Temperature field coupled steel rail three-wire molten nozzle electroslag welding shape control method and system

Abstract

The present application belongs to the field of welding forming, and discloses a temperature field coupled steel rail three-wire welding nozzle electric slag welding shape control method and system, which comprises the following steps: placing the to-be-welded steel rails in the welding position, the to-be-welded steel rails having a welding gap therebetween, the welding gap being capable of accommodating three electric slag welding nozzles, each nozzle having a welding wire therein; placing two water-cooled copper molds on the two sides of the to-be-welded steel rails; designing welding process parameters based on the welding gap, starting three-wire electric slag welding, and acquiring the temperature field in the welding steel rail molten pool in real time, and adjusting the welding process parameters in real time based on the temperature field; the welding process parameters at least include the flow of the water-cooled copper mold. The present application can improve the quality of the steel rail welding joint and ensure near-net forming of the steel rail weld.

Description

Technical Field

[0001] This invention belongs to the field of welding forming, and more specifically, relates to a method and system for controlling the electroslag weldability of a three-wire welding nozzle for rails with temperature field coupling. Background Technology

[0002] To reduce the impact of trains on the tracks, welding is used to construct seamless railway lines, making welded joints a crucial component of high-speed seamless tracks. To ensure train safety, the construction of high-speed, heavy-haul, and complex geological seamless tracks places extremely high demands on the performance of welded joints: the tensile strength of rail welded joints must be no less than 980 MPa, and the microstructure must be free of martensite and bainite. To meet or exceed these performance indicators, the grain size of rail welded joints must be less than 22 μm, the microstructure must consist of more than 98% pearlite and less than 2% ferrite, and the interlamellar spacing of the pearlite fine structure must be less than 0.1 μm.

[0003] Existing rail welding methods include aluminothermic welding, flash welding, gas pressure welding, and electric arc welding. Aluminothermic welding is mainly used for rail turnout welding; it is simple to operate, but slag inclusions are common in the weld, and the joint grains are coarse, making it prone to martensite formation. Flash welding and gas pressure welding are mainly used for rail production in workshops and on-site welding; their weld grains are fine, but the heat-affected zone has larger grains, sometimes exhibiting martensite and bainite. Electric arc welding is mainly used for rail repair; its heat-affected zone has coarse grains, making it highly susceptible to martensite formation and microcracks. Because existing rail welding methods easily produce martensite and bainite in the microstructure of the weld joints, with coarse grains, and because the weld joint formation is difficult to control, it is difficult to obtain high-quality rail weld joints.

[0004] Therefore, there is an urgent need for a method for welding and shape control of rails to improve the quality of rail welded joints and ensure near-net-shape rail welds. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for controlling the weld shape of rail three-wire welding nozzles using temperature field coupling. The purpose is to improve the quality of rail welded joints and ensure near-net-shape rail welds.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for controlling the electroslag weldability of a three-wire welding nozzle for rails coupled with a temperature field is proposed, comprising the following steps:

[0007] The rails to be welded are placed in the welding position, and there is a welding gap between the rails to be welded. The welding gap can accommodate three electroslag welding nozzles, and welding wire passes through each nozzle. The water-cooled copper mold is placed on both sides of the rail to be welded position.

[0008] Based on the welding gap, the welding process parameters are designed, and three-wire electroslag welding is started. The temperature field in the weld pool of the welded rail is acquired in real time, and the welding process parameters are adjusted in real time based on the temperature field. The welding process parameters include at least the flow rate of the water-cooled copper mold.

[0009] As a further preferred option, the welding process is divided into the following three welding stages:

[0010] Phase I, arc initiation and slag formation phase; Phase II, normal welding phase; Phase III, lead-out phase.

[0011] As a further preferred embodiment, the welding process parameters designed based on the welding gap include at least:

[0012] Welding parameter specifications are determined based on the welding gap;

[0013] A reference rail gap is set, and the welding process parameters are corrected once based on the difference between the welding gap and the reference rail gap.

[0014] As a further preferred embodiment, the real-time adjustment of welding process parameters based on the temperature field includes at least:

[0015] Obtain the trained second and third neural networks;

[0016] The welding process parameters are corrected in real time based on the second neural network and the third neural network.

[0017] The second neural network and the third neural network correspond to the process parameter feedforward prediction networks for welding stage II and stage III, respectively. The process parameter feedforward prediction networks are used to generate corresponding welding process parameters based on the acquired temperature field.

[0018] As a further preferred embodiment, when the temperature field is a temperature field image, the second neural network and the third neural network are configured as CNN convolutional neural networks.

