A method, apparatus and device for intelligent control of laser fusion deposition

Abstract

This application discloses an intelligent control method, apparatus, device, and readable storage medium for laser fused deposition, relating to the field of laser fused deposition technology. The method includes: acquiring the target width and height values ​​of a single pass; using laser power, scanning speed, and wire feed speed as optimization variables; optimizing the predicted single pass size to approximate the target width and height values; determining an initial combination of process parameters by combining a machine learning prediction model with a multi-objective optimization algorithm for inverse search; performing laser fused deposition using the initial process parameter combination and acquiring molten pool images in real time; extracting the molten pool region image from the images using an image segmentation model; inputting the molten pool region image into a process parameter prediction model to obtain a set of predicted process parameters reflecting the current molten pool state; calculating the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination; and adjusting the actual process parameters acting on the molten pool in real time based on this deviation.

Description

Technical Field

[0001] This invention relates to the field of laser filament deposition technology, and more specifically to an intelligent control method, apparatus, device, and readable storage medium for laser filament deposition. Background Technology

[0002] Laser filament deposition additive manufacturing technology boasts advantages such as high material utilization, fast deposition rate, and high density of formed parts, making it a promising field for manufacturing complex metal components. However, this technology still faces two major challenges in its industrial application: Optimizing process parameters is challenging: Deposition quality is highly dependent on the matching of process parameters such as laser power, scanning speed, and wire feed speed. Currently, optimal process parameters are mainly determined through trial and error and experience, lacking scientific and efficient optimization methods. Especially when facing multiple target dimensional requirements such as width and height, it is difficult to quickly and accurately obtain one or more sets of optimal process parameter combinations, resulting in long preparation cycles, high costs, and poor consistency in formed dimensions.

[0003] Insufficient stability control during deposition: During deposition, the molten pool state can drift due to factors such as heat accumulation and environmental disturbances. Existing control methods are mostly based on single characteristics such as molten pool width or temperature, adjusting laser power through a PID controller. These methods have significant limitations: 1. Single control dimension: They cannot coordinate the control of multiple geometric features or the overall state of the molten pool. 2. Complex physical quantity conversion: Controller inputs (such as pixel width and temperature) and outputs (laser power) are different physical quantities, making it difficult to establish a control model and resulting in fuzzy adjustments. 3. Lack of foresight: They are "passive response" control methods, unable to predict the required process parameter adjustments based on the overall state of the molten pool.

[0004] Therefore, existing technologies lack a closed-loop control method and system that can intelligently determine initial process parameters before deposition and maintain the stability of the overall state of the molten pool in real time during deposition. This has become a key bottleneck restricting the accuracy and reliability of laser filament deposition technology. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent control method, apparatus, device, and readable storage medium for laser filament deposition. First, through machine learning and multi-objective optimization algorithms, the optimal initial process parameters are automatically and inversely searched based on the target size, replacing the traditional, inefficient trial-and-error approach. Second, during the deposition process, a deep learning model is innovatively used to inversely map real-time molten pool images into predicted process parameter values, and dynamic adjustments are made by comparing the deviations from the initial set values. This strategy, which transforms the control of complex molten pool characteristics into tracking of the same physical quantity parameters, fundamentally simplifies the control logic and significantly improves the accuracy and stability of regulation. This method organically combines offline global optimization with online real-time closed-loop control, systematically ensuring the consistency of the formed dimensions and the stability of the quality throughout the entire process from start to finish, effectively overcoming dynamic interferences such as heat accumulation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent control method for laser filament deposition, the method comprising: The target width and height values ​​of a single pass are obtained. Laser power, scanning speed, and wire feed speed are used as optimization variables. The optimization objective is to make the predicted single pass size approximate the target width and height values. By combining a machine learning prediction model and a multi-objective optimization algorithm, an inverse search is performed to determine a set of initial process parameter combinations. The initial process parameter combinations include laser power, scanning speed, and wire feed speed. Laser fused wire deposition was performed using an initial combination of process parameters, and images of the molten pool were acquired in real time. The molten pool region image was extracted from the images using an image segmentation model. The image of the molten pool region is input into the process parameter prediction model to obtain a set of predicted process parameters that reflect the current state of the molten pool. The deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination is calculated, and the actual process parameters acting on the molten pool are adjusted in real time based on the deviation.

