A laser-assisted robot machining method and device based on digital twinning

Abstract

The application provides a laser-assisted robot machining method and device based on digital twinning, relates to the technical field of intelligent manufacturing and precision machining, and comprises the following steps: collecting machining state data and workpiece surface space coordinates of a current working step through a multi-sensor array; inputting the machining state data and the workpiece surface space coordinates into a forward-looking prediction model to obtain a workpiece deformation amount prediction value; when the workpiece deformation amount prediction value is greater than a maximum deformation amount threshold value, solving and obtaining an optimal adjustment amount through a Bayesian optimization algorithm based on the workpiece deformation amount prediction value and the maximum deformation amount threshold value, and adjusting process parameters of the current working step according to the optimal adjustment amount; collecting actual point cloud data of a current machined surface, and obtaining an ideal model; obtaining a defect position distribution and a deviation amount set according to the actual point cloud data and the ideal model; and adjusting process parameters of a next working step of a double-robot arm robot according to the defect position distribution and the deviation amount set.

Description

Technical Field

[0001] This application relates to the fields of intelligent manufacturing and precision machining technology, and in particular to a laser-assisted robot machining method and apparatus based on digital twins. Background Technology

[0002] With the development of aerospace and other fields, the demand for precision machining of large-sized, complex, thin-walled parts is increasing, and dual-arm laser-assisted machining is widely used in the field of precision machining.

[0003] Existing dual-arm laser-assisted processing methods lack effective digital twin pre-simulation tools, making it difficult to identify risks in advance and leading to frequent failures. In the online processing stage, multi-source data is not integrated, and the control strategy cannot adaptively adjust process parameters according to the material state, making it difficult to suppress processing deformation. In the post-processing stage, there is a lack of a mechanism to feed the detection results back to the next step for parameter compensation, resulting in error accumulation, low processing efficiency and workpiece yield, and affecting workpiece quality. Summary of the Invention

[0004] To address the aforementioned problems, in a first aspect, the present invention provides a laser-assisted robot processing method based on digital twins, comprising: The machining program of the dual-arm robotic robot is optimized to obtain an optimized machining program. The process parameters for the current step are obtained based on the optimized processing program. The dual-arm robot performs the processing operation based on the process parameters for the current step and collects the processing status data and workpiece surface spatial coordinates for the current step through a multi-sensor array. The processing status data and workpiece surface spatial coordinates are input into the look-ahead prediction model to predict the deformation amount and obtain the predicted value of the workpiece deformation amount. A maximum deformation amount threshold is set. When the predicted value of the workpiece deformation amount is greater than the maximum deformation amount threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted value of the workpiece deformation amount and the maximum deformation amount threshold. The process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step. The dual-arm robot performs the processing operation according to the optimized process parameters of the current step. After the current step is completed, the current processed surface is formed. The actual point cloud data of the current processed surface is collected by a laser contour scanner to obtain the ideal model corresponding to the current processed surface. Defect analysis is performed based on actual point cloud data and an ideal model to obtain the defect location distribution and deviation set; based on the defect location distribution and deviation set, the process parameters for the next step of the dual-arm robot are adjusted.

[0005] In one embodiment, the machining program of the dual-arm robot is optimized to obtain an optimized machining program, including: The initial machining program, geometric model parameters, and motion constraint parameters of the dual-arm robot are obtained, and a digital twin model is constructed based on the geometric model parameters and motion constraint parameters. The initial machining program is imported into the digital twin model for machining simulation and program correction to obtain an optimized machining program.

[0006] In one embodiment, after collecting the processing status data of the current step, the method further includes: Input the processing status data of the current step into the thermo-coupling field mapping model to perform thermo-value mapping and obtain the thermo-coupling field; The thermal coupling field is converted into a corresponding color and mapped to the digital twin model using a vertex color mapping algorithm to obtain a visualized workpiece model.

[0007] In one embodiment, the look-ahead prediction model includes a Branch network and a Trunk network. The Branch network includes multi-layer LSTM units and fully connected layers, and the Trunk network includes multi-layer fully connected neural networks. The model inputs processing state data and workpiece surface spatial coordinates into the look-ahead prediction model to predict deformation, obtaining the predicted workpiece deformation value, including: Processing status data is input into a multi-layer LSTM unit for multi-layer temporal feature extraction. The hidden state feature vector is output by the last LSTM unit and mapped to a basis coefficient vector through a fully connected layer. The workpiece surface spatial coordinates are input into a multi-layer fully connected neural network for multi-layer spatial feature extraction, and the last fully connected neural network outputs a basis function vector. The predicted value of workpiece deformation is calculated based on the basis coefficient vector and basis function vector.

[0008] In one embodiment, the optimal adjustment amount is obtained by solving a Bayesian optimization algorithm based on the predicted workpiece deformation and the maximum deformation threshold, including: Laser power, feed rate, and cutting speed in the process parameters are used as decision variables, and the physical constraint range of the decision variables is set. With minimizing the deviation between the predicted workpiece deformation and the maximum deformation threshold as the optimization objective, a target deviation acquisition function is constructed. Within physical constraints, the optimal adjustment amount that minimizes the target deviation acquisition function is iteratively solved using the Bayesian optimization algorithm. The optimal adjustment amount includes the optimal laser power adjustment, the optimal feed rate adjustment, and the optimal cutting speed adjustment.