[0019] As a further preferred embodiment, the second neural network and the third neural network are jointly trained according to the following steps:

[0020] Obtain an initial model of a CNN (Convolutional Neural Network), which includes a first part and a second part. The first part of the initial model includes at least several convolutional layers and two fully connected layers, and the second part of the initial model includes at least several convolutional layers and one fully connected layer. The first part and the second part of the initial model are connected by a fully connected layer.

[0021] Obtain a first training sample set and a second training sample set; the first training sample set includes multiple temperature field images corresponding to welding stage II, corresponding welding parameter features and corresponding temperature field feature labels, and the second training sample set includes multiple temperature field images corresponding to welding stage III, corresponding welding parameter features and corresponding temperature field feature labels.

[0022] The first part of the initial model is trained based on the first training sample set, and the second part of the initial model is trained based on the second training sample set in multiple rounds to obtain the pre-trained first and second parts of the initial model.

[0023] Obtain a joint training sample set, which is a training set that screens out temperature gradients that meet the screening requirements based on the first training sample set and the second training sample set;

[0024] Based on the joint training sample set, the first and second parts of the pre-trained initial model are jointly trained to obtain a trained CNN convolutional neural network.

[0025] As a further preferred embodiment, the welding process parameters include the water flow rate of the water-cooled copper mold, the water flow direction, the water flow pulse speed, and the process parameters of the three welding wires.

[0026] As a further preferred embodiment, the three electroslag welding nozzles include one main nozzle and two auxiliary nozzles, with the two auxiliary nozzles distributed on both sides of the main nozzle; the rail to be welded is in the shape of an "I", which includes a rail head, a rail web and a rail bottom, the two auxiliary nozzles are used to weld the rail bottom, and the main nozzle is used to weld the rail web and the rail head.

[0027] As a further preferred embodiment, the water-cooled copper mold also includes a rail bottom copper mold, which is disposed at the bottom of the rail to be welded to quickly establish a stable welding pool.

[0028] According to another aspect of the present invention, a temperature field coupled rail three-wire melting nozzle electroslag weldability control system is provided, comprising:

[0029] The welding preparation module is used to place the rails to be welded in the welding position. There is a welding gap between the rails to be welded. The welding gap can accommodate three electroslag welding nozzles, and welding wire passes through each nozzle. Two water-cooled copper molds are placed on both sides of the two rails to be welded.

[0030] Welding start-up module, used for three-wire electroslag welding;

[0031] Temperature field acquisition module is used to acquire the temperature field in the weld pool of the welded rail in real time;

[0032] The welding adjustment module is used to design welding process parameters based on the welding gap and to adjust the welding process parameters in real time based on the temperature field; the welding process parameters include at least the flow rate of the water-cooled copper mold.

[0033] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0034] 1. Rail welding is performed using electroslag welding, combined with water-cooled copper molds for forced forming. Electroslag welding has a large heat input and a slow cooling rate, which can avoid the formation of martensite and bainite in the weld joint when welding rails. At the same time, the welding parameters are initially designed based on the welding gap, avoiding the labor costs of a large number of experiments. Furthermore, during welding, the welding process parameters can be adjusted in real time based on the temperature field of the molten pool, which can improve the quality of the rail weld joint and ensure near-net-shape rail weld.

[0035] 2. To ensure the stability of the welding process and to ensure that the accumulated metal fills the welding gap according to the predetermined welding path and oscillation path, the process parameters of the three welding wires can only be finely adjusted. This invention uses a water-cooled copper mold for forced cooling and combines temperature images with neural networks to precisely control the corresponding parameters of the water-cooled copper mold, thereby adjusting the cooling effect of the water-cooled copper mold during the welding process, reducing the welding temperature of the molten pool, so as to better control the formation of the molten pool and avoid the risk of burn-through or insufficient welding.

[0036] 3. Since different welding processes correspond to different neural networks, joint training is adopted to improve the stability of the boundaries of multiple models, avoid parameter instability at the junction of welding processes, and improve welding quality. Attached Figure Description

[0037] Figure 1 This is a system block diagram of an exemplary temperature field coupled rail three-wire melting nozzle electroslag weldability control system according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic flowchart illustrating an exemplary temperature field coupled three-wire electroslag weld shape control method according to an embodiment of the present invention.

