A method for compensating for arc collapse at the arc extinguishing point of a single-pass multi-layer weld in arc additive manufacturing.

Abstract

This invention discloses a method for compensating for arc-extinguishing collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing, relating to the field of arc additive manufacturing. The method includes: determining corresponding welding process parameters based on welding equipment, welding materials, and welding methods; designing a single-pass multi-layer weld experiment based on the corresponding welding process parameters; constructing a deep neural network with the process parameters as input and the difference between the weld's intermediate height and the arc-extinguishing height, and the compensation length at the collapse point as output; training and evaluating the neural network using experimental data to obtain a prediction model for the weld's arc-extinguishing collapse compensation length; collecting and inputting parameters from the weld formation process into the trained weld compensation prediction model, and outputting the corresponding height difference and compensation length, thereby achieving accurate prediction of the arc-extinguishing collapse compensation length for single-pass multi-layer welds in arc additive manufacturing. This invention effectively solves the defect of arc-extinguishing collapse, and experimental verification shows that the height flatness of the component is improved.

Description

Technical Field

[0001] This invention relates to the field of electric arc additive manufacturing, and specifically to a method for compensating for collapse at the arc extinguishing point of a single-pass multi-layer weld in electric arc additive manufacturing. Background Technology

[0002] With the rapid iterative development of the manufacturing industry, the requirements for metal components are becoming increasingly stringent. Existing traditional manufacturing technologies are struggling to meet the demands of modern industrial development. Therefore, additive manufacturing technology has been proposed and widely applied across various sectors of the manufacturing industry. Additive manufacturing technology involves layering materials from the bottom up, following a pre-planned trajectory and driven by a motion actuator. It is a layer-by-layer additive manufacturing technology that creates something from scratch. It integrates many emerging technologies, including machining technology, digital information technology, computer technology, and materials science. Compared to traditional manufacturing technologies, it offers advantages such as simpler equipment, a wider range of materials, unlimited forming sizes, and high-efficiency, rapid production. Furthermore, it is not limited by traditional manufacturing techniques and can quickly produce small quantities of complex products in a short time.

[0003] Currently, many technical challenges in the field of arc additive manufacturing both domestically and internationally require in-depth research. For example, during continuous deposition, factors such as temperature and heat dissipation conditions can cause changes in the geometric parameters of the deposited weld, leading to problems such as complex path programming, stress concentration in the formed part, poor forming accuracy, and increased surface roughness. A common problem is the significant collapse at the weld extinguishing point during single-pass multi-layer deposition experiments. This is mainly because at the extinguishing point, due to the inertia of the welding torch and the force of the shielding gas, the high-temperature liquid metal does not immediately solidify and continues to flow a short distance along the direction of the welding torch. Simultaneously, the arc extinguishes, and there is no longer any cladding metal to fill the gap, resulting in a decrease in the forming height at the very end of the weld bead, i.e., a collapse at the extinguishing point. Without corresponding measures, this not only hinders the control of the morphology and accuracy of the experimental formed part but also leads to localized wire elongation exceeding the appropriate range in multi-layer weld deposition experiments. This results in problems such as the shielding gas failing to properly protect the cladding metal, arc spatter, and the welding torch colliding with the deposited metal.