[0007] In some embodiments, laser power, scanning speed, and wire feed speed are used as optimization variables, with the optimization objective being to make the predicted single-pass dimension approximate the target width and height values. A set of initial process parameter combinations is determined by combining a machine learning prediction model with a multi-objective optimization algorithm for inverse search, including: Historical experimental data was acquired, and based on the historical experimental dataset, a machine learning prediction model was trained with laser power, scanning speed, and wire feeding speed as inputs and single-track width and height as outputs. To optimize the prediction of single-track dimensions to approximate the target width and height values, a multi-objective optimization problem is constructed. A multi-objective optimization algorithm is used to solve the multi-objective optimization problem, and a set of process parameters is selected from the obtained Pareto optimal solution set as the initial process parameter combination.

[0008] In some embodiments, a multi-objective optimization problem is constructed with the optimization objective of making the predicted single-track size approximate the target width and height values, including: The current laser power, scanning speed, and wire feed speed are input into the machine learning prediction model to obtain the predicted single-pass size; To optimize the prediction of single-track dimensions to approximate the target width and height values, a multi-objective optimization problem is constructed.

[0009] In some embodiments, adjusting the process parameters actually acting on the molten pool in real time based on the deviation includes: If the deviation is greater than the preset deviation value, the product of the deviation and the preset proportional gain coefficient will be used as the parameter adjustment amount. The sum of the initial process parameter combination and the parameter adjustment amount is taken as the target process parameter combination for the molten pool.

[0010] In some embodiments, the image segmentation model is a deep learning model based on the YOLO architecture, used to segment the molten pool contour region that is not obscured by the wire feed from the acquired molten pool image.

[0011] In some embodiments, the machine learning prediction model is an XGBoost model, the multi-objective optimization algorithm is a decomposition-based multi-objective evolutionary algorithm, and the process parameter prediction model is a convolutional neural network model.

[0012] Secondly, the present invention also provides an intelligent control device for laser filament deposition, the device comprising: The parameter determination module is used to obtain the target width and target height values ​​of the target single pass. Using laser power, scanning speed, and wire feed speed as optimization variables, the optimization objective is to make the predicted single pass size approximate the target width and height values. By combining a machine learning prediction model and a multi-objective optimization algorithm to perform an inverse search, a set of initial process parameter combinations is determined. The initial process parameter combinations include laser power, scanning speed, and wire feed speed. The image segmentation module is used to perform laser fused wire deposition using an initial combination of process parameters, and to acquire images of the molten pool in real time. The image segmentation model is then used to extract the molten pool region image from the image. The parameter prediction module is used to input the image of the molten pool area into the process parameter prediction model to obtain a set of predicted process parameters that reflect the current state of the molten pool. The parameter adjustment module is used to calculate the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination, and adjust the actual process parameters acting on the molten pool in real time based on the deviation.

[0013] Thirdly, the present invention also 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 computer program to implement the intelligent control method for laser filament deposition provided in the first aspect.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent control method for laser filament deposition provided in the first aspect.

[0015] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent control method for laser filament deposition provided in the first aspect.

[0016] The beneficial effects of this invention are as follows: The intelligent control method for laser fused deposition in this invention first obtains the target width and height values ​​of a single pass, using laser power, scanning speed, and wire feed speed as optimization variables. The optimization objective is to make the predicted single-pass size approximate the target width and height values. By combining a machine learning prediction model with a multi-objective optimization algorithm for inverse search, a set of initial process parameter combinations is determined. These initial process parameter combinations include laser power, scanning speed, and wire feed speed. Laser fused deposition is then performed using these initial process parameter combinations, and molten pool images are acquired in real time. An image segmentation model is used to extract the molten pool region image from the images. The molten pool region image is then input into the process parameter prediction model to obtain a set of predicted process parameters reflecting the current molten pool state. Finally, the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination is calculated, and the actual process parameters acting on the molten pool are adjusted in real time based on this deviation. Firstly, by using machine learning and a multi-objective optimization algorithm, the optimal initial process parameters are automatically searched inversely based on the target size, replacing the traditional inefficient trial-and-error approach. Secondly, during the deposition process, a deep learning model is innovatively used to inversely map real-time molten pool images into predicted process parameters, and dynamic adjustments are made by comparing the deviations from the initial setpoints. This strategy, which transforms the control of complex molten pool characteristics into tracking of physical quantity parameters, fundamentally simplifies the control logic and significantly improves the accuracy and stability of regulation. This method organically combines offline global optimization with online real-time closed-loop control, systematically ensuring the consistency of forming dimensions and quality stability throughout the entire process from start to finish, and effectively overcoming dynamic interferences such as heat accumulation.