[0009] In one embodiment, defect analysis is performed based on actual point cloud data and an ideal model to obtain a set of defect location distributions and deviations, including: The actual point cloud data is registered with the ideal model. The signed distance between each point in the registered actual point cloud data and the surface of the ideal model is calculated. The signed distance is used as the deviation of each point. The deviation of all points constitutes the deviation set. Points whose absolute value of the deviation exceeds the preset range are recorded as defect points. The defect location distribution is formed by all defect points.

[0010] In one embodiment, the process parameters for the next step of the dual-arm robot are adjusted based on the defect location distribution and deviation set, including: The defect area on the currently machined surface is determined based on the defect location distribution. The deviation magnitude and direction of the defect area are determined based on the deviation set. The type of defect area is determined based on the deviation magnitude and direction. The types of defect areas include undercut areas and overcut areas. For undercut areas, increase the laser power or decrease the feed rate in the process parameters of the next step; for overcut areas, decrease the laser power or increase the feed rate in the process parameters of the next step.

[0011] Secondly, the present invention provides a laser-assisted robot processing device based on digital twins, used to implement the aforementioned laser-assisted robot processing method based on digital twins, the device comprising: The machining program optimization module is used to optimize the machining program of the dual-arm robotic robot to obtain an optimized machining program. The data acquisition module is used to obtain the process parameters of the current step according to the optimized processing program. The dual robotic arm robot performs processing operations according to the process parameters of the current step and collects the processing status data and workpiece surface spatial coordinates of the current step through a multi-sensor array. The current step adjustment module is used to input the processing status data and workpiece surface spatial coordinates into the look-ahead prediction model to predict the deformation amount and obtain the predicted value of the workpiece deformation amount; set the maximum deformation amount threshold; when the predicted value of the workpiece deformation amount is greater than the maximum deformation amount threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted value of the workpiece deformation amount and the maximum deformation amount threshold; the process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step; the dual-arm robot performs the processing operation according to the optimized process parameters of the current step. The point cloud acquisition module is used to form the currently processed surface after the current process step is completed. It acquires the actual point cloud data of the currently processed surface through a laser contour scanner to obtain the ideal model corresponding to the currently processed surface. The next step adjustment module is used to perform defect analysis based on actual point cloud data and ideal model to obtain the defect location distribution and deviation set; and to adjust the process parameters of the next step of the dual-arm robot based on the defect location distribution and deviation set.

[0012] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned laser-assisted robot processing method based on digital twins.

[0013] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned laser-assisted robot processing method based on digital twins.

[0014] The present invention has the following beneficial effects: 1. Through digital twin pre-simulation and program correction in the pre-processing stage, risks such as collisions between dual robotic arms, overtravel, and laser misalignment can be identified in advance, avoiding on-site failures, improving processing safety, and shortening the debugging cycle. In the online processing stage, processing status data and workpiece spatial coordinates are integrated, and a look-ahead prediction model is used to predict deformation in real time. When the predicted value exceeds the threshold, Bayesian optimization is used to quickly solve for the optimal adjustment amount, and the process parameters of the current step are controlled in real time in a closed loop, thereby actively suppressing processing deformation and significantly improving processing quality and consistency. After the step is completed, the actual point cloud is registered with the ideal model, and the signed deviation is calculated to obtain the defect location distribution and deviation set. Based on this, the process parameters of the next step are automatically adjusted to achieve defect compensation between steps, avoid error accumulation, and effectively improve processing efficiency and workpiece yield.

[0015] 2. By constructing a look-ahead prediction model based on Branch and Trunk networks, the temporal features extracted by the Branch network are combined with the spatial coordinate features extracted by the Trunk network. The predicted workpiece deformation at any point on the workpiece surface is calculated in real time using the dot product of basis coefficients and basis functions plus bias, achieving high-precision and interpretable online deformation prediction. When the predicted deformation exceeds the threshold, a Bayesian optimization algorithm is used with laser power, feed rate, and cutting speed as decision variables. The goal is to minimize the squared deviation between the predicted deformation and the target threshold. Within the physical constraints, the optimal adjustment amount is solved rapidly through iteration, realizing real-time closed-loop adaptive control of the process parameters of the current step, which significantly reduces the machining deformation of the workpiece.

[0016] 3. By registering the actual point cloud with the ideal model and calculating the signed distance as the deviation, it is possible to accurately distinguish between undercut and overcut areas. The defect location distribution is formed by points whose absolute value of the deviation exceeds the preset tolerance range, realizing the spatial positioning and orientation identification of processing defects, and providing a reliable basis for differentiated compensation. The laser power is increased or the feed speed is reduced to enhance the removal capacity for undercut areas, and the laser power is reduced or the feed speed is increased to slow down the cutting for overcut areas. Thus, the defects existing on the currently processed surface are automatically compensated in the next process step, forming a closed-loop error compensation between processes. This effectively avoids the accumulation of defects and significantly improves the consistency of multi-process processing and the final yield. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a structural diagram of a dual-arm robotic robot. Figure 3 This is a structural diagram of the device according to an embodiment of the present invention; Explanation of reference numerals in the attached figures: 100-Six-DOF industrial robotic arm; 200-Laser generator; 300-Cutting tool; 400-Cutting force sensor; 500-Laser profile scanner; 600-Infrared thermal imager. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0020] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0021] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0022] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0023] Reference Figure 1 The present invention provides a laser-assisted robot processing method based on digital twin, including steps S100-S500.

