Welding process optimization method for improving fracture toughness of weld joint at extremely low temperature

Abstract

The invention belongs to the technical field of extremely low temperature welding, and particularly relates to a welding process optimization method for improving weld fracture toughness at extremely low temperature, which comprises the following steps: 1) optimizing welding material chemical components with optimal mechanical properties through machine learning; 2) establishing a temperature field finite element model in the multi-layer and multi-pass welding process at extremely low temperature; (3) collecting mechanical property parameters of the welding seam at the extremely low temperature and setting multi-layer and multi-pass welding heat input and welding modes; and 4) determining welding speed and wire feeding speed parameters according to the actual welding condition, determining the heat distribution condition in the welding process by using a finite element calculation method, and carrying out residual stress and nondestructive testing analysis on the welding test plate. According to the multi-layer and multi-pass welding process optimization method at the extremely low temperature, the welding process is optimized in the mode that the welding material chemical components are optimized through machine learning and the heat distribution in the welding process is determined through finite element simulation, and therefore the optimal welding process for controlling the welding mechanical property and the excellent welding quality is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of ultra-low temperature welding, and in particular to a welding process optimization method for improving the fracture toughness of a weld at an ultra-low temperature. Background Art

[0002] Fusion energy is an ultimate clean energy source, and many countries in the world are conducting related research. The coil box of the fusion reactor is large in shape and has a "D"-shaped structure. Welding is the main manufacturing method for connecting the coil box. Due to the complex structure and working under extremely low temperature conditions (4.2K), welding is prone to introduce defects, residual stress and non-uniformity of organizational properties. The extremely low temperature toughness of the weld is an important performance indicator to ensure structural safety.

[0003] Usually, steel for extremely low temperatures is mainly austenitic stainless steel with a high Ni content. The factors affecting austenitic stainless steel welds are very complex and are usually related to factors such as ferrite content, solubility of alloying elements, welding process parameters, thermal cracks, and carbide precipitation. In traditional multi-layer and multi-pass welding methods, it is inevitable that the subsequent welding will heat the previous weld, causing unevenness in the structure of the weld and the heat-affected zone, generating higher post-weld thermal stress, and even defects such as cracks, pores, or slag inclusions, which reduce the fracture toughness of the material. In addition, the fracture toughness at extremely low temperatures is very sensitive to the chemical composition of the welding material. Traditional welding process control methods are mainly based on experience and trial and error methods, which cannot meet the requirements of extremely low temperature fracture toughness. At present, there is still no welding process optimization method for extremely low temperature fracture toughness, and it is urgent to develop new methods to provide technical support for the manufacture of related equipment.

[0004] In order to meet the above problems, in order to meet the requirements of faster welding material selection and achieve excellent mechanical properties of ultra-low temperature welds, machine learning provides a new idea for solving the above problems. The machine learning method can make full use of all welding process data and associate the chemical composition of welding materials with the mechanical properties of welds, thereby optimizing the welding process of ultra-low temperature welds. Summary of the invention

[0005] Based on the above-mentioned existing technical problems, the present invention proposes a welding process optimization method for improving the fracture toughness of welds at extremely low temperatures.

[0006] The present invention proposes a welding process optimization method for improving the fracture toughness of welds at extremely low temperatures, and the welding process optimization method comprises:

[0007] S1. Learning to optimize the chemical composition of welding materials with the best mechanical properties through machine learning models: Welds with different compositions will form welds with different hardness and toughness during the welding process. Based on the optimal mechanical property parameters of ultra-low temperature steel welding joints, machine learning methods are used to determine welding materials whose strength and toughness meet the requirements of low-temperature engineering.

[0008] S2. Establish a finite element model of the temperature field during multi-layer and multi-pass welding of ultra-low temperature steel: Use multi-layer and multi-pass welding to penetrate ultra-low temperature and ultra-thick plates. During the welding process, when the next layer is welded after the previous layer is welded, use the established temperature field finite element model to analyze the temperature distribution during the welding process to determine whether the heat from the previous layer of welding has been completely dissipated.