[0004] To address the issue of arc-extinguishing collapse in single-pass multi-layer welded parts, Xiong Jun from Harbin Institute of Technology proposed a parallel reciprocating cladding method in his research on the influence of GAM additive manufacturing technology on the morphology of single-pass multi-layer welded parts. He Jianbin from Xinjiang University proposed an alternating path to address the large height difference between the arc-extinguishing and arc-starting ends in his research on the process of arc additive manufacturing of 5356 aluminum alloy. Yin Fan from Nanjing University of Science and Technology proposed a fixed-distance compensation method to ensure forming in his research on multi-layer deposition forming process and dimensional control. Xiao Yuchen from Yanshan University derived a double-layer parallel reciprocating welding method to solve the collapse problem in his research on thin-walled wall welding. In summary, among the solutions proposed by many representative scholars, it can be concluded that most scholars adopt a parallel reciprocating deposition method or an improved method based on parallel reciprocating, such as a double-layer parallel reciprocating welding method, to address the problem of arc-extinguishing collapse in single-pass multi-layer welded parts. However, existing solutions, whether parallel reciprocating or improved based on parallel reciprocating, not only fail to accurately fill defects but also often exhibit a phenomenon where the middle is higher and the two ends are lower as the number of deposition layers increases. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for compensating for arc-extinguishing collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing. Based on a neural network for process parameter prediction, this method compensates for collapse by welding back a certain distance from the extinguishing point without extinguishing the arc before extinguishing it at each layer. This effectively solves the problem of arc-extinguishing collapse during single-layer multi-pass forming, ensuring the high flatness of the finished part in arc additive manufacturing, and thus resolving the issue of part appearance being affected by arc-extinguishing collapse.

[0006] To achieve the above objectives, the present invention can adopt the following technical solutions:

[0007] A method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing, comprising the following steps:

[0008] Design the corresponding welding process based on the welding materials, structural form, and welding method;

[0009] Based on the welding process design, corresponding process experiments are conducted, and the process experiments are carried out in batches to obtain process experiment data; wherein, the process experiment data includes welding current, welding voltage, welding speed, weld formation size, and weld collapse compensation distance at the arc extinguishing point;

[0010] A dataset for predicting the collapse compensation distance at the weld arc extinguishing point is established based on the data combination method of the welding current, the welding voltage, the welding speed, the weld formation size, and the collapse compensation distance at the weld arc extinguishing point;

[0011] A deep neural network is constructed with the welding current, welding voltage, and welding speed as inputs, and the weld formation size and the weld arc extinguishing collapse compensation distance as outputs. The deep neural network is trained and evaluated by dividing the experimental dataset into training and validation sets to obtain a prediction model for the weld arc extinguishing collapse compensation distance.

[0012] The welding current, welding voltage, and welding speed obtained during the weld formation process are collected and input into the weld arc extinguishing collapse compensation distance prediction model. The corresponding weld formation size and weld arc extinguishing collapse compensation distance are output, thereby accurately predicting the weld arc extinguishing collapse compensation distance.

[0013] As described above, in the arc additive manufacturing single-pass multi-layer weld arc extinguishing collapse compensation method, the weld forming dimensions further include any one or any combination of the weld center reinforcement height, the weld minimum reinforcement height, and the compensation reinforcement height difference.

[0014] The above-described method for compensating for arc collapse at the arc extinguishing point of a single-pass multi-layer weld in arc additive manufacturing further includes the following steps for designing corresponding process experiments based on the welding process: on the basis of the corresponding materials and equipment, selecting a range of process parameters that can ensure good weld formation, and using orthogonal experimental design to design multi-variable, multi-level process experiments to minimize the number of process experiments and the training difficulty when building the neural network.

[0015] The above-described method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing further includes a deep neural network specifically a multi-input multi-output network, which consists of an input layer, a hidden layer, and an output layer. The number of neurons in the output layer is determined by the process parameters. The output layer includes the weld formation size and the arc collapse compensation distance at the extinguishing point. The number of layers and neurons in the hidden layer are jointly determined by the degree of nonlinearity, the accuracy of the formation prediction, and the computational efficiency of the network model weights and thresholds in the multi-input multi-output network composed of process parameters and weld formation results. The model structure and network parameters are optimized and adjusted according to the prediction error.

[0016] The above-described method for compensating for the collapse at the arc extinguishing point of a single-pass multi-layer weld in arc additive manufacturing further includes, in part, the step of optimizing and adjusting the model structure and network parameters, specifically comprising: optimizing the model structure and network parameters based on the size of the dataset, the network structure, the model calculation speed, and / or the accuracy of the prediction results.