[0024] The S100 performs machining program optimization on the dual-arm robot to obtain an optimized machining program.

[0025] In some embodiments, the machining program of the dual-arm robot is optimized to obtain an optimized machining program, including: The initial machining program, geometric model parameters, and motion constraint parameters of the dual-arm robot are obtained, and a digital twin model is constructed based on the geometric model parameters and motion constraint parameters. The initial machining program is imported into the digital twin model for machining simulation and program correction to obtain an optimized machining program.

[0026] In some embodiments, the hardware layout of the dual-arm robotic robot is as follows: Figure 2 As shown, two six-degree-of-freedom industrial robotic arms 100 are arranged face-to-face on both sides of the workpiece. A laser generator 200 is mounted at the end of the left robotic arm, and a cutting tool 300 is mounted at the end of the right robotic arm. A laser contour scanner 500 is mounted on a vertical guide rail on the side of the worktable and can move up and down along the rail. Its field of view covers the entire workpiece surface, used to collect three-dimensional point cloud data of the processed surface. An infrared thermal imager 600 is arranged on a sliding frame on the outside of the worktable and can move with the processing area. Its field of view always covers the laser-heated area and the surrounding area, used to collect the two-dimensional temperature distribution field of the workpiece surface in real time. A cutting force sensor 400 is integrated between the cutting tool 300 and the robotic arm flange, collecting the cutting force components in three orthogonal directions in real time. A three-axis accelerometer is mounted at the end of the right robotic arm near the cutting tool, used to collect vibration signals during processing. All sensors are connected to the industrial control computer via an EtherCAT bus. The digital twin platform is developed based on the Unity3D engine, runs on the same industrial control computer, and communicates with the physical equipment in real time via the OPCUA protocol.

[0027] Obtain the initial machining program, geometric model parameters, and motion constraint parameters for the dual-arm robot. The initial machining program is standard G-code or a custom robot language instruction, including motion commands for each axis, laser switch and power commands, spindle speed commands, etc. Geometric model parameters include: the lengths of each link in the left arm, the symmetry parameters of the right arm, the laser generator mounting position, and the tool mounting dimensions. Motion constraint parameters include: the range of motion of each joint, the maximum motion speed, and the allowable range of relative velocity deviation at the end effector when the two arms work together.

[0028] Based on the above parameters, a digital twin model completely identical to the physical device is constructed in the Unity digital twin platform. Specifically, 3D CAD models of the robotic arm, laser, tool, workpiece, and worktable are imported and laid out according to the actual assembly relationships. The same kinematic chain relationships are configured via scripts: the transformation matrices of the left robotic arm from the base to the end effector and the right robotic arm are matched with the forward kinematics model in the entity controller. Each joint variable in the digital twin model is bound to the entity controller variables in the OPCUA node, enabling real-time mapping of the physical device's position and attitude data to the twin model. External data input can also drive the twin model for pre-simulation.

[0029] The initial machining program is imported into the digital twin model for machining pre-simulation. During the pre-simulation, the digital twin model runs visually at the actual machining speed, simulating the entire process of dual robotic arm linkage machining. During the pre-simulation, the system automatically monitors the following items:

[0030] Coaxiality of the two robotic arms: whether the laser irradiation point and the cutting point of the tool coincide on the same trajectory line on the workpiece surface, with a deviation threshold set at ±0.5mm.

[0031] Relative motion speed consistency: The instantaneous speed difference between the left and right robotic arms along the processing direction, with a threshold of ±10mm / min.

[0032] Laser irradiation position accuracy: Whether the center of the laser spot falls at a predetermined distance in front of the tool, with a deviation threshold of ±1mm.

[0033] Robotic arm collision risk: If the minimum distance between any two links is less than 10mm, an alarm will be triggered and highlighted.

[0034] Robotic arm overtravel or abnormal posture: whether the angles of each joint exceed the physical limits or approach an unusual posture.

[0035] Consistency between the workpiece shape and the target model after virtual machining: Material removal is simulated through Boolean operations, and the simulation results are compared with the CAD model to detect undercut or overcut areas.

[0036] If any anomaly is detected during the rehearsal, the system will display a warning window on the interface, highlight the location of the anomaly, and pause the rehearsal. The operator must then modify the initial machining program according to the prompts, such as adjusting the trajectory, speed, laser parameters, or the timing of the dual robotic arm coordination, and then rehearse again. This process is repeated until no anomalies are detected. At this point, the current machining program is saved as an optimized machining program to the industrial control computer and sent to the physical robot controller for execution.