[0009] S3. Complete the collection of mechanical properties parameters of ultra-low temperature steel welds and the setting of multi-layer and multi-pass welding heat input and welding methods: Establish a machine learning model through the collected weld mechanical properties parameters and multi-layer and multi-pass welding heat input data to find the welding parameters with the greatest correlation to the mechanical properties. By recombining the welding parameters, the mechanical properties of the welds can be predicted in turn, thereby finding the welding process with the best mechanical properties.

[0010] S4. Determine the welding speed and wire feeding speed parameters according to the actual welding situation: Use the finite element calculation method to determine the heat distribution during welding, and perform residual stress and non-destructive testing analysis on the welding test plate. If the results are reasonable, the best welding process is obtained. If the results are unreasonable, re-execute S1, S2, S3 and S4 until the welding process parameters of the weld with the best mechanical properties are found.

[0011] Preferably, the machine learning model in S1 is constructed using an artificial neural network.

[0012] Through the above technical scheme, the mechanical properties of the weld are predicted by the chemical composition of the welding material, and then the chemical composition of the welding material is predicted in turn by the predicted mechanical property data of the weld with the best mechanical properties, thereby obtaining the chemical composition of the weld with the best mechanical properties.

[0013] Preferably, before analyzing the temperature distribution during welding using the established temperature field finite element model in S2, the three-dimensional model of the weldment is imported into the corresponding finite element calculation module through software, and the corresponding material properties are assigned, and the corresponding boundary conditions and initial conditions are added.

[0014] Through the above technical solution, appropriate boundary conditions, including constraint conditions and loading conditions, are added to the model according to the actual situation in the welding process. For example, the constraint relationship between the workpiece and the supporting structure, as well as the heat input conditions of the heat source, can be defined. The initial conditions of the welding process can be set in the finite element model, including the initial temperature distribution, the initial stress state, etc.

[0015] Preferably, when performing finite element model simulation of the temperature field during multi-layer and multi-pass welding of ultra-low temperature steel in S2, the position where the welding gun moves is divided into spatial physical layers, and the calculated temperature and displacement data along with the movement of the welding gun are exported as data in csv format.

[0016] Through the above technical solution, the data is used as input data for learning the machine learning model.

[0017] Preferably, when establishing the finite element model of the temperature field during the multi-layer and multi-pass welding process of ultra-low temperature steel in S2, a denser grid is used in the area closer to the welding area to capture the change of the temperature field, and a sparser grid is used in the farther area.

[0018] Through the above technical solution, in finite element analysis, meshing refers to dividing a complex structure or area into multiple small units, namely finite elements, for numerical calculation and simulation. By meshing the structure or area, the entire problem can be decomposed into many small units. The physical behavior in each unit can be described by a simple mathematical equation. Then, by establishing a mathematical model on each unit and connecting all the units, the entire structure can be numerically solved to obtain the response and behavior of the structure under different conditions. By using meshes of different densities in different areas, the calculation efficiency can be improved while ensuring the simulation accuracy, making the established finite element model more practical and operational. Such comprehensive considerations can better balance the contradiction between accuracy and efficiency, and provide better support for the simulation and analysis of multi-layer and multi-pass welding processes of ultra-low temperature steel.

[0019] Preferably, the mechanical properties of the weld in S3 include fracture toughness, yield strength, tensile strength, elastic modulus, and magnetic index.

[0020] Through the above technical solution, in order to ensure that machine learning predicts more comprehensive physical information, the temperature data used in finite element simulation calculations must be imported as input data for machine learning model learning, that is, a machine learning model is established through the collected weld mechanical properties parameters and heat input data of multi-layer and multi-pass welding, and the welding parameters with the greatest correlation with the mechanical properties are found. The mechanical properties of the weld are predicted in turn by recombining the welding parameters, thereby finding the welding process parameters with the best mechanical properties.

[0021] Preferably, the derivation relationship between the fracture toughness and the yield strength YS (MPa), Ni content (%) and inclusion spacing λ (mm) is as follows:

[0022] K IC =150-0.314YS+16.6[Ni]+2300λ

[0023] Preferably, the welding method described in S3 includes automatic tungsten inert gas arc welding and laser welding.

[0024] Preferably, when the finite element calculation method is used in S4 to determine the heat distribution during the welding process, the simulation of the multi-layer and multi-pass welding process is analyzed using a unit gradient activation technology.