[0039] Figure 3 This is a schematic flowchart illustrating the application of a CNN convolutional neural network in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the welding process shown in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram of the neural network training process shown in an embodiment of the present invention;

[0042] Figure 6This is a schematic flowchart illustrating the training process of a CNN convolutional neural network according to an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0044] Flowcharts are used in this embodiment of the invention to illustrate the operations performed by the system according to the embodiments of the invention. It should be understood that the preceding or following operations are not necessarily executed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0045] Figure 1 This is a system block diagram of an exemplary temperature field coupled three-wire melting nozzle electroslag weld shape control system for rails.

[0046] like Figure 1 As shown, the system 100 may include: a welding preparation module 110, a welding start module 120, a temperature field acquisition module 130, and a welding adjustment module 140, wherein:

[0047] The welding preparation module 110 is used to place two workpieces to be welded in the welding position. There is a welding gap between the two workpieces to be welded. The welding gap contains at least three electroslag welding nozzles and welding wires. The nozzles are tubular and the welding wires pass through the tubes. The three electroslag welding nozzles include one main nozzle and two auxiliary nozzles. The two auxiliary nozzles are respectively distributed on both sides of the main nozzle. Two water-cooled copper molds are respectively placed on the back side of the two workpieces to be welded.

[0048] The welding start-up module 120 is used for three-wire electroslag welding. The welding process is divided into three stages: Stage I, the arc ignition and slag formation stage, in which the welding wires of the two auxiliary welding nozzles and the welding wire of the main welding nozzle are simultaneously ignited to form a stable slag pool; Stage II, the normal welding stage, in which the arcs of the two auxiliary wires and the main wire are extinguished, and the process transitions to the electroslag welding stage; Stage III, the lead-out stage, in which the welding nozzle continues to weld with a small current and a small heat input when the welding of the rail head is close to completion to ensure that the rail head has a good shape.

[0049] Temperature field acquisition module 130 is used to acquire the temperature field in the molten pool of the workpiece to be welded in real time.

[0050] The welding adjustment module 140 is used to adjust the welding process parameters in real time based on the temperature field of the molten pool.

[0051] In some embodiments, the temperature field acquisition module 130 is used to acquire a temperature field image. At this time, the temperature field acquisition module 130 should also acquire a trained second neural network and a third neural network. The second neural network and the third neural network correspond to the process parameter feedforward prediction networks of welding stage II and welding stage III, respectively. The process parameter feedforward prediction networks are used to generate corresponding water-cooled copper mold parameters based on the acquired molten pool temperature field image.

[0052] In some embodiments, when the temperature field is a temperature field image, the second neural network and the third neural network are configured as CNN convolutional neural networks.

[0053] In some embodiments, when the second neural network and the third neural network are CNN convolutional neural networks, the second neural network and the third neural network are jointly trained according to the following steps: obtaining an initial model of the convolutional neural network; the initial model of the convolutional neural network includes a first part and a second part, wherein the first part of the initial model includes at least several convolutional layers and two fully connected layers, and the second part of the initial model includes at least several convolutional layers and one fully connected layer; the first part and the second part of the initial model are connected by the fully connected layer; obtaining a first training sample set and a second training sample set; the first training sample set includes multiple temperature field images corresponding to the welding stage II, and the second training sample set includes multiple temperature field images corresponding to the welding stage III; training the first part of the initial model based on the first training sample set and training the second part of the initial model based on the second training sample set for multiple rounds to obtain pre-trained initial model first and second parts; obtaining a joint training sample set, wherein the joint training sample set is used to screen out training sets whose temperature gradients meet the screening requirements based on the first training sample set and the second training sample set; and jointly training the first part and the second part of the pre-trained initial model based on the joint training sample set to obtain a trained CNN convolutional neural network.

[0054] In some embodiments, the parameters of the water-cooled copper mold include, but are not limited to: water flow rate, water flow direction, and water flow pulse speed.

[0055] In some embodiments, the welding adjustment module 140 is also used to determine the welding gap between the rails and to determine the welding parameter specifications based on the welding gap.

[0056] In some embodiments, the welding adjustment module 140 is further configured to set a reference rail gap and correct the welding process parameters based on the difference between the welding gap between the rails and the reference rail gap.

[0057] In some embodiments, the rail is in the shape of an "I" and includes a rail head, a rail web, and a rail base. The two auxiliary wires are used to weld the rail base, and the main welding wire is used to weld the rail head and rail web.

[0058] In some embodiments, the water-cooled copper mold further includes a rail-bottom copper mold, which is used to quickly establish a stable weld pool.

[0059] The systems and modules thereof in one or more embodiments of the present invention can be implemented in various ways. For example, in some embodiments, the systems and modules thereof can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the methods and systems described above can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The systems and modules thereof of the present invention can be implemented not only by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., but also by software, for example, executed by various types of processors, or by a combination of the aforementioned hardware circuits and software (e.g., firmware).