[0037] The S200 obtains the process parameters of the current step based on the optimized machining program. The dual-arm robot executes the machining operation based on the process parameters of the current step and collects the machining status data and workpiece surface spatial coordinates of the current step through a multi-sensor array.

[0038] In some embodiments, after actual processing begins, the dual-arm robotic arm obtains the process parameters for the current step, including laser power, based on the optimized processing program. Feed rate Cutting speed The spindle speed, etc., are used to execute the machining operation of the current step.

[0039] The multi-sensor array includes an infrared thermal imager, a cutting force sensor, and a triaxial accelerometer. The infrared thermal imager acquires the surface temperature distribution of the workpiece. ,in The coordinates are two-dimensional pixel coordinates, which are calibrated and converted into three-dimensional coordinates of the workpiece surface. In practical applications, the average temperature of the workpiece surface within the current field of view of the thermal imager is used. .

[0040] Triaxial cutting force collected by cutting force sensor The unit is Newton (N). In practical applications, the magnitude of the resultant cutting force is used. .

[0041] Vibration data collected by a triaxial accelerometer The unit is gravitational acceleration. In practical applications, the root mean square value of vibration data within a time window is used. ,Depend on , , Composition of processing status data.

[0042] Workpiece surface spatial coordinates Generated offline, corresponding to the current position of the tool cutting point in the workpiece coordinate system. Each sampling time... Aligned with sensor data, forming a synchronized data stream. All data is buffered in a circular buffer at a frequency of 100Hz, retaining the most recent data. Historical data at a specific moment.

[0043] In some embodiments, after collecting the processing status data of the current step, the method further includes: Input the processing status data of the current step into the thermo-coupling field mapping model to perform thermo-value mapping and obtain the thermo-coupling field; The thermal coupling field is converted into a corresponding color and mapped to the digital twin model using a vertex color mapping algorithm to obtain a visualized workpiece model.

[0044] In some embodiments, the thermo-coupled field mapping model employs a Fourier Neural Operator (FNO), whose network structure includes: an input layer receiving local temperature field and cutting force data, which, after passing through four layers of Fourier convolutions, takes the machining state data as input and outputs the temperature field of the entire workpiece surface. and stress field The resolution is 256×256. The FNO model has been pre-trained using a triplet database. During training, the input is local sampled data from the simulation, and the output is the complete full-field data. The learning rate is... The training process was repeated 200 times. After training, the FNO was able to extrapolate the entire field distribution from sparse sensor measurements with an error of <5%.

[0045] Vertex color mapping algorithms include temperature mapping, stress mapping, and rendering, specifically: Temperature mapping: Set minimum temperature value Corresponding to blue (RGB(0,0,255)), the maximum value Corresponding to red (RGB(255,0,0)). For any temperature ,exist Linear interpolation within the interval yields RGB values: Red component = Blue component = The green component is 0.

[0046] Stress mapping: Setting the minimum stress value Corresponding to cyan (RGB(0,255,255)), the maximum value For the red area, linear interpolation is also performed.

[0047] Rendering: For each mesh vertex of the digital twin model, a corresponding color is calculated based on its predicted temperature and stress values ​​at the current moment, and rendered onto the model surface in a semi-transparent overlay. The refresh rate is consistent with the sensor frame rate, enabling three-dimensional dynamic visualization of the temperature and stress fields during processing. Operators can observe high-temperature areas or stress concentration areas through a graphical interface to help determine whether the processing status is abnormal.

[0048] The S300 inputs the processing status data and workpiece surface spatial coordinates into the look-ahead prediction model to predict the deformation amount and obtain the predicted workpiece deformation amount. A maximum deformation amount threshold is set. When the predicted workpiece deformation amount is greater than the maximum deformation amount threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted workpiece deformation amount and the maximum deformation amount threshold. The process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step. The dual-arm robot performs the processing operation according to the optimized process parameters of the current step.

[0049] In some embodiments, the look-ahead prediction model includes a Branch network and a Trunk network. The Branch network includes multi-layer LSTM units and fully connected layers, and the Trunk network includes multi-layer fully connected neural networks. The model inputs processing state data and workpiece surface spatial coordinates into the look-ahead prediction model to predict deformation, obtaining the predicted workpiece deformation value, including: Processing status data is input into a multi-layer LSTM unit for multi-layer temporal feature extraction. The hidden state feature vector is output by the last LSTM unit and mapped to a basis coefficient vector through a fully connected layer. The workpiece surface spatial coordinates are input into a multi-layer fully connected neural network for multi-layer spatial feature extraction, and the last fully connected neural network outputs a basis function vector. The predicted value of workpiece deformation is calculated based on the basis coefficient vector and basis function vector.

[0050] In some embodiments, the look-ahead prediction model employs an LSTM-DeepONet hybrid neural network architecture, including a Branch network and a Trunk network. This model has been pre-trained offline and deployed on the GPU of an industrial control computer, with a single inference time of less than 10ms, meeting real-time requirements.

