is as follows: ； ； ； ； ； ； ； ； in, From the weld End point to weld Estimated travel time from the starting point; For weld Welding time; Represents the time-dependent heat cost; node set Where node 0 represents the initial position, node Indicates weld seam Directed edge set , indicating from the weld To the weld The transfer; Indicates weld seam The initial temperature.

6. The heat-sensing autonomous welding path planning method according to claim 5, characterized in that, The reward function of the near-end policy optimization algorithm is designed as follows: ； in: For the cost of travel time; This represents the current welding time of the weld seam. This represents an increase in heat cost; For value function estimation, This represents the discount factor, used to measure the weight of future rewards on current decisions; The PPO algorithm uses a policy network. Generate an action probability distribution and perform stable training by pruning the proxy objective function: ； in For probability ratios, For the dominant function, It is the clipping coefficient for controlling the update magnitude of the policy in the PPO algorithm.

7. The heat-sensing autonomous welding path planning method according to claim 6, characterized in that, The continuous programming model is as follows: ； ； ； ； in, Indicates the first Parametric curves of the path, for The first derivative of represents the second derivative. Indicates the first Path in parameters curvature at that point Indicates the maximum permissible curvature of the path.

8. The heat-sensing autonomous welding path planning method according to claim 7, characterized in that, The step of inputting the optimal weld sequence into a continuous programming model and generating a smooth path that satisfies obstacle avoidance and kinematic constraints based on the CHOMP algorithm is as follows: First, the optimal weld sequence is parameterized as follows: And construct the objective function for the CHOMP algorithm optimization: ； Where: the first term is the smoothing term, which minimizes the trajectory kinetic energy; the second term... The potential field function includes: obstacle potential field: Encourage staying away from obstacles; thermal field: Avoid high-temperature areas; Then, the weld path is updated using the covariant gradient descent method. The update formula for the covariant gradient descent method is as follows: ; in, For the new weld path, This is the quality matrix.

9. A heat-sensing autonomous welding path planning system, characterized in that, The device includes a memory and a processor. The memory includes a program for a heat-sensing autonomous welding path planning method. When the processor executes the program for the heat-sensing autonomous welding path planning method, it implements the steps of a heat-sensing autonomous welding path planning method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a thermal sensing autonomous welding path planning method program, which, when executed by a processor, implements the steps of a thermal sensing autonomous welding path planning method as described in any one of claims 1 to 8.

Images