[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an intelligent control method for laser filament deposition according to an embodiment of the present invention; Figure 2 This is a comparison diagram of different segmentation methods shown in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a process parameter prediction model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the online verification experiment results shown in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the dynamic change of laser power according to an embodiment of the present invention; Figure 6 This is a flowchart illustrating another intelligent control method for laser filament deposition according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an intelligent control device for laser filament deposition according to an embodiment of the present invention; Figure 8 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics; however, not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0021] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0022] In some embodiments, such as Figure 1 As shown, an intelligent control method for laser filament deposition is provided, the specific method including: S101, obtain the target width and target height values ​​of the target single pass, and use laser power, scanning speed and wire feeding speed as optimization variables to make the predicted single pass size approximate the target width and height values ​​as the optimization objective. By combining machine learning prediction model and multi-objective optimization algorithm to perform inverse search, determine a set of initial process parameter combinations.

[0023] The initial process parameter combination includes laser power, scanning speed, and wire feed speed.

[0024] Optionally, one could first obtain the target width and height values ​​of the target single track uploaded by the user, then obtain historical experimental data, and based on the historical experimental dataset, train a machine learning prediction model with laser power, scanning speed, and wire feeding speed as inputs and single track width and height as outputs; construct a multi-objective optimization problem with the optimization objective of making the predicted single track size approximate the target width and height values; use a multi-objective optimization algorithm to solve the multi-objective optimization problem, and select a set of process parameters as the initial process parameter combination from the obtained Pareto optimal solution set.

[0025] The machine learning prediction model is the XGBoost model, and the multi-objective optimization algorithm is a decomposition-based multi-objective evolutionary algorithm.

[0026] To optimize the prediction of single-track dimensions by making them approximate the target width and height values, a multi-objective optimization problem is constructed, which includes: inputting the current laser power, scanning speed, and wire feed speed into the machine learning prediction model to obtain the predicted single-track dimensions; and constructing a multi-objective optimization problem with the goal of making the predicted single-track dimensions approximate the target width and height values.

[0027] For example, firstly, a machine learning prediction model is trained based on historical deposition experimental data. Historical data is obtained by conducting comprehensive, full-factor single-pass deposition experiments on selected materials (such as Ti6Al4V) using a laser fused wire deposition platform. The experiment systematically changes three key parameters: laser power (P), scanning speed (v_s), and wire feed speed (v_f) (e.g., P is set at 9 levels: 1800-2600W, v_s at 7 levels: 4-10mm / s, and v_f at 5 levels: 1.1-1.9m / min), completing a total of 315 effective depositions. For each single pass formed in the experiment, after cutting, grinding, and polishing, the actual width (W_meas) and actual height (H_meas) of its cross-section are precisely measured using an optical microscope, thus constructing a dataset containing the correspondence between "input process parameters [P, v_s, v_f]" and "output actual dimensions [W_meas, H_meas]".

[0028] Using this dataset, a regression model was trained using the XGBoost algorithm. This model takes process parameters [P, v_s, v_f] as input and outputs the predicted single-pass width (W_pred) and height (H_pred), thus establishing a positive nonlinear mapping from the process parameter space to the single-pass size space. Testing showed that the model has high prediction accuracy, with a width prediction accuracy (ACC ~95%~) of 95.45% and a height prediction accuracy (ACC ~90%~) of 93.18%.

[0029] When a new deposition task is required, input the target width (W_target, e.g., 5.0 mm) and target height (H_target, e.g., 0.9 mm) of the single channel to be deposited.

[0030] Then, a multi-objective optimization problem is constructed: Decision variables: laser power (P), scanning speed (v_s), wire feed speed (v_f).

[0031] Optimization objective: Minimize two objective functions: f1 = |W_pred - W_target| and f2 = |H_pred - H_target|. Here, W_pred and H_pred are the predicted channel sizes calculated by the XGBoost prediction model established in the previous step based on the current decision variable values.

[0032] Constraints: The range of values ​​for decision variables is limited to the effective process window explored in historical experiments, for example: 1800≤P≤2600 (W), 4≤v_s≤10 (mm / s), 1.1≤v_f≤1.9 (m / min).