[0025] Through the above technical solution, the unit gradient activation technology means that before welding begins, the unit at the weld is in a zero gradient unactivated state. As welding proceeds, the units are activated in sequence according to the heat transfer gradient theory, and the material temperature data of the corresponding welding material is applied.

[0026] Preferably, when the welding speed and the wire feeding speed are set in S4, the welding speed and the wire feeding speed parameters are determined by the relationship between the welding voltage, the welding current and the heat input.

[0027] Through the above technical scheme, when determining the welding speed and wire feeding speed parameters, it is necessary to comprehensively consider the relationship between welding voltage, welding current and heat input, and make adjustments based on specific materials and welding requirements. By reasonably adjusting these parameters, high-quality welding effects can be achieved, ensuring the firmness and appearance quality of the weld.

[0028] The beneficial effects of the present invention are:

[0029] 1. The multi-layer and multi-pass welding process of the ultra-low temperature steel of the present invention optimizes the welding process by combining machine learning to optimize the chemical composition of the welding material with finite element simulation to determine the heat distribution during welding, thereby reducing the trial and error cost of manually adjusting the welding process parameters and obtaining the optimal welding process for controlling welding mechanical properties and excellent welding quality.

[0030] 2. By setting up machine learning, it can learn from existing data and predict unknown data through learning. Combined with finite element calculation simulation data, it can learn from the chemical composition data of existing welding materials to predict the chemical composition of non-existent welding materials. The chemical element design of welding materials can be carried out, thereby reducing the experimental exploration cost of welding material development, improving the quality of welded joints, and providing a very convenient and economical optimization method for the development of welding technology for ultra-low temperature steel. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a system structure diagram of the present invention;

[0032] Figure 2 is a flow chart of optimizing welding material composition by machine learning in an embodiment of the present invention;

[0033] Figure 3 It is a flow chart of finite element simulation temperature field in an embodiment of the present invention.

[0034] Figure 4 It is a neural network diagram for optimizing welding process parameters in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0036] Reference Figure 1-Figure 4 , a welding process optimization method for improving the fracture toughness of welds at extremely low temperatures, the welding process optimization method comprising:

[0037] S1. The machine learning model is used to learn and optimize the chemical composition of welding materials with the best mechanical properties. Specifically, because welding materials with different compositions have different hardness and toughness in the welding process, the welding materials with strength and toughness that meet the requirements of low-temperature engineering are determined through machine learning methods based on the optimal mechanical performance parameters of ultra-low temperature steel welded joints.

[0038] The machine learning model is constructed using an artificial neural network. It predicts the mechanical properties of the weld through the chemical composition of the welding material, and then predicts the chemical composition of the welding material through the predicted mechanical properties data of the weld with the best mechanical properties, thereby obtaining the chemical composition of the weld with the best mechanical properties.

[0039] S2. Establish a finite element model of the temperature field in the multi-layer and multi-pass welding process of ultra-low temperature steel. Specifically, the ultra-low temperature and ultra-thick plates are welded through by multi-layer and multi-pass welding. During the welding process, when the next layer of welding is carried out after the previous layer of welding is completed, the established temperature field finite element model is used to analyze the temperature distribution in the welding process to determine whether the heat of the previous layer of welding has been completely dissipated. When simulating the finite element model of the temperature field in the multi-layer and multi-pass welding process of ultra-low temperature steel, the position where the welding gun moves is divided into spatial physical layers, and the calculated temperature and displacement data along with the movement of the welding gun are exported as data in csv format, which is used as input data for learning the machine learning model.

[0040] Before using the established temperature field finite element model to analyze the temperature distribution during welding, the three-dimensional model of the weldment is imported into the corresponding finite element calculation module through the software, and the corresponding material properties are assigned, and the corresponding boundary conditions and initial conditions are added. According to the actual situation in the welding process, appropriate boundary conditions are added to the model, including constraint conditions and loading conditions. For example, the constraint relationship between the workpiece and the supporting structure, as well as the heat input conditions of the heat source, can be defined. The initial conditions of the welding process are set in the finite element model, including the initial temperature distribution, the initial stress state, etc.