[0017] The above-described method for compensating for the collapse at the arc extinguishing point of a single-pass multi-layer weld in arc additive manufacturing further includes the following steps for evaluating the neural network: comprehensively considering the size of the network structure, network training time, model computational efficiency, and forming prediction accuracy to evaluate the computational cost and efficiency of the weld forming prediction model, ensuring the accuracy and stability of the model. The model accuracy is determined by linear regression analysis of the predicted data and the actual data.

[0018] As described above, the method for compensating for arc collapse at the arc extinction point of a single-pass multi-layer weld in arc additive manufacturing further includes the following weld structure: thick-walled single-pass multi-layer stacking and / or thin-walled single-pass multi-layer stacking. The welding method includes any one or any combination of non-consumable electrode gas shielded welding (TIG), consumable electrode gas shielded welding (MIG / MAG), cold metal transfer welding (CMT), or plasma arc welding (PAW).

[0019] In the above-described method for compensating for the collapse at the arc-extinguishing point of a single-pass multi-layer weld in arc additive manufacturing, the collapse compensation distance at the arc-extinguishing point is further defined as the compensation distance that yields the optimal weld smoothness after compensation.

[0020] Compared with existing technologies, the advantages of this invention are as follows: The collapse compensation method of this invention is based on building a neural network for process parameter prediction. During each layer of welding, collapse compensation is achieved by welding back a certain distance at the arc-extinguishing point without extinguishing the arc before extinguishing the arc. This method can effectively solve the problem of arc-extinguishing point collapse that occurs in single-layer multi-pass forming processes, thereby ensuring the high flatness of the finished part in arc additive manufacturing, and thus solving the problem of the part's appearance being affected by arc-extinguishing point collapse. In addition, compared with existing solutions such as parallel reciprocating welding, the method of this invention also has advantages such as low cost, convenient operation, and accurate collapse compensation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0022] Figure 1 This is a flowchart illustrating the steps involved in building a weld arc extinguishing point collapse compensation distance prediction model according to an embodiment of the present invention.

[0023] Figure 2 This is a flowchart illustrating the operational logic of the weld arc extinguishing point collapse compensation distance prediction model according to an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the MIG weld prediction neural network structure for ER308L stainless steel welding wire according to an embodiment of the present invention.

[0025] Figure 4 These are experimental morphological photographs of the MIG weld seam of ER308L stainless steel welding wire according to an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of the prediction error results for MIG weld formation of ER308L stainless steel welding wire according to an embodiment of the present invention.

[0027] Figure 6 This is a comparative schematic diagram of the application results of MIG welds using ER308L stainless steel welding wire according to an embodiment of the present invention. (a) is a single-pass multi-layer weld component manufactured by the traditional stacking method; (b) is a single-pass multi-layer weld component manufactured using the method of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention 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. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0029] Example:

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, in the embodiments of this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] The word “exemplary” as used below means “serving as an example, embodiment, or illustration.” Any embodiment illustrated as an “exemplary” need not be construed as superior to or better than other embodiments.

[0032] See Figures 1 to 5This invention provides a method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing. The method mainly includes steps such as preparing welding plates and welding equipment, designing and implementing process experiments, preprocessing process parameters and experimental data, collecting and establishing a weld formation experimental dataset, building a neural network, and deploying a weld formation result parameter prediction model. Specifically, it may include the following steps:

[0033] Step 101: Design the corresponding welding process based on the welding materials, structural form, and welding method.

[0034] Specifically, the preparation of welding plates and welding equipment involves preparing suitable experimental materials such as welding plates, welding wire, welding shielding gas, welding robots, welding power sources, and water-cooling machines, based on the selected arc welding equipment, welding materials, structural form, welding method, and process parameter design scheme.