[0060] It should be noted that the above description of the processing device and its modules is for ease of description only and should not be construed as limiting the invention to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principle of the system, may arbitrarily combine the various modules or construct subsystems connected to other modules without departing from this principle.

[0061] Figure 2 This is a schematic flowchart of an exemplary temperature field coupled three-wire electroslag weld form control method. In some embodiments, this method 200 may be further executed by system 100.

[0062] Step 210: Place the two workpieces to be welded in the welding position.

[0063] In some embodiments, step 210 may be performed by welding preparation module 110.

[0064] In some implementation scenarios, this technology can be used for welding rails in railways, bullet trains, high-speed trains, and other similar applications. In some embodiments, this technology can also be used for welding large, thick parts (such as large boilers, large trusses, and steel structures).

[0065] The workpieces to be welded can be two long sections of welding rails that have already been placed and ground smooth before being placed at the welding position. In some embodiments, the workpieces to be welded need to undergo certain pre-welding preparation work before welding to ensure welding quality. In some embodiments, pre-welding preparation work includes, but is not limited to: pre-welding oxide scale removal, pickling treatment, and laying of welding lining.

[0066] After the pre-welding preparations are completed, the two workpieces to be welded are placed in the welding position. The welding position refers to the relative distance between the two workpieces and the welding location. In some embodiments, the two workpieces are placed horizontally. In some embodiments, the two workpieces may also be at any angle.

[0067] Furthermore, the gap to be welded contains at least three electroslag welding wires, the three electroslag welding wires including one main wire and two auxiliary wires, the two auxiliary wires being distributed on both sides of the main wire, and the two water-cooled copper molds being placed on the back side of the two rails to be welded.

[0068] In some embodiments, the welding preparation module 110 can be clamped and placed using fixed tooling, such as welding assembly fixtures or welding fixing fixtures. In some embodiments, the welding preparation module 110 can also be a freely movable tool, such as a 4-axis, 5-axis, or 6-axis robot. In some embodiments, the workpiece to be welded can be placed manually, guided and positioned by the welding preparation module. Such methods are still within the scope of protection of this invention and will not be elaborated further here.

[0069] In some embodiments, the workpiece to be welded is a steel rail, which is I-shaped and includes a rail head, a rail web, and a rail base. Two auxiliary wires are used to weld the rail base, and the main welding wire is used to weld the rail head and rail web. In this embodiment, the water-cooled copper mold also includes a rail base copper mold, which is used to quickly establish a stable weld pool.

[0070] Step 220: Perform three-wire electroslag welding.

[0071] In some embodiments, step 220 may be performed by welding initiation module 120.

[0072] In some embodiments, the printed part digital model processing module 120 can directly perform the three-wire electroslag welding based on a fixed process specification. In some embodiments, the printed part digital model processing module 120 can perform the three-wire electroslag welding based on preset parameters. The welding process is divided into the following three welding stages, as follows: Figure 4 As shown, specifically, the process includes the following steps: Step 410, the welding wires of the two auxiliary welding nozzles and the welding wire of the main welding nozzle simultaneously ignite, forming a stable slag pool; Step 420, the arcs of the two auxiliary wires and the main wire extinguish, transitioning to the electroslag welding stage; Step 430, when the welding of the rail head is nearing completion, the welding nozzle continues welding with a small current and low heat input to ensure good rail head formation. In steps 420 and 430, the second neural network and the third neural network are applied respectively. The second and third neural networks are process parameter feedforward prediction networks used to generate corresponding welding process parameters based on the acquired molten pool temperature field image. Specifically, the welding process parameters include, but are not limited to: water flow rate, water flow direction, water flow pulse velocity, and the process parameter specifications of the three welding wires.

[0073] In some embodiments, the digital model processing module 120 for printing parts can slice the digital model corresponding to the weld spacing, obtain the shape contour of the current weld layer corresponding to each slice from bottom to top, and then obtain the corresponding welding process parameters based on the shape contour of the current weld layer.

[0074] In this scenario embodiment, the model corresponding to the weld can be modeled and drawn using 3D drawing software such as UG, SOLIDWORKS, and CATIA. In some embodiments, the digital model retains modeling driving information, which may be the angle information, distance information, etc., between the two workpieces to be welded.

[0075] Step 230: Real-time acquisition of the temperature field within the molten pool of the workpiece to be welded.

[0076] In some embodiments, step 230 may be performed by the temperature field acquisition module 130.

