A welding process optimization method and system for 7Ni steel machining

Abstract

The application belongs to the field of advanced non-ferrous metal material welding technology, and discloses a welding process optimization method and system for 7Ni steel processing; welding process optimization features are obtained based on a three-party feature vector fusion method by using the detected electrode feature vector, base material characteristic parameters and welding process parameters; a welding process optimization model is constructed by using welding process optimization features, welding process performance features composed of the welding process method used at the time and weld defect features; under the processing of a multiple neural network model, the obtained welding process is more reliable, and welding defects are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of advanced nonferrous metal material welding, and in particular relates to a welding process optimization method and system for 7Ni steel processing. Background Art

[0002] Welding is both a key and challenging aspect of LNG tank manufacturing. Numerous scholars, both domestically and internationally, have conducted extensive research on welding methods, welding processes, and post-weld inspection. In the field of LNG tank welding, the main welding methods currently used are arc welding, submerged arc welding, gas metal arc welding, tungsten inert gas welding, and laser welding. Arc welding is currently the main method, followed by submerged arc welding and tungsten inert gas welding. Arc welding has relatively low welding efficiency, and gas metal arc welding generally uses pulsed current to ensure quality. Tungsten inert gas welding can ensure high-quality weld joints, but has low productivity and high costs. Submerged arc welding is a highly efficient welding method, especially when welding girth welds. Due to the use of a girth welding mechanical system, its advantages are even more prominent, making it suitable for welding almost all transverse and horizontal welds. This paper aims to promote the application of 7Ni steel in LNG tank manufacturing and reduce manufacturing costs. For the welding of 7Ni steel, there is currently only theoretical research on welding using the method of arc welding with electrodes, and there are no practical application cases. There are also no research cases using submerged arc welding or other high-efficiency welding methods for 7Ni steel. Therefore, the existing technology has the following problems: Most welding process methods are selected based on the consideration of the base material, but the fact that welding is a project promoted by the joint efforts of three parties is ignored. This requires consideration of factors such as welding materials, base materials and the environment. How to link these three with welding process optimization is an urgent problem to be solved. At the same time, there are few artificial intelligence methods used in the existing technology to optimize the selection of welding processes. Most neural network models are used for welding defect diagnosis, and there are still certain gaps in the optimization selection of welding methods. At the same time, there is also a certain neglect in the acquisition of image features of the welding rod before welding, and industrial images cannot be well utilized in the selection of welding optimization process methods. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a welding process optimization method and system for 7Ni steel processing.

[0004] In a first aspect of the present invention, a welding process optimization method for 7Ni steel processing is provided, the method comprising:

[0005] P1. Detect and process the electrode feature vector, collect the welding base material characteristic parameters, obtain the welding process parameters, and obtain the corresponding welding process method and welding process performance composed of weld defect characteristics;

[0006] P2. Based on a three-way feature vector fusion method, high-dimensional fusion of the welding rod feature vector, the parent material characteristic parameters, and the welding process parameters is used to obtain welding process optimization features;

[0007] P3. Constructing a welding process optimization model based on the welding process optimization characteristics and the welding process performance;

[0008] P4. Output new welding process performance for the 7Ni steel to be subsequently welded according to the welding process optimization model.

[0009] Furthermore, the welding process optimization model is constructed by inputting the welding process optimization features and the welding process performance into a multiple activation function neural network model improved based on welding process parameters.

[0010] Furthermore, the welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters, and the welding rod optical image feature is obtained by acquiring and processing the feature through a CCD camera.

[0011] Furthermore, the basic parameters of the welding rod include the welding rod diameter, the coating radius and the welding rod eccentricity, wherein the welding rod eccentricity measurement formula is:

[0012]

[0013] Where E is the eccentricity of the welding rod, The distance from the center of the welding rod core to the outside of the welding rod, is the distance between the center of the electrode coating and the outside of the electrode, D is the coating radius, and the electrode feature vector is constructed by combining the optical image features of the electrode with the electrode diameter, coating radius, and electrode eccentricity.

[0014] Furthermore, the parent material characteristic parameters include the content of reversed austenite, carbon content and density of 7Ni steel, and the welding process parameters include welding speed, welding environment temperature and welding environment humidity.

[0015] Furthermore, the multiple activation function neural network model improved based on welding process parameters adopts a dual activation function input method to construct the model.

[0016] Furthermore, the high-dimensional fusion method based on the tripartite feature vector fusion method is obtained by longitudinally splicing the feature vectors of the welding rod feature vector, the parent material characteristic parameters and the welding process parameters and then supplementing the matrix vector values.