[0035] In some embodiments, the weld structure may include thick-walled single-pass multi-layer stacking, thin-walled single-pass multi-layer stacking, and other structural types, which can be supported by different materials and different base plate thicknesses depending on the type. The welding method, depending on the arc type, may include any one or any combination of non-consumable electrode gas shielded welding (TIG), consumable electrode gas shielded welding (MIG / MAG), cold metal transfer welding (CMT), and plasma arc welding (PAW).

[0036] Step 102: Design corresponding process experiments according to the welding process, and implement the process experiments in batches to obtain process experiment data; wherein, the process experiment data includes welding current, welding voltage, welding speed, weld formation size and weld arc extinguishing collapse compensation distance.

[0037] Specifically, the process experiment design and implementation involves designing a single-pass multi-layer weld deposit process experiment based on experimental conditions such as welding power supply load performance, welding plate thickness, welding wire material and diameter, and conducting batch experiments using the selected arc additive manufacturing equipment.

[0038] Process parameter preprocessing involves preprocessing the experimental data into structured data required for the weld formation morphology dataset. The process parameter data involved in the experiment includes welding current, welding voltage, welding speed, weld formation size, and weld collapse compensation distance at the arc extinguishing point. For example, process parameters may include welding current, welding voltage, welding speed, etc., and experimental result data may include weld formation size and weld collapse compensation distance at the arc extinguishing point.

[0039] In some embodiments, the weld formation dimensions include the weld mid-section reinforcement height, the weld minimum reinforcement height, and the compensation reinforcement height difference. The weld arc-extinguishing collapse compensation distance is the compensation distance that yields the optimal weld smoothness after compensation.

[0040] In other embodiments, the specific steps of designing corresponding process experiments according to the welding process include: based on the corresponding materials and equipment, within the range of process parameters that ensure good weld formation, using orthogonal experimental design to design multivariate, multi-level, large-scale additive manufacturing process experiments, minimizing the number of process experiments and the training difficulty of the neural network while ensuring experimental results, and maximizing the accuracy and efficiency of the neural network model for predicting the collapse compensation distance at the weld arc extinguishing point while avoiding the situation of non-fitting or overfitting of the neural network due to insufficient dataset.

[0041] Step 103: Establish a prediction dataset for the weld arc extinguishing collapse compensation distance based on the data combination method of the welding current, the welding voltage, the welding speed, the weld formation size, and the weld arc extinguishing collapse compensation distance.

[0042] Specifically, the experimental dataset for weld formation is collected and established. The collected process parameters and experimental results data are combined in a way that includes welding current, welding voltage, welding speed, weld formation size, and weld arc extinguishing collapse compensation distance to create a predicted dataset for weld arc extinguishing collapse compensation distance.

[0043] Step 104: Construct a deep neural network with the welding current, welding voltage, and welding speed as inputs, and the weld formation size and the weld arc extinguishing collapse compensation distance as outputs. Use the experimental dataset to divide the training set and validation set to train and evaluate the deep neural network, and obtain the weld arc extinguishing collapse compensation distance prediction model.

[0044] Specifically, the neural network is built by designing and implementing functions such as building the neural network structure, adjusting the neural network training parameters, evaluating the network training efficiency and model prediction efficiency, and optimizing the network structure and training parameters. The network structure and training parameters are adjusted according to the error of the prediction results to obtain the prediction model of the collapse compensation distance at the arc extinguishing point of the weld.

[0045] In some embodiments, the deep neural network is specifically a multiple-input multiple-output network, which consists of an input layer, a hidden layer, and an output layer. The number of neurons in the output layer is determined by the process parameters. The output layer includes the weld formation size and the collapse compensation distance at the weld arc extinguishing point. The number of layers and neurons in the hidden layer are jointly determined by the degree of nonlinearity in the multiple-input multiple-output network composed of process parameters and weld formation results, the accuracy of forming prediction, and the computational efficiency of network model weights and thresholds. The model structure and network parameters are optimized and adjusted according to the prediction error.