[0051] Branch network uses processing status data For input, where the time window length is... The input vector at each time step Here This represents the magnitude of the resultant cutting force at the current moment (in N). This represents the average temperature (in °C) of the workpiece surface within the current field of view of the thermal imager. The root mean square value (in grams) of the vibration data within a 0.1-second window. The Branch network consists of two LSTM layers, each with 128 hidden units. The input sequence passes through both LSTM layers sequentially. Each LSTM unit contains a forget gate, an input gate, and an output gate, enabling it to selectively remember long-term dependencies. The hidden state vector at the last time step is taken. Then input a fully connected layer (weight matrix size) Bias vector size Output basis coefficient vector The number of basis functions .

[0052] Trunk networks use workpiece surface spatial coordinates For input, a four-layer fully connected neural network is used, with 64, 64, 64 neurons in each layer. The first three layers use ReLU activation functions, while the last layer has no activation function and uses linear output. The Trunk network outputs a basis function vector. Since the Trunk network responds to spatial coordinates statically, the basis function values ​​of the discrete point set on the workpiece surface can be pre-calculated and stored in a lookup table to accelerate inference during actual deployment.

[0053] The expression for the predicted workpiece deformation is: in, Points on the surface of the workpiece The predicted value of workpiece deformation at the location is in millimeters (mm). This represents the number of basis functions, with a value of 10. The summation index ranges from 1 to 10; The first output of the Branch network Individual base coefficients, dimensionless; The first output of the Trunk network The basis functions in coordinates The value at that location is dimensionless; Here, is the bias term, a learnable scalar, and its value after pre-training is [value missing]. .

[0054] The training of the prospective prediction model is divided into two stages: pre-training and fine-tuning. Pre-training uses a triplet database, which is constructed through ABAQUS finite element simulation. A three-dimensional thermo-mechanical coupled model of laser-assisted cutting is established in ABAQUS: the workpiece material is titanium alloy Ti-6Al-4V, the mesh size is 0.5mm, and sequential coupled thermo-mechanical analysis is employed. Full-factor experiments are conducted for different combinations of process parameters: laser power... Six power levels are selected: 500W, 1000W, 1500W, 2000W, 2500W, and 3000W; feed rate Five levels were selected: 100, 200, 300, 400, and 500 mm / min; cutting speed Four levels were selected: 50, 100, 150, and 200 mm / min. Total number of combinations. Each simulation group outputs: temperature field, stress field, and deformation field. The temperature field, stress field, and deformation data are associated and stored as a triplet database, using the SQLite embedded database engine, and integrated into the digital twin system as a mobile plugin.

[0055] During pre-training, the loss function is the mean square error between the predicted deformation and the simulated actual deformation. The optimizer uses Adam, with an initial learning rate of... Batch size 32, training for 100 epochs. After training, the model's MSE on the test set is [value missing]. The average absolute error is 0.005 mm.

[0056] After deploying the pre-trained model, 200 sets of data with actual deformation labels were collected during actual processing. The actual deformation was obtained by measuring the processed workpiece using a coordinate measuring machine and aligned with the sensor data. 80% of the data was used for fine-tuning, and 20% for validation. During fine-tuning, the LSTM layer parameters were kept fixed, and only the parameters of the fully connected layers and the Trunk network were fine-tuned, with the learning rate reduced to [missing value]. The model was trained for 10 epochs. After fine-tuning, the mean absolute error of the model on the validation set was reduced to 0.012 mm, which meets the engineering accuracy requirements.

[0057] In some embodiments, the optimal adjustment amount is obtained by solving a Bayesian optimization algorithm based on the predicted workpiece deformation value and the maximum deformation threshold, including: Laser power, feed rate, and cutting speed in the process parameters are used as decision variables, and the physical constraint range of the decision variables is set. With minimizing the deviation between the predicted workpiece deformation and the maximum deformation threshold as the optimization objective, a target deviation acquisition function is constructed. Within physical constraints, the optimal adjustment amount that minimizes the target deviation acquisition function is iteratively solved using the Bayesian optimization algorithm. The optimal adjustment amount includes the optimal laser power adjustment, the optimal feed rate adjustment, and the optimal cutting speed adjustment.

[0058] In some embodiments, the system sets a maximum deformation threshold. During the processing, the predicted value of workpiece deformation output by the look-ahead prediction model is calculated in real time. .when At that time, the process parameters remain unchanged. At that time, process parameter adjustments are triggered.

[0059] The decision variables are the adjustments to three process parameters: laser power adjustment. (Unit: W) Feed rate adjustment (Unit: mm / min) Cutting speed adjustment amount (Unit: mm / min). Current process parameters are denoted as laser power. Feed rate Cutting speed The adjusted parameters are:

[0060] Set physical constraints on decision variables to ensure that parameters do not exceed equipment capacity and process safety limits: In this embodiment, , ; , ; , The adjustment amount itself is limited by the upper and lower bounds of the parameter, for example... Cannot make Exceeding .

[0061] The optimization objective is to minimize the deviation between the predicted deformation and the maximum deformation threshold. The expression for the target deviation acquisition function is as follows: in, The predicted value of workpiece deformation output by the look-ahead prediction model, i.e. .