[0077] The temperature field acquisition module 130 can acquire the temperature field on the workpiece to be welded based on one or more thermistors or thermocouples coupled to the workpiece. It is understood that when the required data is the temperature field of the entire surface of the workpiece to be welded, using multiple thermistors or thermocouples is preferred. In some embodiments, the temperature field acquisition module 130 can also use an infrared temperature sensor to acquire the temperature field within the molten pool of the workpiece to be welded in real time.

[0078] Step 240: Adjust the welding process parameters in real time based on the molten pool temperature field.

[0079] In some embodiments, step 240 can be performed by the welding adjustment module 140. The welding adjustment module 140 can determine the corresponding welding parameter specifications based on the rail gap obtained in step 210.

[0080] In some implementation scenarios, the welding gap between the rails is a fixed width that is uniform from top to bottom. In an embodiment of this scenario, the welding preparation module 110 can retrieve the corresponding reference welding specification parameters from a pre-stored database based on the welding gap. For example, a 100mm gap corresponds to reference welding specification 1, and a 200mm gap corresponds to reference welding specification 2.

[0081] In some implementation scenarios, the parameters corresponding to the welding gap do not fall within the parameters corresponding to the benchmark welding specification. In this case, the welding process parameters can be corrected based on the difference between the welding gap between the rails and the benchmark rail gap. For example, if the welding gap is 140mm, different weights can be obtained based on the difference between the welding gap and welding specification 1 and welding specification 2, and finally, the welding parameters corresponding to the current welding gap (i.e., the corrected welding parameters) can be obtained by using (0.6*welding parameter 1) + (0.4*welding parameter 2).

[0082] In some embodiments, the rails are arranged in parallel, but the welding gap is not uniform from top to bottom. For example, the cross-section between the rails presents a trapezoidal cross-section that is thicker at one end and thinner at the other. In an embodiment of this scenario, the method further includes obtaining the width of the head and tail of the trapezoid. For example, if the head of the trapezoid is 100mm and the tail is 200mm, and if the 100mm gap corresponds to welding specification 1 and the 200mm gap corresponds to welding specification 2, then the parameters on both sides are set according to these two welding specifications. Further, for the parameters of the middle part of the trapezoid, the corresponding welding specifications can be obtained point by point based on interpolation. For example, for a point A in the middle of the trapezoid with a gap of 130mm, different weights are obtained based on the difference between the current layer gap and welding specification 1 and welding specification 2, and finally (0.7*welding parameter 1) + (0.3*welding parameter 2) is used to obtain the welding parameters of the current layer. In some alternative embodiments, a neural network can also be used to obtain the corresponding weight coefficients.

[0083] It is understandable that welding thick plates requires a long welding time, and due to the large gaps between welds, the welding wire often needs to be oscillated. Typically, the oscillation speed of the welding wire is a fixed speed. Therefore, it is necessary to adaptively adjust the welding parameters to ensure that the workpiece can be welded without defects such as burn-through, overheating, or insufficient weld. Thus, the variation of welding parameters in this scenario should be a changing curve.

[0084] In some embodiments, the rails are not parallel, such as forming a 10° angle, meaning the weld seam has a shape that is either smaller at the bottom and larger at the top, or larger at the top and smaller at the bottom, from bottom to top. In other words, the weld width of the same layer is not uniform. In this scenario, it is necessary to slice the weld layer and obtain the shape contour of the current weld layer, and then obtain the weld gap of the current layer through the shape contour information. Different weld gaps are often difficult to exhaustively measure experimentally, so interpolation can be used to determine them. However, although the interpolation method can adjust the welding parameters, it introduces fluctuations in heat input, which poses risks to weld formation and welding quality. Therefore, one or more embodiments of the present invention involve a convolutional neural network that can adaptively correct parameters based on images captured by a thermal imager. Specifically, the welding adjustment module 140 can acquire trained second and third neural networks; and adjust the welding process parameters in real time based on the trained second and third neural networks. The second neural network and the third neural network correspond to the process parameter feedforward prediction networks for welding stage II and welding stage III, respectively. These process parameter feedforward prediction networks are used to generate the corresponding welding process parameters based on the acquired molten pool temperature field image. For more details on the process parameter feedforward prediction networks, please refer to [link to relevant documentation]. Figure 3 The corresponding explanations will not be repeated here.

[0085] Figure 3 This is a schematic flowchart of an exemplary CNN convolutional neural network application. In some embodiments, process 300 may be further executed by system 100. Specifically, process 300 may be executed by welding adjustment module 140.

[0086] Step 310: Obtain the trained second and third neural networks.

[0087] The training process for the second and third neural networks can be found in the subsequent explanations, and will not be repeated here.