[0041] At the same time, meshing is related to the accuracy and efficiency of the solution calculation. Accuracy and efficiency are often contradictory. Considering comprehensively, when establishing the finite element model of the temperature field in the multi-layer and multi-pass welding process of ultra-low temperature steel, a denser mesh is used in the area close to the welding area to capture the change of the temperature field, and a sparser mesh is used in the farther area. In finite element analysis, meshing refers to dividing a complex structure or area into multiple small units, namely finite elements, for numerical calculation and simulation. By meshing the structure or area, the entire problem can be decomposed into many small units. The physical behavior in each unit can be described by simple mathematical equations. Then, by establishing a mathematical model on each unit and connecting all the units, the entire structure can be numerically solved to obtain the response and behavior of the structure under different conditions. By using meshing of different densities in different areas, the calculation efficiency can be improved while ensuring the simulation accuracy, making the established finite element model more practical and operable. Such comprehensive consideration can better balance the contradiction between accuracy and efficiency, and provide better support for the simulation and analysis of the multi-layer and multi-pass welding process of ultra-low temperature steel.

[0042] The multi-layer and multi-pass welding process for ultra-low temperature steel of the present invention optimizes the welding process by combining machine learning to optimize the chemical composition of the welding material with finite element simulation to determine the heat distribution during welding, thereby reducing the trial and error cost of manually adjusting the welding process parameters and obtaining the optimal welding process for controlling welding mechanical properties and excellent welding quality.

[0043] S3. Complete the collection of mechanical property parameters of ultra-low temperature steel welds and the setting of multi-layer and multi-pass welding heat input and welding methods. Specifically, a machine learning model is established through the collected data such as weld mechanical property parameters and multi-layer and multi-pass welding heat input to find the welding parameters with the greatest correlation to the mechanical properties. The mechanical properties of the weld are predicted in turn by recombining the welding parameters, thereby finding the welding process parameters with the best mechanical properties. The welding methods include automatic tungsten inert gas welding and laser welding.

[0044] In order to ensure that machine learning predicts more comprehensive physical information, the temperature data used in finite element simulation calculations must be imported as input data for machine learning model learning. That is, a machine learning model is established through the collected data on weld mechanical properties parameters and heat input of multi-layer and multi-pass welding to find the welding parameters with the greatest correlation with the mechanical properties. The mechanical properties of the weld are predicted in turn by recombining the welding parameters, thereby finding the welding process parameters with the best mechanical properties. The mechanical properties of the weld include fracture toughness, yield strength, tensile strength, elastic modulus, and magnetic index.

[0045] The derived relationship between fracture toughness and yield strength YS (MPa), Ni content (%) and inclusion spacing λ (mm) is as follows:

[0046] K IC =150-0.314YS+16.6[Ni]+2300λ

[0047] S4. Determine the welding speed and wire feeding speed parameters according to the actual welding situation. Specifically, use the finite element calculation method to determine the heat distribution during the welding process, and perform residual stress and non-destructive testing analysis on the welding test plate. If the result is reasonable, the best welding process is obtained. If the result is unreasonable, re-execute S1, S2, S3 and S4 until the welding process of the weld with the best mechanical properties is found. Specifically, it needs to be determined based on the relationship between welding voltage, welding current and heat input, because when determining the welding speed and wire feeding speed parameters, it is necessary to comprehensively consider the relationship between welding voltage, welding current and heat input, and adjust them according to specific materials and welding requirements. By reasonably adjusting these parameters, high-quality welding effects can be achieved to ensure the firmness and appearance quality of the weld.

[0048] When the finite element calculation method is used to determine the heat distribution during the welding process, the simulation of the multi-layer and multi-pass welding process is analyzed using the unit gradient activation technology. The unit gradient activation technology means that before welding begins, the unit at the weld is in a zero gradient and unactivated state. As welding proceeds, the units are activated in turn according to the heat transfer gradient theory, and the material temperature data of the corresponding welding material is applied.