[0062] The Bayesian optimization framework is used for iterative solution, and the specific process is as follows: Initialization: Latin hypercube sampling is used in the first 5 evaluations to generate 5 initial candidate points within the constraints of the decision variables. .

[0063] Constructing a Gaussian process surrogate model: The mean function of the Gaussian process is taken as a constant, and the covariance function (kernel function) is selected from the Matérn5 / 2 kernel. This kernel function is suitable for physical process modeling and has second-order differentiability. The hyperparameters of the kernel function are obtained by maximizing the marginal likelihood estimation.

[0064] Acquisition function: A variant of Expected Improvement (EI) is used, and its expression is: ,in Collect function values ​​for the target deviation. This is the current optimal value. Since the optimization objective is to minimize the deviation, the direction of EI improvement is towards a smaller deviation.

[0065] Iteration: In each iteration, the acquisition function is maximized to determine the next evaluation point. Local optimization is performed using the L-BFGS-B algorithm within the physical constraints. The process parameters corresponding to the new evaluation point are sent to the controller, a short-term machining operation (lasting 0.5 seconds) is performed, the predicted deformation value under these parameters is measured, and the Gaussian process model is updated.

[0066] Convergence criterion: When the change in the target deviation acquisition function value is less than 5 times in 5 consecutive iterations. Stop when the maximum number of iterations (20) is reached. Output the optimal combination of adjustment values. Even the decision variable that minimizes the target deviation acquisition function.

[0067] The process parameters of the dual-arm robot for the current step are adjusted in real time according to the optimal adjustment amount: the optimized process parameters are... , , The data is written to the robot controller and laser controller via OPCUA and takes effect immediately. Adjustments are completed in the next control cycle. The system continuously collects new processing status data within the current step, repeating the prediction and optimization steps to form real-time closed-loop control until the current step ends. Experimental results show that, using this method, the maximum workpiece deformation decreased from 0.18 mm without control to 0.048 mm, meeting the requirement of being below the 0.05 mm threshold.

[0068] After the current step of the S400 process is completed, the current processed surface is formed. The actual point cloud data of the current processed surface is collected by the laser contour scanner to obtain the ideal model corresponding to the current processed surface.

[0069] In some embodiments, after the current step is completed, the dual-arm robotic arm pauses its movement, and the laser contour scanner begins to collect actual point cloud data. Based on the machining program of the current step, the system determines the workpiece surface geometry that should be obtained after the step. The system reads the CAD model file for the corresponding step and converts it into an ideal model.

[0070] The S500 performs defect analysis based on actual point cloud data and an ideal model to obtain the defect location distribution and deviation set; based on the defect location distribution and deviation set, it adjusts the process parameters for the next step of the dual-arm robot.

[0071] In some embodiments, defect analysis is performed based on actual point cloud data and an ideal model to obtain a set of defect location distributions and deviations, including: The actual point cloud data is registered with the ideal model. The signed distance between each point in the registered actual point cloud data and the surface of the ideal model is calculated. The signed distance is used as the deviation of each point. The deviation of all points constitutes the deviation set. Points whose absolute value of the deviation exceeds the preset range are recorded as defect points. The defect location distribution is formed by all defect points.

[0072] In some embodiments, the ICP (Iterative Closest Point) algorithm is used to register the actual point cloud data with the ideal model. The steps are as follows: Preprocessing: Downsampling and outlier removal are performed on the actual point cloud.

[0073] Initial alignment: Coarse registration is performed using reference points on the workpiece.

[0074] ICP iteration: Minimize the sum of Euclidean distances from the actual point cloud to the surface of the ideal model, iterating 20 times or until the convergence threshold is reached. mm. After registration, the actual point cloud coordinates are transformed into the local coordinate system of the ideal model.

[0075] The signed distance between each point in the actual point cloud data after point cloud registration and the surface of the ideal model is calculated. The signed distance is defined as follows: along the normal direction of the workpiece surface, the distance is positive when the actual point is outside the ideal model (i.e., material residue causing undercutting); and negative when it is inside the ideal model (i.e., excessive material removal causing overcutting). In engineering, the normal direction is obtained by interpolating the normals of the vertices of the triangular mesh of the ideal model. The signed distance is used as the deviation of each point. The set of deviations consists of the deviations at all points. .

[0076] Set the preset tolerance range as The absolute value of the deviation The points are denoted as defect points, and the defect location distribution is formed by all defect points. Furthermore, continuous defect points (spatial distance less than 1 mm) are clustered into defect regions, and each defect region contains location (centroid coordinates), boundary contour, and deviation statistics.

[0077] In some embodiments, the process parameters for the next step of the dual-arm robot are adjusted based on the defect location distribution and the set of deviations, including: The defect area on the currently machined surface is determined based on the defect location distribution. The deviation magnitude and direction of the defect area are determined based on the deviation set. The type of defect area is determined based on the deviation magnitude and direction. The types of defect areas include undercut areas and overcut areas. For undercut areas, increase the laser power or decrease the feed rate in the process parameters of the next step; for overcut areas, decrease the laser power or increase the feed rate in the process parameters of the next step.