[0033] A decomposition-based multi-objective evolutionary algorithm (MOEA / D) is used to solve the above optimization problem. The search process is as follows: a) Initialization: Within the constraints of the decision variables, the MOEA / D algorithm randomly generates an initial set of process parameter combinations (population). b) Evaluation: Each process parameter combination [P_i, v_si, v_fi] in the population is input into the trained XGBoost prediction model to calculate the corresponding prediction size [W_pred_i, H_pred_i], and then its two corresponding objective values ​​f1_i and f2_i are calculated according to the objective function. c) Evolution: The MOEA / D algorithm performs non-dominated sorting and decomposition based on the objective values ​​of the individuals in the population. Through evolutionary operations such as selection, crossover, and mutation, a new generation of process parameter combination populations with better performance is generated. d) Iteration: Steps b) and c) are repeated for multiple generations of evolution. e) Output: When the evolution reaches a preset termination condition (such as the maximum number of generations), the algorithm outputs a Pareto optimal solution set. The solution set contains multiple sets of process parameters, each of which achieves an optimal trade-off between the two objectives of "width error" and "height error," meaning that one objective cannot be further improved without worsening the other.

[0034] Finally, from the Pareto optimal solution set provided by the MOEA / D algorithm, a set of process parameters is selected according to preset rules (e.g., selecting the solution with the minimum sum of f1+f2, or selected by the operator based on process preferences), such as [P0=2150W, v_s0=6.2mm / s, v_f0=1.5m / min]. This set of parameters is determined as the initial combination of process parameters to start this deposition. The effectiveness of this result has been experimentally verified (see Table 1), and the actual single-pass size deposited using this parameter combination is very close to the target size.

[0035] S102 uses an initial combination of process parameters to perform laser fused wire deposition and acquires images of the molten pool in real time. The molten pool region image is extracted from the image using an image segmentation model.

[0036] Among them, the image segmentation model is a deep learning model based on the YOLO architecture, which is used to segment the molten pool contour region that is not obscured by the wire feed from the acquired molten pool image.

[0037] For example, a predetermined combination of initial process parameters (e.g., [P0=2150W, v_s0=6.2mm / s, v_f0=1.5m / min]) is sent to the actuator. The laser emits light at the set power (P0), the wire feeder feeds the wire at the set speed (v_f0), and the motion mechanism (such as a robot) drives the processing head to move along a predetermined trajectory at the set speed (v_s0), thus initiating laser filament deposition.

[0038] Simultaneously, a rangefinder industrial camera (such as a Basler ace area array camera, equipped with a 590-710nm bandpass filter to suppress laser and background interference), mounted above the machining head, begins continuously acquiring dynamic images of the molten pool area at a high frame rate (e.g., 50 fps). The camera transmits the image data to the host computer (data processing and control module) in real time via a USB 3.0 interface. To ensure strict synchronization between the image and the process, the timestamp of the image and the sending time of the process parameter commands are coordinated by the same clock source.

[0039] Upon receiving the real-time image stream, each frame is immediately processed as follows: a) Image cropping: To reduce computation and focus on the region of interest, the original high-resolution image (e.g., 2448×2048 pixels) is uniformly cropped into a fixed-size (e.g., 640×640 pixels) central region, which covers the molten pool and its surrounding key features. b) Model inference: The cropped image is input into a pre-trained image segmentation model for inference. In this embodiment, the model is a deep learning model based on the YOLOv8n architecture and fine-tuned for the molten pool segmentation task. The core function of this model is to identify and segment the molten pool region partially obscured by the metal wire in the image. c) Output result: The model outputs a binary mask image with the same resolution as the input image. In this mask image, pixels belonging to the molten pool region are marked as foreground (e.g., pixel value 255), while the background (including wire, substrate, spatter, etc.) is marked as background (pixel value 0). Figure 2 As shown, Figure 2 The comparison diagram shows the effect of different segmentation methods. This method can effectively overcome the problem of broken molten pool contour caused by wire occlusion in the traditional threshold segmentation method and extract the complete molten pool area.