[0046] In some embodiments, the step of optimizing and adjusting the model structure and network parameters specifically includes: optimizing the model structure and network parameters based on the size of the dataset, the network structure, the model computation speed, and / or the accuracy of the prediction results. The main purpose of optimizing and adjusting the model structure and network parameters is to adjust them based on the dataset size, prediction error, and computational efficiency, thereby achieving comprehensive control over the feedback and adjustment of the neural network-based prediction model parameters.

[0047] In other embodiments, evaluating the neural network specifically includes: evaluating the size of the neural network structure, evaluating the training time when building the neural network, evaluating the computational efficiency of the neural network model, and evaluating the accuracy of the neural network's prediction results. The computational overhead and efficiency of the arc extinction compensation distance prediction model are comprehensively evaluated to ensure the accuracy and stability of the neural network prediction model. The accuracy of the network model is determined by linear regression analysis of the prediction result data and experimental test data.

[0048] Step 105: Collect the welding current, welding voltage, and welding speed obtained during the weld formation process, and input the above process parameters into the weld arc extinguishing collapse compensation distance prediction model, and output the corresponding weld formation size and weld arc extinguishing collapse compensation distance, thereby accurately predicting the weld arc extinguishing collapse compensation distance.

[0049] Specifically, the deployment of the weld formation result parameter prediction model involves deploying the weld arc extinguishing collapse compensation distance prediction model onto a computer platform, designing a human-computer interaction interface and ports. The input ports include welding equipment, material type, structural form, welding method, and process parameters, while the output ports include weld formation size and weld arc extinguishing collapse compensation distance. An expansion port is also reserved.

[0050] To better understand this invention, the operating logic of the weld arc extinguishing point collapse compensation distance prediction model of this invention will be explained below.

[0051] See Figure 2 First, multiple sets of corresponding data for process parameters and weld formation data are read from the welding parameter and forming dataset. Second, the neural network structure and training parameters are preset. Third, based on the welding process parameters and weld formation data, the preset neural network structure and training parameters are placed in the neural network model for predicting the collapse compensation distance at the weld arc extinguishing point for training. Next, the performance of the neural network is evaluated. If the evaluation fails, the network structure and network training parameters are optimized and adjusted and fed back to the preset stage of network structure and training parameters. The neural network is trained again until the neural network model passes the performance evaluation. Finally, after passing the evaluation, the neural network model for predicting the collapse compensation distance at the weld arc extinguishing point is deployed.

[0052] To better understand the present invention, the following specific embodiments further illustrate the method for compensating for arc collapse at the arc extinguishing point of a single-pass multi-layer weld in arc additive manufacturing.

[0053] This invention provides a neural network-based system for predicting the collapse compensation distance at the arc extinction point of MIG welds using ER308L stainless steel welding wire. The welding equipment is a TM-1400GⅢ, the arc welding robot is a Panasonic robot-AUR01062, the teach pendant system is a Windows CE system, the welding power supply is a YD-500GS, the wire is ER308L, and the test plate size is 400mm×200mm×10mm.

[0054] The structure is a single-pass multi-layer stacked plate, the welding method is front wire flat welding, the welding torch is perpendicular to the test plate, the arc type is gas metal inert gas (MIG) welding, the base plate is Q345 low carbon steel, the welding wire is ER308L stainless steel welding wire with a diameter of 1.2mm, the arc voltage is 22.5V, the welding speed is 0.5~0.8m / min, the welding current range is 160~220A, the welding wire extension is 15mm, the shielding gas is a mixture of (98% Ar + 2% CO2) gas with a flow rate of 20L / min.

[0055] Table 1 shows the dimensional data of MIG weld formation using ER308L stainless steel welding wire, and the corresponding weld morphology photographs are shown below. Figure 4 As shown.