[0088] In some embodiments, the second neural network and the third neural network correspond to the process parameter feedforward prediction networks for welding stage II and welding stage III, respectively. These process parameter feedforward prediction networks are used to generate corresponding welding process parameters based on the acquired molten pool temperature field image. Specifically, the welding process parameters include, but are not limited to: water flow rate, water flow direction, water flow pulse velocity, and the process parameter specifications for the three welding wires.

[0089] Step 320: Determine the welding process parameters after secondary correction in real time based on the second neural network and the third neural network.

[0090] In some embodiments, the welding parameter specifications of the three welding wires can be determined based on the welding gap between the rails. It is understood that using the welding gap to determine the corresponding process specifications of the three welding wires only ensures that the welding parameters are sufficient to completely fill the crater during stable welding, and does not characterize issues such as increased weld width and prolonged high-temperature time of the weld caused by the accumulation of heat during welding. In one or more embodiments of the present invention, corresponding temperature field features can be obtained based on a temperature field image, and correction values ​​for the welding parameters can be further determined based on these features. It is understood that the correction values ​​for the welding parameters should further include the water flow rate, water flow direction, water flow pulse velocity, and the process specifications of the three welding wires in the water-cooled copper block. It should be noted that in some implementation scenarios, in order to ensure the stability of the welding process and ensure that the accumulated metal can fill the welding gap according to the predetermined welding path and oscillation path, the process parameters of the three welding wires can only be fine-tuned. In this case, the main function of the neural network is to adjust the cooling effect of the water-cooled copper mold during welding, reduce the welding temperature of the molten pool, better control the molten pool formation, and avoid the risk of burn-through or insufficient welding.

[0091] Specifically, the neural network can determine the temperature distribution within the weld pool based on the brightness of the temperature field image combined with color band information from the camera. For example, dark red can be defined as 1500 degrees Celsius, light red as 1200 degrees Celsius, and orange as 800 degrees Celsius. This transforms the temperature field image into a temperature field matrix. Further, such as... Figure 6 As shown, the temperature field matrix is ​​input into the neural network, and the welding specifications corresponding to the temperature field image are coupled into the neural network to output the corrected welding specifications. It can be understood that since the welding specifications and the temperature field matrix may have different dimensions, one-hot encoding or other methods can be used to make them the same dimension when inputting them into the neural network, so as to facilitate subsequent forward propagation of the neural network.

[0092] Specifically, one round of training for the second and third neural networks can be... Figure 5 The neural network model shown is an example. It includes three network layers and a total of 6 neurons. The operation at each neuron is similar to that of neuron 6, and the forward propagation process of the neural network can be described by the following two formulas:

[0093] z = f(y)

[0094]

[0095] Where y represents the input data of the activation function f(·) of the neuron, z represents the output of the neuron, and for neurons in the output layer of the model, z can be the prediction result of the model on the training samples or the object to be predicted; the subscript n or m represents the index of the neuron, and in(n) represents the set of indices of the neurons in the layer preceding neuron n, with... Figure 5 For example, neuron 4 receives the outputs of neuron 1, neuron 2, and neuron 3, and in(4) = {1,2,3}. m,n b represents the weights that map neuron m to neuron n. n Let w be the constant term corresponding to neuron n. m,n and b n The model parameters that make up the neural network model can be obtained through training.

[0096] Through forward propagation, the feature data of the training samples can be processed layer by layer through each network layer of the neural network model to obtain the prediction results.

[0097] The backpropagation algorithm compares the prediction results of a specific training sample with the labeled data to determine the update magnitude of each weight in the network. In other words, the backpropagation algorithm is used to determine how the loss function changes relative to each weight (also known as the gradient or error derivative), denoted as .

[0098] by Figure 5 Taking an exemplary neural network model as an example, firstly, the gradient of the loss function value relative to the output of output neuron 6 can be calculated. When the loss function is the mean squared error loss function hour, Where z6 is the prediction result. This is the labeled data. Subsequently, the weight w between neuron 6 and neuron 5 can be calculated using the chain rule. 5,6 The gradient of the loss function value relative to the gradient of the output of neuron 5

[0099]

[0100]

[0101]

[0102] By analogy, the gradient of the loss function value with respect to each weight can be calculated one by one.

[0103] In view of the above process, the welding adjustment module 140 can adjust the welding based on the loss function value Loss. iBackpropagation of gradients is performed until the gradient of the loss function value relative to each element in the initial output matrix is ​​calculated. Then, the model is updated based on the gradient of each element, thus realizing one round of model update.