[0040] The image segmentation model is established as follows: First, a large number of melt pool image samples (e.g., 2850 images) are simultaneously acquired through a series of single-pass deposition experiments under different process parameters (e.g., using a central composite design experimental scheme). Then, annotation tools (such as Labelme) are used to accurately annotate the melt pool contours in the images using artificial polygons, generating label data. Based on this dataset, the YOLOv8n model is transferred and trained, enabling it to learn the features for accurately segmenting the melt pool from complex deposition scenarios. The trained model is then packaged and deployed in the inference engine of the host computer for real-time invocation.

[0041] After the above steps, the final image of the molten pool region corresponding to each frame of real-time image, after removing the wire occlusion (i.e., a binary mask, or one that can be converted into contour coordinates), is obtained. This region image is the direct input for subsequent process parameter prediction and is immediately passed to the next processing module (i.e., the process parameter prediction model).

[0042] This embodiment demonstrates that the process achieves high-frame-rate, automated, and high-precision perception and feature extraction of visual information of the melt pool while deposition is in progress, providing a stable and reliable real-time input for closed-loop control.

[0043] S103, input the image of the molten pool region into the process parameter prediction model to obtain a set of predicted process parameters reflecting the current state of the molten pool.

[0044] The process parameter prediction model is a convolutional neural network model.

[0045] For example, a real-time image of the molten pool region (i.e., a binary mask or contour map) from the previous step is received. To meet the input requirements of the process parameter prediction model, this image undergoes normalization preprocessing: First, the image size is uniformly scaled to a fixed resolution (e.g., 64×64 pixels). Then, the pixel values ​​are normalized to the [0, 1] interval. Simultaneously, the molten pool temperature measurement value (T) is read in real time from a coaxially mounted dual-color infrared thermometer. This temperature data has been preprocessed through filtering and outlier removal (e.g., using interquartile range) to reflect a stable molten pool thermal state.

[0046] Finally, a data sample integrating multimodal features is constructed: the preprocessed molten pool region image is used as a two-dimensional input channel, and the synchronized molten pool temperature value (T) is used as a scalar input feature.

[0047] The constructed fused data samples are input into a pre-trained process parameter prediction model. In this embodiment, the model is a specially designed convolutional neural network, and its network structure is as follows: Figure 3 As shown, Figure 3This is a schematic diagram of the process parameter prediction model, mainly consisting of: Feature extraction: Composed of three convolutional modules connected in series, each module includes a convolutional layer (Conv2D), a batch normalization layer (BatchNorm2d), and a ReLU activation function. This part is responsible for automatically extracting deep visual features related to process parameters (such as melt pool morphology and contour complexity) from the input melt pool region image. Feature fusion and mapping: The visual feature map extracted by the CNN is flattened into a feature vector, and this vector is concatenated with the aforementioned melt pool temperature scalar feature (T) at a specific network layer. The concatenated fused feature vector is then subjected to nonlinear transformation and information integration through two fully connected layers. Output layer: The last fully connected layer outputs three consecutive values, representing the three predicted process parameters mapped by the model to the current melt pool state: predicted laser power (P_pred), predicted scanning speed (v_s_pred), and predicted wire feed speed (v_f_pred). These three values ​​together constitute the set of predicted process parameters [P_pred, v_s_pred, v_f_pred].

[0048] The CNN model was built based on a large amount of offline training data. The training dataset was constructed as follows: during full-factor single-pass deposition experiments, multiple frames of molten pool images and average temperatures were simultaneously collected for each set of stable process parameters [P_true, v_s_true, v_f_true], forming a paired sample library of "molten pool image-temperature" and "real process parameters". The CNN model was trained in a supervised manner using this dataset, with the mean squared error (MSE) between the model's predicted parameters and the actual set parameters as the loss function. The model was validated on the test set, demonstrating high prediction accuracy. For the prediction of laser power, scanning speed, and wire feed speed, the ACC (accuracy) was ~90%~ (prediction error within 10%), all exceeding 98% (see Table 2). Figure 4 The schematic diagram of the online verification experiment shown demonstrates that during the deposition process, the model can output stable and nearly true setpoint predicted parameters in real time, and can respond to dynamic changes in process parameters. Figure 5 ), Figure 5 The diagram shows the dynamic changes in laser power, where (a) represents the forming result and (b) represents the prediction result.

[0049] After the model inference is completed, the set of predicted process parameters [P_pred(k), v_s_pred(k), v_f_pred(k)] corresponding to the current time (the k-th frame) is immediately obtained. This set is passed to the next control loop (i.e., the deviation calculation and adjustment module) in real time as a quantitative indicator characterizing the actual state of the current molten pool.