[0049] The above-mentioned ultra-low temperature steel welding process is subjected to finite simulation solution calculation to obtain the corresponding temperature field data. The welding process is optimized by taking the temperature data and the weld mechanical property data as the characteristic parameters of machine learning, and the optimization results are judged. Specifically, if the obtained results meet the requirements of the ultra-low temperature steel welding structure, the welding process that meets the welding requirements can be determined. Otherwise, re-learning is performed to obtain a new optimized welding process by changing the weld mechanical property parameters and adjusting the machine learning hyperparameters, thereby achieving the purpose of optimizing the welding process.

[0050] By setting up machine learning, it can learn from existing data and predict unknown data through learning. Combined with finite element calculation simulation data, it can learn based on the chemical composition data of existing welding materials to predict the chemical composition of non-existent welding materials. The chemical element design of welding materials can be carried out, thereby reducing the experimental exploration cost of welding material development, improving the quality of welded joints, and providing a very convenient and economical optimization method for the development of welding processes for ultra-low temperature steels.

[0051] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A welding process optimization method for improving the fracture toughness of welds at extremely low temperatures, characterized in that: The welding process optimization method comprises: S1. Learning and optimizing the chemical composition of welding materials with the best mechanical properties through machine learning models: welding materials with different compositions have different hardness and toughness in the welds formed during welding. Based on the optimal mechanical property parameters of ultra-low temperature steel welded joints, machine learning methods are used to determine welding materials whose strength and toughness meet the requirements of cryogenic engineering. S2. Establish a finite element model of the temperature field during multi-layer and multi-pass welding of ultra-low temperature steel: Use multi-layer and multi-pass welding to penetrate ultra-low temperature and ultra-thick plates. During the welding process, when the next layer is welded after the previous layer is welded, use the established temperature field finite element model to analyze the temperature distribution during the welding process to determine whether the heat from the previous layer of welding has been completely dissipated; S3. Complete the collection of mechanical property parameters of ultra-low temperature steel welds and the setting of multi-layer multi-pass welding heat input and welding method: Establish a machine learning model through the collected weld mechanical property parameters and multi-layer multi-pass welding heat input data to find the welding parameters with the greatest correlation to the mechanical properties, and predict the mechanical properties of the welds in turn through the recombination of the welding parameters, so as to find the welding process parameters with the best mechanical properties; S4. Determine the welding speed and wire feeding speed parameters according to the actual welding situation: Use the finite element calculation method to determine the heat distribution during welding, and perform residual stress and non-destructive testing analysis on the welding test plate. If the results are reasonable, the best welding process is obtained. If the results are unreasonable, re-execute S1, S2, S3 and S4 until the welding process parameters of the weld with the best mechanical properties are found.

2. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: The machine learning model in S1 is constructed using an artificial neural network.

3. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: Before analyzing the temperature distribution during welding using the established temperature field finite element model in S2, the three-dimensional model of the weldment is imported into the corresponding finite element calculation module through software, and the corresponding material properties are assigned, and the corresponding boundary conditions and initial conditions are added.

4. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: When the finite element model simulation of the temperature field during the multi-layer and multi-pass welding of ultra-low temperature steel is performed in S2, the position where the welding gun moves is divided into spatial physical layers, and the temperature and displacement data calculated along with the movement of the welding gun are exported as data in csv format.

5. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: When establishing the finite element model of the temperature field during the multi-layer and multi-pass welding of ultra-low temperature steel in S2, a denser grid is used in the area closer to the welding area to capture the change of the temperature field, and a sparser grid is used in the farther area.

6. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: The mechanical properties of the weld in S3 include fracture toughness, yield strength, tensile strength, elastic modulus, and magnetic index.

7. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 5, characterized in that: The derivation relationship between the fracture toughness and the yield strength YS (MPa), Ni content (%) and inclusion spacing λ (mm) is as follows: K IC =150-0.314YS+16.6[Ni]+2300λ。 8. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: The welding method described in S3 includes automatic tungsten inert gas arc welding and laser welding.

9. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: When the finite element calculation method is used in S4 to determine the heat distribution during the welding process, the simulation of the multi-layer and multi-pass welding process is analyzed using the unit gradient activation technology.

10. The welding process optimization method for improving the fracture toughness of welds at extremely low temperatures according to claim 1, characterized in that: When the welding speed and the wire feeding speed are set in S4, the welding speed and the wire feeding speed parameters are determined by the relationship between the welding voltage, the welding current and the heat input.

Images