[0104] The welding adjustment module 140 then performs multiple iterations based on the above algorithm until the model converges or the model performance index meets the threshold requirement, thereby obtaining the trained first neural network model.

[0105] In some embodiments, the decision to proceed to the next iteration or to determine the final trained model can be based on the difference in performance and the sample labels. The criteria for this decision may include whether the preset number of iterations has been reached, whether the updated model meets a preset performance threshold, or whether a termination instruction has been received. If it is determined that the next iteration is necessary, it can be performed based on the updated model from the current iteration. If it is determined that the next iteration is not necessary, the updated model obtained during the current iteration can be used as the final trained model.

[0106] Figure 6 This is a schematic flowchart illustrating an exemplary CNN convolutional neural network training process.

[0107] Step 610: Obtain the initial model of the convolutional neural network.

[0108] The initial model of the convolutional neural network includes a first part and a second part, wherein the first part of the initial model has at least a number of convolutional layers, and the second part of the initial model includes at least a number of convolutional layers.

[0109] The initial model of a convolutional neural network includes initialized model parameters and a complete model structure. In some embodiments, the initial model can be an untrained convolutional neural network or a convolutional neural network that has not been fully trained. Each layer of the initial model can be set with initial parameters, which can be continuously adjusted during training until training is complete.

[0110] Step 620: Obtain the first training sample set and the second training sample set.

[0111] The first training sample set includes multiple temperature field images corresponding to the second stage of welding, corresponding welding parameter features, and corresponding temperature field feature labels. The second training sample set includes multiple temperature field images corresponding to the third stage of welding, corresponding welding parameter features, and corresponding temperature field feature labels. In some embodiments, the temperature field image labels can be manually annotated. In some alternative embodiments, the temperature field image labels can be machine-annotated. For example, another pre-trained temperature field image machine model can convert the temperature field images into corresponding temperature field feature labels. The welding parameter features include welding information such as the current, voltage, and welding time used in welding. These can be calculated using methods such as one-hot encoding to ensure that the calculation is performed in the same dimension as the temperature field feature label matrix, and then input into a convolutional neural network for subsequent calculations using methods such as cross product, dot product, or matrix expansion.

[0112] Step 630: Perform multiple rounds of training on the first part of the initial model based on the first training sample set, and train the second part of the initial model based on the second training sample set, to obtain the pre-trained first and second parts of the initial model.

[0113] It is understandable that this step involves training the corresponding training samples separately to obtain the first and second parts of the pre-trained initial model.

[0114] Step 640: Obtain a joint training sample set, which is a training set that meets the screening requirements based on the first training sample set and the second training sample set.

[0115] Because welding process II and welding process III differ significantly in temperature, yet they correspond to continuous welding processes, simply using a summation of two models with separate neural network outputs can lead to a considerable jump in welding parameters when transitioning from welding process II to welding process III. For welding, such large parameter jumps can pose a risk of arc interruption, affecting the final weld quality. Therefore, this invention relates to a method that uses one or more neural networks for joint training to eliminate abrupt changes in weld quality during the welding transition region and improve the applicability of the neural network.

[0116] Specifically, when selecting training samples, the sample set should consist entirely of temperature field images from the transition from welding process II to welding process III. However, such a sample size is relatively small, and therefore can only be used for joint training, not for the sample size requirements of model pre-training.

[0117] Step 650: Based on the joint training sample set, the first part and the second part of the pre-trained initial model are jointly trained to obtain the trained CNN convolutional neural network.

[0118] The first and second parts of the pre-trained initial model can be concatenated and then forward propagated through joint training samples to obtain the shape feature prediction value H, based on the labels of the joint training samples. Constructing the loss function Then, backpropagation is performed to obtain the corrected values ​​(or gradients) of the model parameters at each layer. These values ​​include multiple matrix elements (such as gradient elements), each corresponding one-to-one with a model parameter. Each gradient element reflects the direction (increase or decrease) and amount of parameter correction. For more information on forward and backward propagation, please refer to [link to documentation / reference]. Figure 4 The corresponding description will not be repeated here.

[0119] In some embodiments, the decision to proceed to the next iteration or to determine the final trained model can be based on the difference in performance and the sample labels. The criteria for this decision may include whether the preset number of iterations has been reached, whether the updated model meets a preset performance threshold, or whether a termination instruction has been received. If it is determined that the next iteration is necessary, it can be performed based on the updated model from the current iteration. If it is determined that the next iteration is not necessary, the updated model obtained during the current iteration can be used as the final trained model.