[0017] A welding process optimization system for 7Ni steel processing is also provided, which includes a welding rod feature processing generation module, a welding base material characteristic parameter acquisition module, a welding process parameter acquisition module, a welding process optimization feature generation module, a welding process optimization model construction module, and a welding process and performance control module:

[0018] The welding rod feature processing and generating module is used to detect and process the welding rod feature vector;

[0019] The welding base material characteristic parameter acquisition module is used to collect the welding base material characteristic parameters;

[0020] The welding process parameter acquisition module is used to obtain welding process parameters;

[0021] The welding process optimization feature generation module is configured to obtain the welding process optimization feature through high-dimensional fusion of the welding rod feature vector, the base material characteristic parameters and the welding process parameters through a three-way feature vector fusion method;

[0022] The welding process optimization model construction module is used to construct a welding process optimization model based on the welding process optimization characteristics and welding process performance;

[0023] The welding process and performance control module is configured to obtain welding process performance and obtain new welding process performance of the 7Ni steel to be welded according to the welding process optimization model.

[0024] Furthermore, the welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters, and the welding rod optical image feature is obtained by acquiring and processing the feature through a CCD camera.

[0025] Furthermore, the multiple activation function neural network model improved based on welding process parameters adopts a dual activation function input method to construct the model.

[0026] The present invention considers three factors, namely welding rod, base material and environment, and performs high-dimensional and efficient processing of the three-party input features for the preferred welding process method, and considers feature fusion of the three-party welding process preferred features. At the same time, taking into account the lack of output processing of welding process optimization methods in existing deep learning models, and the gradient influence of high-dimensional feature input features on the activation function in the deep learning model under the scenario application, thereby affecting the robustness of the deep learning model in the selection of welding optimization process methods, the present invention adopts a dual activation function to reduce the influence of the activation function gradient, and based on the welding process method preferred scenario, dynamically combines two activation functions, ReLU activation function and Tanh activation function improved based on welding process parameters, to enhance the expression ability, convergence ability and generalization performance of the model. The present invention also processes the welding rod image data, processes the entire image according to the huge differences between the welding rod coating, flux core and other image pixels, and removes the corresponding edge pixel image according to the pixel defects of the edge image during shooting to reduce the generalization of pixel features. The optical image features of the welding rod obtained after processing are more in line with the selection of the welding optimization process method of the deep learning model in the present invention, thereby improving the model's processing quality of the input features and improving the accuracy of welding method selection.

[0027] In addition, the beneficial effect of the present invention is to construct a multiple neural network model based on the three-party characteristics of welding rods, welding mother parts and welding process parameters, so as to process the relevant input features that affect welding performance in a high-dimensional and efficient manner. By processing the three-party influencing feature vectors, the welding process optimization characteristics suitable for 7Ni steel processing and welding are obtained. The constructed process optimization model is more robust and has an optimized output of the welding process method under the 7Ni steel welding scenario. Under the processing of optical image features of the input image features and the use of the dual neural network model, the obtained welding process is more reliable and reduces welding defects.

[0028] More embodiments and improved effects of the present invention will be further introduced in conjunction with the drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of the welding process optimization method for 7Ni steel processing of the present invention;

[0030] Figure 2 Schematic diagram of the welding process optimization system for 7Ni steel processing according to the present invention;

[0031] Figure 3 This is an example of an optical image of a welding rod obtained after processing in an embodiment of the present invention;

[0032] Figure 4 Schematic diagram of the ReLU activation function and the dual activation function in the embodiment of the present invention. DETAILED DESCRIPTION

[0033] Below, the invention is further described in conjunction with the accompanying drawings and specific implementation methods. The model used in the present invention is an improved version of the dual activation function of the neural network model.

[0034] The 7Ni steel in the present invention is composed of 7% nickel and a certain amount of carbon, silicon, manganese and other elements, and therefore belongs to the field of advanced non-ferrous metal materials.

[0035] On the one hand, by welding 7Ni steel using different welding methods and different welding materials, the type, shape, size, etc. of joint welding defects can be determined through macroscopic inspection and non-destructive testing, and the defect generation process and mechanism can be analyzed;

[0036] On the other hand, the microstructure morphology, composition phase, grain size, and second phase of different areas of welded joints with different welding methods and materials were studied by optical microscopy, scanning electron microscopy, energy dispersive spectrometer, X-ray diffractometer, EBSD, transmission electron microscopy and other analytical methods. The amount, distribution and morphology of reversed austenite were analyzed to reveal the welding metallurgical reaction process and organizational transformation mechanism under different working conditions.