[0050] This embodiment demonstrates that the process, through a deep learning multimodal fusion model, enables the real-time and automatic reverse mapping of unstructured and difficult-to-control molten pool visual and thermal state information into structured and directly controllable process parameters, providing crucial decision-making basis for subsequent precise closed-loop control based on parameter deviations.

[0051] S104, calculate the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination, and adjust the actual process parameters acting on the molten pool in real time based on the deviation.

[0052] The process parameters that actually act on the molten pool are adjusted in real time based on the deviation, including: if the deviation is greater than the preset deviation value, the product of the deviation and the preset proportional gain coefficient is used as the parameter adjustment amount; the sum of the initial process parameter combination and the parameter adjustment amount is used as the target process parameter combination of the molten pool.

[0053] For example, in each control cycle k, the control module receives two key inputs: an initial combination of process parameters [P0, v_s0, v_f0] characterizing the desired target state, and a set of predicted process parameters [P_pred(k), v_s_pred(k), v_f_pred(k)] characterizing the current sensing state.

[0054] First, calculate the real-time deviation value e(k) between the two in each process parameter dimension: laser power deviation: e_P(k) = P0 - P_pred(k); scanning speed deviation: e_vs(k) = v_s0 - v_s_pred(k); wire feed speed deviation: e_vf(k) = v_f0 - v_f_pred(k).

[0055] To avoid overreacting to sensor noise or minor fluctuations in model predictions, a permissible deviation threshold (i.e., dead zone) is set for each process parameter. For example, the activation threshold Th_P for laser power is set to 60W. The absolute value of each deviation is checked against its corresponding threshold: an adjustment to the laser power is triggered only if |e_P(k)| > Th_P. Otherwise, the current laser power setting remains unchanged.

[0056] In the multi-pass overlap verification experiment of this embodiment, to simplify control and avoid affecting the amount of material input, only the laser power was selected as the controlled variable. Therefore, only e_P(k) was judged and subsequently adjusted, while the scanning speed and wire feed speed remained unchanged at their initial settings v_s0 and v_f0 during the deposition process.

[0057] When an adjustment is deemed necessary, an incremental proportional control strategy is used to calculate the adjustment amount. For laser power, the formula for calculating the adjustment amount ΔP(k) is: Wherein, K_P is the preset proportional gain coefficient, which is set to 0.01 in this embodiment. Subsequently, the new laser power setting value P_new(k) is calculated: P_new(k) = P_current(k-1) + ΔP(k); where, P_current(k-1) is the power value actually output to the laser in the previous control cycle.

[0058] New process parameter commands [P_new(k), v_s0, v_f0] are sent in real time to the actuators (laser, robot, and wire feeder) via a communication interface (such as TCP / IP). After receiving the command, the laser updates its internal power setpoint to P_new(k), thereby realizing real-time correction of the actual laser energy acting on the molten pool within the control cycle.

[0059] The above process (acquisition → segmentation → prediction → deviation calculation → judgment → adjustment → output) is executed cyclically within each control cycle until single-pass deposition is completed. This closed-loop control mechanism can effectively suppress the growth of the molten pool size caused by heat accumulation, ensuring that the predicted process parameters and the actual molten pool state remain dynamically stable around the initial set values, ultimately obtaining uniformly sized formed parts (as shown in Table 3).

[0060] This embodiment demonstrates that the process transforms the "parameter deviation" output by the predictive model, which characterizes the deviation from the state, into "adjustment instructions" for the actuator through explicit mathematical rules and judgment logic, thereby achieving fully automatic, real-time closed-loop control from state perception to precise execution.