[0056] Table 1. MIG weld formation dimensions of ER308L stainless steel welding wire

[0057]

[0058] The neural network prediction model is a three-input, two-output network. The inputs are welding voltage, welding current, and welding speed; the outputs are weld formation dimensions and weld collapse compensation distance at the arc extinguishing point. Due to the significant differences in the value ranges of different features in the samples, the iteration speed of the neural network may be severely affected, and outliers may appear. To reduce the influence of feature values ​​and ensure the performance of the neural network, a maximum-minimum normalization method is used to process the feature data. Specifically, for each feature value x, after calculating the mean (x), maximum (x), and minimum (x), a linear transformation, i.e., normalization, is performed.

[0059]

[0060] Normalization reduces the training difficulty of neural network models by processing data into the range [0, 1], while ensuring their iteration speed and preventing overfitting.

[0061] See Figure 3 , Figure 3 A neural network structure for predicting MIG weld seams using ER308L stainless steel welding wire, specifically for predicting the collapse compensation distance at the weld seam extinguishing point, is constructed. This structure consists of an input layer, hidden layer I, hidden layer II, and an output layer. The input layer has 3 neurons, the hidden layer has 11 neurons, and the output layer has 2 neurons. The final training parameters for the neural network are: batch_size = 6, epochs = 2000, training set size = 50, test set size = 6, and learning rate = 0.1. To more significantly measure the prediction accuracy of this neural network, the mean squared error (MSE) and regression value R are used as the standards for evaluating the accuracy of the neural network prediction model, which can be expressed as:

[0062]

[0063]

[0064] The prediction error results of the MIG weld prediction neural network for ER308L stainless steel welding wire are as follows: Figure 5 As shown, the neural network for predicting the collapse compensation distance at the weld extinguishing point, which is to be deployed in the final stage, belongs to a multi-input multi-output highly nonlinear function relationship from the welding voltage U, welding current I, and welding speed V to the weld reinforcement difference and weld compensation distance in the weld forming size, which can be denoted as:

[0065]

[0066] To evaluate the effectiveness and accuracy of the neural network prediction model, a large number of data points were generated densely within the reasonable range of three process parameters: welding current I, welding voltage U, and welding speed V. After normalization, the data were compared with experimental data. Then, predictions were made, and the correlation and error between the measured and predicted values ​​were analyzed.

[0067] Importing the data into the neural network prediction model and running the trained model yields a mean squared error (MSE) of 0.0052606. The correlation coefficients for each dataset are shown below. Figure 5 (a) is a schematic diagram of neural network regression fitting. It can be seen from the figure that the regression R value of all datasets is greater than 0.9 and close to 1, which indicates that the accuracy and effect of the established model prediction are excellent. Figure 5(b) is a comparison chart of the neural network prediction results and actual values. Predicted value 1 and actual value 1 refer to the predicted and actual values ​​of the weld reinforcement difference, with a maximum error of <= 0.2 mm. Predicted value 2 and actual value 2 refer to the predicted and actual values ​​of the weld arc extinguishing compensation distance, with a maximum error of <= 0.2 mm. It can be seen that the actual and predicted values ​​of the two output parameters match very well. Therefore, the error assessment of this neural network prediction model is qualified, and it can be used.

[0068] See Figure 6 , Figure 6 This is a schematic diagram comparing the application results of MIG welds using ER308L stainless steel welding wire in an embodiment of the present invention. The compensation distance and welding process parameters were predicted using a trained BP neural network prediction model. The program instructions used were A = [15; 150; 22.5; 0.5] and y = sim(net0, A), where net0 is the trained neural network. The results obtained were 0.1899 and 0.3088, respectively. The predicted values ​​were relatively small, so based on the prediction results, the back-welding distance was set to 15mm, and the current during back-welding was adjusted to 150A. The welding voltage and welding speed remained unchanged and consistent with the previous parameters. A single-pass multi-layer straight wall forming experiment was conducted. During the forming of the same layer, the welding current was reduced to 150A and then extinguished at the arc extinguishing point while welding back 15mm. The interlayer temperature was controlled at around 150 degrees Celsius. A comparison diagram of the experimental results using the compensation method of the present invention and the results of traditional stacking experiments is shown below. Figure 6 As shown, Figure 6 (a) shows the forming result using the traditional stacking forming method. Figure 6 (b) shows the forming result achieved using the collapse compensation method. From Figure 6 It can be seen that the surface of the molded part obtained by the collapse compensation method is relatively flat, and there is no obvious collapse at both ends, which is a great improvement over the molding result of the traditional method.