[0037] Through the above description, it can be determined that the characteristic vectors considered in the present invention are all characteristics that have a significant impact on the 7Ni steel welding process.

[0038] In a first aspect of the present invention, a welding process optimization method for 7Ni steel processing is provided, the method comprising:

[0039] P1. Detect and process the electrode feature vector, collect the welding base material characteristic parameters, obtain the welding process parameters, and obtain the corresponding welding process method and welding process performance composed of weld defect characteristics;

[0040] P2. Based on a three-way feature vector fusion method, high-dimensional fusion of the welding rod feature vector, the parent material characteristic parameters, and the welding process parameters is used to obtain welding process optimization features;

[0041] P3. Constructing a welding process optimization model based on the welding process optimization characteristics and the welding process performance;

[0042] P4. Output new welding process performance for the 7Ni steel to be subsequently welded according to the welding process optimization model.

[0043] In this embodiment, since 7Ni steel is optimized on the basis of the alloy composition of 9Ni steel, the Ni content is reduced by 2%, but the content of Cr and Mo is increased, which reduces the low-temperature toughness and improves the hardenability. Considering the welding structural characteristics of the LNG storage tank and the chemical composition, strength and low-temperature toughness of 7Ni steel, and analogous to the welding of 9Ni steel, the welding process methods respectively correspond to the three welding methods of arc welding, submerged arc welding and tungsten inert gas welding and their matching welding materials. The GTAW method uses the Anhui Aotai argon arc welding machine with a model of ZX7-400STG, the SMAW method uses the Shanghai Weiteli welding machine with a model of ZX7-500, and the SAW method uses the Anhui Aotai submerged arc welding machine with a model of MZ-1250. This project adopts GTAW flat welding (1G), SMAW flat welding (1G), horizontal welding (2G), vertical welding (3G), overhead welding (4G), and SAW flat welding (1G) for 7Ni steel plates.

[0044] The weld defect signature is set based on the weld defect condition after welding. Its label value can be one of three, corresponding to -1, 0, and 1, respectively. This is a conventional design by those skilled in the art based on the defect condition. In this embodiment, a virtual, predictive weld defect condition is generated for subsequent welding of 7Ni steel for staff to assess, thereby improving the accuracy of welding process selection and weld quality.

[0045] Furthermore, the welding process optimization model is constructed by inputting the welding process optimization features and the welding process performance into a multiple activation function neural network model improved based on welding process parameters.

[0046] Furthermore, the welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters. The welding rod optical image feature is obtained by acquiring and processing the CCD camera. This is because the shooting resolution and detection accuracy of the CCD camera have problems. Therefore, it is necessary to perform a certain degree of image processing on the welding rod image acquired by the CCD camera to obtain the welding rod optical image feature. The overall image is processed based on the huge difference between the welding rod coating, flux core and the rest of the image pixels. The corresponding edge pixel image is removed based on the pixel defects of the edge image during shooting to reduce the generalization of pixel features. The welding rod optical image feature formula obtained after processing is:

[0047]

[0048] Where, is the optical image feature of the welding rod, which is the calculated pixel value of the overall processing of the welding rod image. m and n are the maximum values ​​of the horizontal and vertical coordinates in the Cartesian coordinate system constructed by the two-dimensional welding rod image. i and j represent the horizontal and vertical coordinate values ​​of the constructed pixel Cartesian coordinates. represents the pixel value of the i-th horizontal and j-th column, Represents the maximum pixel value of the entire welding rod image, Indicates the minimum pixel value of the entire welding rod image.

[0049] Furthermore, the basic parameters of the welding rod include the welding rod diameter, the coating radius and the welding rod eccentricity, wherein the welding rod eccentricity measurement formula is:

[0050]

[0051] Where E is the eccentricity of the welding rod, The distance from the center of the welding rod core to the outside of the welding rod, is the distance between the center of the electrode coating and the outside of the electrode, D is the coating radius, and the electrode feature vector is constructed by combining the optical image features of the electrode with the electrode diameter, coating radius, and electrode eccentricity.

[0052] Furthermore, the parent material characteristic parameters include the content of reversed austenite, carbon content and density of 7Ni steel, and the welding process parameters include welding speed, welding environment temperature and welding environment humidity.