[0061] The intelligent control method for laser filament deposition in the above embodiments first obtains the target width and height values ​​of a single pass. Using laser power, scanning speed, and wire feed speed as optimization variables, the optimization objective is to make the predicted single-pass size approximate the target width and height values. By combining a machine learning prediction model with a multi-objective optimization algorithm for inverse search, a set of initial process parameter combinations is determined. These initial process parameter combinations include laser power, scanning speed, and wire feed speed. Laser filament deposition is then performed using these initial process parameter combinations, and molten pool images are acquired in real time. An image segmentation model is used to extract the molten pool region image from the images. The molten pool region image is then input into the process parameter prediction model to obtain a set of predicted process parameters reflecting the current molten pool state. Finally, the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination is calculated, and the actual process parameters acting on the molten pool are adjusted in real time based on this deviation. Firstly, by using machine learning and a multi-objective optimization algorithm, the optimal initial process parameters are automatically searched inversely based on the target size, replacing the traditional inefficient trial-and-error approach. Secondly, during the deposition process, a deep learning model is innovatively used to inversely map real-time molten pool images into predicted process parameters, and dynamic adjustments are made by comparing the deviations from the initial setpoints. This strategy, which transforms the control of complex molten pool characteristics into tracking of physical quantity parameters, fundamentally simplifies the control logic and significantly improves the accuracy and stability of regulation. This method organically combines offline global optimization with online real-time closed-loop control, systematically ensuring the consistency of forming dimensions and quality stability throughout the entire process from start to finish, and effectively overcoming dynamic interferences such as heat accumulation.

[0062] To more comprehensively demonstrate this solution, this embodiment presents an optional method for intelligent control of laser filament deposition, such as... Figure 6 As shown: S201, obtain the target width and target height values ​​of a single target lane.

[0063] S202: Acquire historical experimental data and, based on the historical experimental dataset, train a machine learning prediction model with laser power, scanning speed, and wire feeding speed as inputs and single-track width and height as outputs.

[0064] S203 inputs the current laser power, scanning speed, and wire feed speed into the machine learning prediction model to obtain the predicted single-pass size.

[0065] S204. To optimize the prediction of a single track size by making it approximate the target width and height values, a multi-objective optimization problem is constructed.

[0066] S205 uses a multi-objective optimization algorithm to solve the multi-objective optimization problem, and selects a set of process parameters from the obtained Pareto optimal solution set as the initial process parameter combination.

[0067] S206 uses an initial combination of process parameters for laser fused wire deposition and acquires images of the molten pool in real time. The molten pool region image is then extracted from the image using an image segmentation model.

[0068] S207, Input the image of the molten pool region into the process parameter prediction model to obtain a set of predicted process parameters reflecting the current state of the molten pool.

[0069] S208, calculate the deviation between the predicted set of process parameters and the corresponding parameters in the initial set of process parameters.

[0070] S209, if the deviation is greater than the preset deviation value, the product of the deviation and the preset proportional gain coefficient will be used as the parameter adjustment amount.

[0071] S210, the sum of the initial process parameter combination and the parameter adjustment amount is taken as the target process parameter combination of the molten pool.

[0072] The specific processes of S201-S210 described above can be found in the description of the above method embodiments. Their implementation principles and technical effects are similar, and will not be repeated here.

[0073] Based on the same inventive concept, this application also provides an intelligent control device for laser filament deposition, which implements the intelligent control method for laser filament deposition described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the intelligent control device for laser filament deposition provided below can be found in the limitations of the intelligent control method for laser filament deposition described above, and will not be repeated here.

[0074] In one embodiment, such as Figure 7 As shown, an intelligent control device for laser filament deposition is provided, the device comprising: The parameter determination module 30 is used to obtain the target width and target height values ​​of the target single pass. Using laser power, scanning speed and wire feeding speed as optimization variables, the optimization objective is to make the predicted single pass size approximate the target width and height values. By combining a machine learning prediction model and a multi-objective optimization algorithm to perform an inverse search, a set of initial process parameter combinations is determined. The initial process parameter combinations include laser power, scanning speed and wire feeding speed. The image segmentation module 31 is used to perform laser fused wire deposition using an initial combination of process parameters, and to acquire molten pool images in real time, and to extract the molten pool region image from the image using an image segmentation model. The parameter prediction module 32 is used to input the image of the molten pool area into the process parameter prediction model to obtain a set of predicted process parameters that reflect the current state of the molten pool. The parameter adjustment module 33 is used to calculate the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination, and adjust the actual process parameters acting on the molten pool in real time based on the deviation.

[0075] This application also provides an electronic device, in some embodiments, referring to... Figure 7 As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions that can be executed on the processor 730. The processor 730 can execute the intelligent control method and / or technical solution for laser filament deposition based on the foregoing embodiments by calling the program instructions. The electronic device 700 can be a mobile terminal device such as a mobile phone or a computer.

[0076] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program that performs an intelligent control method for laser filament deposition. For example, computer program instructions, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions that invoke the methods of this application may be stored in a fixed or removable storage medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in a storage medium that operates according to the program instructions.