[0120] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling the weld shape of a three-wire welding nozzle electroslag welder for rails using temperature field coupling, characterized in that, Includes the following steps: The rails to be welded are placed in the welding position, and there is a welding gap between the rails to be welded. The welding gap can accommodate three electroslag welding nozzles, and welding wire passes through each nozzle. The water-cooled copper mold is placed on both sides of the rail to be welded position. Based on the welding gap, welding process parameters are designed, three-wire electroslag welding is started, and the temperature field in the weld pool of the steel rail is acquired in real time. The welding process parameters are adjusted in real time based on the temperature field. The welding process parameters include at least the flow rate of the water-cooled copper mold. Welding is divided into the following three stages: Stage I, Arc Initiation and Slag Formation Stage; Phase II, normal welding phase; Phase III, Introduction Phase; Real-time adjustment of welding process parameters based on temperature field includes at least: Obtain the trained second and third neural networks; The welding process parameters are corrected in real time based on the second neural network and the third neural network. Wherein, the second neural network and the third neural network correspond to the process parameter feedforward prediction networks of welding stage II and stage III, respectively. The process parameter feedforward prediction networks are used to generate corresponding welding process parameters based on the acquired temperature field. When the temperature field is a temperature field image, the second neural network and the third neural network are set as CNN convolutional neural networks. The second neural network and the third neural network are jointly trained according to the following steps: Obtain an initial model of a CNN (Convolutional Neural Network), which includes a first part and a second part. The first part of the initial model includes at least several convolutional layers and two fully connected layers, and the second part of the initial model includes at least several convolutional layers and one fully connected layer. The first part and the second part of the initial model are connected by a fully connected layer. Obtain a first training sample set and a second training sample set; the first training sample set includes multiple temperature field images corresponding to welding stage II, corresponding welding parameter features and corresponding temperature field feature labels, and the second training sample set includes multiple temperature field images corresponding to welding stage III, corresponding welding parameter features and corresponding temperature field feature labels. The first part of the initial model is trained based on the first training sample set, and the second part of the initial model is trained based on the second training sample set in multiple rounds to obtain the pre-trained first and second parts of the initial model. Obtain a joint training sample set, which is a training set that screens out temperature gradients that meet the screening requirements based on the first training sample set and the second training sample set; Based on the joint training sample set, the first and second parts of the pre-trained initial model are jointly trained to obtain a trained CNN convolutional neural network.

2. The method for controlling the shape characteristics of electroslag welds of three-wire welding nozzles for rails by temperature field coupling as described in claim 1, characterized in that, The welding process parameters designed based on the welding gap include at least: Welding parameter specifications are determined based on the welding gap; A reference rail gap is set, and the welding process parameters are corrected once based on the difference between the welding gap and the reference rail gap.

3. The method for controlling the electroslag weld characteristics of a three-wire welding nozzle for rails using temperature field coupling as described in claim 1, characterized in that... The welding process parameters include the water flow rate of the water-cooled copper mold, the direction of the water flow, the pulse speed of the water flow, and the process parameters of the three welding wires.

4. The method for controlling the weld shape of a three-wire welding nozzle for rails using temperature field coupling as described in any one of claims 1-3, characterized in that, The three electroslag welding nozzles include one main nozzle and two auxiliary nozzles, with the two auxiliary nozzles distributed on both sides of the main nozzle. The rail to be welded is in the shape of an "I", which includes a rail head, a rail web and a rail bottom. The two auxiliary nozzles are used to weld the rail bottom, and the main nozzle is used to weld the rail web and the rail head.

5. The method for controlling the shape characteristics of electroslag welds with three-wire welding nozzles for rails using temperature field coupling as described in claim 4, characterized in that... The water-cooled copper mold also includes a rail bottom copper mold, which is set at the bottom of the rail to be welded to quickly establish a stable welding pool.

6. A temperature field-coupled electroslag weld form control system for rail three-wire welding nozzles used to implement the temperature field-coupled electroslag weld form control method for rail three-wire welding nozzles as described in any one of claims 1-5, characterized in that, include: The welding preparation module is used to place the rails to be welded in the welding position. There is a welding gap between the rails to be welded. The welding gap can accommodate three electroslag welding nozzles, and welding wire passes through each nozzle. Two water-cooled copper molds are placed on both sides of the two rails to be welded. Welding start-up module, used for three-wire electroslag welding; Temperature field acquisition module is used to acquire the temperature field in the weld pool of the welded rail in real time; The welding adjustment module is used to design welding process parameters based on the welding gap and to adjust the welding process parameters in real time based on the temperature field; the welding process parameters include at least the flow rate of the water-cooled copper mold.

Images