[0078] In some embodiments, the magnitude of the deviation is taken as the average of the absolute values ​​of the deviations at all points within the region. The direction of the deviation is determined by the sign of the deviation amount. If the majority (positive deviation points account for >70%), it is considered an undercut region; if If it is dominant, it is an overcut region.

[0079] For undercut areas (where the actual material is higher than the ideal surface and more needs to be removed), the specific compensation strategy is as follows: If the average undercut amount exist Within the range, laser power increases Or the feed rate is reduced .

[0080] If the average undercut amount Increased laser power Or the feed rate is reduced Furthermore, an additional compensation step could be considered.

[0081] For overcut regions (where the actual material is below the ideal surface and too much has been removed), the compensation strategy is as follows: If the absolute value of the overcut exist Within the range, laser power is reduced. Or the feed rate increases .

[0082] like Laser power reduced Or the feed rate increases Overcutting defects cannot be completely eliminated by subsequent cutting, but the parameters of subsequent steps can be adjusted to avoid further overcutting, and the overcutting location can be recorded for manual evaluation.

[0083] The adjusted process parameters serve as the initial parameters for the next step, executed by a dual-arm robotic system to compensate for machining defects on the currently processed surface. Simultaneously, the system stores the defect location distribution, deviation set, and the process parameters corresponding to the current step in the digital twin system's database, and labels the defects. This database accumulates with each batch of workpieces, providing knowledge support for process planning of subsequent workpieces.

[0084] After all steps are completed, the digital twin system records all process data, including pre-processing simulation logs, online machining sensor data and optimization records, and post-processing defect analysis results. When machining the same or similar workpieces subsequently, the system automatically queries the database during the pre-processing stage, prioritizing the avoidance of process parameter combinations marked with defects and recommending historically optimal parameters. This forms a closed-loop iterative optimization of process knowledge, reducing trial cutting costs and shortening the process debugging cycle.

[0085] The laser-assisted robot machining method described in this embodiment, in the machining verification of titanium alloy aerospace structural parts, improves machining efficiency by about 20% and tool life by 30% compared with traditional methods, reduces surface roughness Ra from 0.8μm to 0.4μm, and achieves surface accuracy tolerance of ±0.05mm, meeting the machining requirements of aerospace parts.

[0086] Reference Figure 3 This invention provides a laser-assisted robot processing device 30 based on digital twins, used to implement a laser-assisted robot processing method based on digital twins. The device includes: The machining program optimization module 31 is used to optimize the machining program of the dual-arm robot to obtain an optimized machining program. The data acquisition module 32 is used to obtain the process parameters of the current step according to the optimized processing program. The dual robotic arm robot performs processing operations according to the process parameters of the current step and collects the processing status data and workpiece surface spatial coordinates of the current step through a multi-sensor array. The current step adjustment module 33 is used to input the processing status data and workpiece surface spatial coordinates into the look-ahead prediction model to predict the deformation amount and obtain the predicted value of the workpiece deformation amount; set the maximum deformation amount threshold; when the predicted value of the workpiece deformation amount is greater than the maximum deformation amount threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted value of the workpiece deformation amount and the maximum deformation amount threshold; the process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step; the dual-arm robot performs the processing operation according to the optimized process parameters of the current step. The point cloud acquisition module 34 is used to form the currently processed surface after the current process step is completed. It acquires the actual point cloud data of the currently processed surface through a laser contour scanner and obtains the ideal model corresponding to the currently processed surface. The next step adjustment module 35 is used to perform defect analysis based on actual point cloud data and ideal model to obtain the defect location distribution and deviation set; and to adjust the process parameters of the next step of the dual-arm robot based on the defect location distribution and deviation set.

[0087] This application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements any of the above-described schemes of a laser-assisted robot processing method based on digital twins.

[0088] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0089] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0090] This application also provides a computer-readable medium storing a computer program that, when executed by a processor, implements the laser-assisted robotic processing method based on digital twins, as described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0091] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0092] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A laser-assisted robot processing method based on digital twins, characterized in that, include: The machining program of the dual-arm robotic robot is optimized to obtain an optimized machining program. The process parameters for the current step are obtained based on the optimized processing program. The dual-arm robot performs the processing operation based on the process parameters for the current step and collects the processing status data and workpiece surface spatial coordinates for the current step through a multi-sensor array. The processing status data and workpiece surface spatial coordinates are input into the look-ahead prediction model to predict the deformation amount and obtain the predicted value of workpiece deformation. Set a maximum deformation threshold. When the predicted workpiece deformation value is greater than the maximum deformation threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted workpiece deformation value and the maximum deformation threshold. The process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step. The dual-arm robot performs the processing operation according to the optimized process parameters of the current step. After the current step is completed, the current processed surface is formed. The actual point cloud data of the current processed surface is collected by a laser contour scanner to obtain the ideal model corresponding to the current processed surface. Defect analysis is performed based on actual point cloud data and ideal models to obtain the defect location distribution and deviation set. Based on the defect location distribution and deviation set, the process parameters for the next step of the dual-arm robot are adjusted.