[0077] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0078] The technical features of the above embodiments can be arbitrarily integrated. For the sake of brevity, not all possible integrations of the technical features in the above embodiments are described. However, as long as the integration of these technical features does not contradict each other, they should be considered to be within the scope of this specification.

[0079] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A smart control method for laser filament deposition, characterized in that, The method includes: The target width and height values ​​of a single pass are obtained. Laser power, scanning speed, and wire feed speed are used as optimization variables. The optimization objective is to make the predicted single pass size approximate the target width and height values. By combining a machine learning prediction model and a multi-objective optimization algorithm, an inverse search is performed to determine a set of initial process parameter combinations. The initial process parameter combinations include laser power, scanning speed, and wire feed speed. Laser fused wire deposition is performed using the initial process parameter combination, and molten pool images are acquired in real time. The molten pool region image is extracted from the images using an image segmentation model. The image of the molten pool region is input into the process parameter prediction model to obtain a set of predicted process parameters that reflect the current state of the molten pool. The deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination is calculated, and the actual process parameters acting on the molten pool are adjusted in real time based on the deviation.

2. The intelligent control method for laser filament deposition as described in claim 1, characterized in that, Using laser power, scanning speed, and wire feed speed as optimization variables, and aiming to make the predicted single-pass dimension approximate the target width and height values, a set of initial process parameter combinations is determined through inverse search by combining a machine learning prediction model and a multi-objective optimization algorithm, including: Historical experimental data was acquired, and based on the historical experimental dataset, a machine learning prediction model was trained with laser power, scanning speed, and wire feeding speed as inputs and single-track width and height as outputs. To optimize the prediction of a single track size by approximating the target width and height values, a multi-objective optimization problem is constructed. A multi-objective optimization algorithm is used to solve the multi-objective optimization problem, and a set of process parameters is selected from the obtained Pareto optimal solution set as the initial process parameter combination.

3. The intelligent control method for laser filament deposition as described in claim 2, characterized in that, To optimize the prediction of single-track dimensions to approximate the target width and height values, a multi-objective optimization problem is constructed, including: The current laser power, scanning speed, and wire feed speed are input into the machine learning prediction model to obtain the predicted single-pass size; To optimize the prediction of a single track dimension to approximate the target width and height values, a multi-objective optimization problem is constructed.

4. The intelligent control method for laser filament deposition as described in claim 1, characterized in that, Based on this deviation, the actual process parameters acting on the molten pool are adjusted in real time, including: If the deviation is greater than the preset deviation value, the product of the deviation and the preset proportional gain coefficient will be used as the parameter adjustment amount. The sum of the initial process parameter combination and the parameter adjustment amount is taken as the target process parameter combination for the molten pool.

5. The intelligent control method for laser filament deposition as described in claim 1, characterized in that, The image segmentation model is a deep learning model based on the YOLO architecture, used to segment the molten pool contour region that is not obscured by the wire feed from the acquired molten pool image.

6. The intelligent control method for laser filament deposition as described in claim 1, characterized in that, The machine learning prediction model is the XGBoost model, the multi-objective optimization algorithm is a decomposition-based multi-objective evolutionary algorithm, and the process parameter prediction model is a convolutional neural network model.

7. An intelligent control device for laser filament deposition, characterized in that, The device includes: The parameter determination module is used to obtain the target width and target height values ​​of the target single pass, and uses laser power, scanning speed and wire feeding speed as optimization variables. The optimization objective is to make the predicted single pass size approximate the target width and height values. By combining a machine learning prediction model and a multi-objective optimization algorithm to perform an inverse search, a set of initial process parameter combinations is determined. The initial process parameter combinations include laser power, scanning speed and wire feeding speed. The image segmentation module is used to perform laser fused wire deposition using the initial process parameter combination, and to acquire molten pool images in real time, and to extract the molten pool region image from the images using the image segmentation model; The parameter prediction module is used to input the image of the molten pool area into the process parameter prediction model to obtain a set of predicted process parameters that reflect the current state of the molten pool. The parameter adjustment module is used to calculate the deviation between the predicted process parameter set and the corresponding parameters in the initial process parameter combination, and adjust the actual process parameters acting on the molten pool in real time based on the deviation.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent control method for laser filament deposition as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent control method for laser filament deposition as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent control method for laser filament deposition as described in any one of claims 1 to 6.

Images