[0053] A neural network is a computational model composed of a large number of interconnected neurons. Inspired by the human nervous system, it combines and trains multiple layers of neurons to achieve highly adaptive and nonlinear mapping from input data to output.

[0054] Neural networks typically consist of multiple layers, including input, hidden, and output layers. The input layer receives input data, while the hidden and output layers are responsible for calculating the output. Each neuron receives multiple inputs from the previous layer, weights and calculates these inputs, adds a bias, and then performs a nonlinear transformation using an activation function to ultimately produce an output. The connections between neurons are typically represented by weights, which indicate the strength of the connection. Weights can be updated through training to adjust the strength of the connections between neurons. Neural networks have strong adaptability and nonlinear mapping capabilities, making them adaptable to a variety of input data and complex problems. In practical applications, neural networks have been widely used in image recognition, speech recognition, natural language processing, intelligent control, and other fields, achieving considerable success.

[0055] The present invention is based on the general neural network model and selects the activation function used in conventional principles to perform nonlinear transformation of input feature data. Therefore, it is an improvement on the activation function of the general conventional neural network, transforming the use of only one activation function into the use of multiple activation functions in this application.

[0056] Tanh activation function: The Tanh (hyperbolic tangent) activation function is a nonlinear activation function commonly used in deep learning. Its shape is similar to the Sigmoid activation function. This function maps input values ​​to a range between -1 and 1. This zero-centered nature makes the Tanh function more powerful than the Sigmoid function in terms of representational power.

[0057] ReLU activation function: ReLU (rectified linear unit) is a very popular activation function in deep learning. It is mainly used in the hidden layers of neural networks. ReLU is simple and efficient in design, which can effectively deal with the vanishing gradient problem and enable the training of deep neural networks.

[0058] In another embodiment of the present invention, the multiple activation function neural network model based on welding process parameter improvement adopts a dual activation function input method to construct the model. The welding process optimization feature is first input into the Tanh activation function based on welding process parameter improvement and then into the ReLU activation function. By dynamically combining these two activation functions, since the ReLU activation function has the advantages of fast calculation and can solve the gradient disappearance, while Tanh will cause the gradient to disappear, the combination of the two can enhance the expression ability, convergence ability and generalization performance of the model. The calculation formula of the dual activation function is:

[0059]

[0060] Where, is the dual activation function value, is the input welding process optimization feature, V is the welding speed, T is the welding environment temperature, and H is the welding environment humidity. Since the welding environment has a great influence on the welding process, the influence of the environment on the generalization of the activation function is fully considered during the model construction process, and the trace influence of the welding process parameters is applied to the influence of the activation function to make the model more robust. Since the welding speed in the welding process parameters is generally expressed in centimeters per minute, and the welding environment temperature of 7Ni steel is generally between 15℃ and 30℃, and the welding environment humidity is generally best maintained between 30% and 50%, in order to ensure dimensional consistency, the logarithmic function is used to remove the dimension.

[0061] Furthermore, the high-dimensional fusion method based on the tripartite feature vector fusion method is obtained by longitudinally splicing the feature vectors of the welding rod feature vector, the parent material characteristic parameters, and the welding process parameters, and then supplementing the matrix vector value. The supplemented matrix vector value is 1. When supplementing the input features, supplementing 1 is a method to improve the model input quality, which provides a more standardized input format for the processing of the model's input features, so as to facilitate the subsequent model input processing to obtain accurate and reliable output. The specific high-dimensional fusion longitudinal splicing matrix vector is:

[0062] .

[0063] A welding process optimization system for 7Ni steel processing is also provided, which includes a welding rod feature processing generation module, a welding base material characteristic parameter acquisition module, a welding process parameter acquisition module, a welding process optimization feature generation module, a welding process optimization model construction module, and a welding process and performance control module:

[0064] The welding rod feature processing and generating module is used to detect and process the welding rod feature vector;

[0065] The welding base material characteristic parameter acquisition module is used to collect the welding base material characteristic parameters;

[0066] The welding process parameter acquisition module is used to obtain welding process parameters;

[0067] The welding process optimization feature generation module is configured to obtain the welding process optimization feature through high-dimensional fusion of the welding rod feature vector, the base material characteristic parameters and the welding process parameters through a three-way feature vector fusion method;

[0068] The welding process optimization model construction module is used to construct a welding process optimization model based on the welding process optimization characteristics and welding process performance;

[0069] The welding process and performance control module is configured to obtain welding process performance and obtain new welding process performance of the 7Ni steel to be welded according to the welding process optimization model.