[0069] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0070] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing, characterized in that: Includes the following steps: The corresponding welding process is designed based on the welding materials, structural form, and welding method; the weld structure includes: thick-walled single-pass multi-layer stacking and / or thin-walled single-pass multi-layer stacking; the welding method includes any one or any combination of non-consumable electrode gas shielded welding (TIG), consumable electrode gas shielded welding (MIG / MAG), cold metal transfer welding (CMT), or plasma arc welding (PAW). According to the welding process design, corresponding process experiments are conducted, and the process experiments are carried out in batches to obtain process experiment data; wherein, the process experiment data includes welding current, welding voltage, welding speed, weld formation size and weld collapse compensation distance at the arc extinguishing point; the weld formation size includes any one or any combination of the weld middle reinforcement height, weld lowest point reinforcement height and compensation reinforcement height difference; A dataset for predicting the collapse compensation distance at the weld arc extinguishing point is established based on the data combination method of the welding current, the welding voltage, the welding speed, the weld formation size, and the collapse compensation distance at the weld arc extinguishing point; A deep neural network is constructed, taking the welding current, welding voltage, and welding speed as inputs, and the weld formation size and the weld arc extinguishing collapse compensation distance as outputs. The deep neural network is trained and evaluated using an experimental dataset divided into training and validation sets. The model structure and network parameters are optimized and adjusted based on the prediction error to obtain a prediction model for the weld arc extinguishing collapse compensation distance. Specifically, the steps for optimizing and adjusting the model structure and network parameters include: optimizing the model structure and network parameters based on the dataset size, network structure, model computation speed, and / or prediction result accuracy. The welding current, welding voltage, and welding speed obtained during the weld formation process are collected and input into the weld arc extinguishing collapse compensation distance prediction model. The corresponding weld formation size and weld arc extinguishing collapse compensation distance are output. The weld arc extinguishing collapse compensation distance is the compensation distance with the optimal weld flatness after compensation. During each layer of welding, collapse compensation is performed by welding back a certain distance without extinguishing the arc at the arc extinguishing point before extinguishing the arc.

2. The method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing according to claim 1, characterized in that: The specific steps for designing the corresponding process experiments based on the welding process include: selecting a range of process parameters that can ensure good weld formation based on the corresponding materials and equipment, and designing multi-variable and multi-level process experiments using orthogonal experimental design to minimize the number of process experiments and the training difficulty when building the neural network.

3. The method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing according to claim 1, characterized in that: The deep neural network is specifically a multiple-input multiple-output network, which consists of an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is determined by the process parameters. The output layer includes the weld formation size and the collapse compensation distance at the weld arc extinguishing point. The number of layers and neurons in the hidden layer are jointly determined by the degree of nonlinearity, the accuracy of forming prediction, and the computational efficiency of the network model weights and thresholds in the multiple-input multiple-output network composed of process parameters and weld formation results.

4. The method for compensating for arc collapse at the extinguishing point of a single-pass multi-layer weld in arc additive manufacturing according to claim 1, characterized in that: The evaluation steps for the neural network specifically include: comprehensively considering the size of the network structure, network training time, model computational efficiency, and forming prediction accuracy to evaluate the computational cost and efficiency of the weld forming prediction model, ensuring the accuracy and stability of the model. The model accuracy is determined by linear regression analysis of the predicted data and the actual data.

Images