2. The laser-assisted robot processing method based on digital twin according to claim 1, characterized in that, The machining program of the dual-arm robotic robot is optimized to obtain an optimized machining program, including: The initial machining program, geometric model parameters, and motion constraint parameters of the dual-arm robot are obtained, and a digital twin model is constructed based on the geometric model parameters and motion constraint parameters. The initial machining program is imported into the digital twin model for machining simulation and program correction to obtain an optimized machining program.

3. The laser-assisted robot processing method based on digital twins according to claim 2, characterized in that, After collecting the processing status data for the current step, the following is also included: Input the processing status data of the current step into the thermo-coupling field mapping model to perform thermo-value mapping and obtain the thermo-coupling field; The thermal coupling field is converted into a corresponding color and mapped to the digital twin model using a vertex color mapping algorithm to obtain a visualized workpiece model.

4. The laser-assisted robot processing method based on digital twin according to claim 1, characterized in that, The look-ahead prediction model includes a Branch network and a Trunk network. The Branch network consists of multi-layer LSTM units and fully connected layers, while the Trunk network consists of multi-layer fully connected neural networks. Processing state data and workpiece surface spatial coordinates are input into the look-ahead prediction model to predict deformation, obtaining the predicted workpiece deformation value, including: Processing status data is input into a multi-layer LSTM unit for multi-layer temporal feature extraction. The hidden state feature vector is output by the last LSTM unit and mapped to a basis coefficient vector through a fully connected layer. The workpiece surface spatial coordinates are input into a multi-layer fully connected neural network for multi-layer spatial feature extraction, and the last fully connected neural network outputs a basis function vector. The predicted value of workpiece deformation is calculated based on the basis coefficient vector and basis function vector.

5. The laser-assisted robot processing method based on digital twin according to claim 1, characterized in that, The optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted value of workpiece deformation and the maximum deformation threshold, including: Laser power, feed rate, and cutting speed in the process parameters are used as decision variables, and the physical constraint range of the decision variables is set. With minimizing the deviation between the predicted workpiece deformation and the maximum deformation threshold as the optimization objective, a target deviation acquisition function is constructed. Within physical constraints, the optimal adjustment amount that minimizes the target deviation acquisition function is iteratively solved using the Bayesian optimization algorithm. The optimal adjustment amount includes the optimal laser power adjustment, the optimal feed rate adjustment, and the optimal cutting speed adjustment.

6. The laser-assisted robot processing method based on digital twin according to claim 1, characterized in that, Defect analysis is performed based on actual point cloud data and an ideal model to obtain the defect location distribution and deviation set, including: The actual point cloud data is registered with the ideal model. The signed distance between each point in the registered actual point cloud data and the surface of the ideal model is calculated. The signed distance is used as the deviation of each point. The deviation of all points constitutes the deviation set. Points whose absolute value of the deviation exceeds the preset range are recorded as defect points. The defect location distribution is formed by all defect points.

7. The laser-assisted robot processing method based on digital twin according to claim 1, characterized in that, Based on the defect location distribution and deviation set, the process parameters for the next step of the dual-arm robot are adjusted, including: The defect area on the currently machined surface is determined based on the defect location distribution. The deviation magnitude and direction of the defect area are determined based on the deviation set. The type of defect area is determined based on the deviation magnitude and direction. The types of defect areas include undercut areas and overcut areas. For undercut areas, increase the laser power or decrease the feed rate in the process parameters of the next step; for overcut areas, decrease the laser power or increase the feed rate in the process parameters of the next step.

8. A laser-assisted robot processing device based on digital twins, used to implement the laser-assisted robot processing method based on digital twins as described in any one of claims 1 to 7, characterized in that, The device includes: The machining program optimization module is used to optimize the machining program of the dual-arm robotic robot to obtain an optimized machining program. The data acquisition module is used to obtain the process parameters of the current step according to the optimized processing program. The dual robotic arm robot performs processing operations according to the process parameters of the current step and collects the processing status data and workpiece surface spatial coordinates of the current step through a multi-sensor array. The current step adjustment module is used to input the processing status data and workpiece surface spatial coordinates into the look-ahead prediction model to predict the deformation amount and obtain the predicted value of the workpiece deformation amount; set the maximum deformation amount threshold; when the predicted value of the workpiece deformation amount is greater than the maximum deformation amount threshold, the optimal adjustment amount is obtained by solving the Bayesian optimization algorithm based on the predicted value of the workpiece deformation amount and the maximum deformation amount threshold; the process parameters of the current step of the dual-arm robot are adjusted according to the optimal adjustment amount to obtain the optimized process parameters of the current step; the dual-arm robot performs the processing operation according to the optimized process parameters of the current step. The point cloud acquisition module is used to form the currently processed surface after the current process step is completed. It acquires the actual point cloud data of the currently processed surface through a laser contour scanner to obtain the ideal model corresponding to the currently processed surface. The next step adjustment module is used to perform defect analysis based on actual point cloud data and ideal model to obtain the defect location distribution and deviation set; and to adjust the process parameters of the next step of the dual-arm robot based on the defect location distribution and deviation set.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the laser-assisted robot processing method based on digital twin as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the laser-assisted robot processing method based on digital twin as described in any one of claims 1 to 7.

Images