[0070] Furthermore, the welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters. The welding rod optical image feature is obtained by acquiring and processing the CCD camera. This is because the shooting resolution and detection accuracy of the CCD camera have problems. Therefore, it is necessary to perform a certain degree of image processing on the welding rod image acquired by the CCD camera to obtain the welding rod optical image feature. The image is processed as a whole based on the huge difference between the welding rod coating, flux core and the remaining image pixels. The corresponding edge pixel image is removed based on the pixel defects of the edge image during shooting to reduce the generalization of pixel features. Therefore, the number of image edge pixels is removed when calculating the welding rod optical image feature. The welding rod optical image feature formula obtained after processing is:

[0071]

[0072] Where, is the optical image feature of the welding rod, which is the calculated pixel value of the overall processing of the welding rod image. m and n are the maximum values ​​of the horizontal and vertical coordinates in the Cartesian coordinate system constructed by the two-dimensional welding rod image. i and j represent the horizontal and vertical coordinate values ​​of the constructed pixel Cartesian coordinates. represents the pixel value of the i-th horizontal and j-th column, Represents the maximum pixel value of the entire welding rod image, Indicates the minimum pixel value of the entire welding rod image. The pixel value of the present invention is between 1-255. The characteristic value of the welding rod optical image obtained after processing is 145 in one embodiment of the present invention.

[0073] Furthermore, the multiple activation function neural network model based on welding process parameter improvement adopts the dual activation function input method to construct the model. The welding process optimization feature is first input into the Tanh activation function based on welding process parameter improvement and then into the ReLU activation function. By dynamically combining these two activation functions, since the ReLU activation function has the advantages of fast calculation and can solve the gradient disappearance problem, while Tanh will cause the gradient to disappear, the combination of the two can enhance the model's expression ability, convergence ability and generalization performance. The dual activation function calculation formula is:

[0074]

[0075] Where, is the dual activation function value, is the input welding process optimization feature, V is the welding speed, T is the welding environment temperature, H is the welding environment humidity, and the max function is a general form known to those skilled in the art. In this application, it is a function from 0 to Take the maximum value. Since the welding environment has a significant impact on the welding process, the impact of the environment on the generalization of the activation function is fully considered during the model construction process. The slight influence of the welding process parameters is applied to the influence of the activation function to make the model more robust. Since the welding speed in the welding process parameters is generally expressed in centimeters per minute, the welding environment temperature of 7Ni steel is generally between 15°C and 30°C, and the welding environment humidity is generally maintained between 30% and 50%, in order to ensure dimensional consistency, the logarithmic function is used to remove the dimension.

[0076] After obtaining the aforementioned welding process optimization feature vector value, one of the neural network training feature vectors is:

[0077]

[0078] At this time, the corresponding welding process method characteristic vector value 1.5 and the weld defect characteristic vector value 1 constitute the welding process performance. There are six welding process methods: GTAW flat welding (1G), SMAW flat welding (1G), horizontal welding (2G), vertical welding (3G), overhead welding (4G), and SAW flat welding (1G), which correspond to corresponding values ​​1, 1.5, 2, 3, 4, and 1.8, respectively, and have corresponding label values ​​of -1, 0, and 1. The model parameters required for the neural network model are trained using this set of training data. In this embodiment, the data for training the neural network model is more than this set of data. After obtaining the final neural network model parameters through multiple sets of such training data, another set of welding process optimization characteristic vector values ​​is input. In this embodiment, the input model obtains the corresponding output value 1.8 and the corresponding label value -1, indicating that the SAW flat welding (1G) method is recommended for welding process optimization.

[0079] The present invention considers three factors, namely welding rod, base material and environment, and performs high-dimensional and efficient processing of the three-party input features for the preferred welding process method, and considers feature fusion of the three-party welding process preferred features. At the same time, taking into account the lack of output processing of welding process optimization methods in existing deep learning models, and the gradient influence of high-dimensional feature input features on the activation function in the deep learning model under the scenario application, thereby affecting the robustness of the deep learning model in the selection of welding optimization process methods, the present invention adopts a dual activation function to reduce the influence of the activation function gradient, and based on the welding process method preferred scenario, dynamically combines two activation functions, ReLU activation function and Tanh activation function improved based on welding process parameters, to enhance the expression ability, convergence ability and generalization performance of the model. The present invention also processes the welding rod image data, processes the entire image according to the huge differences between the welding rod coating, flux core and other image pixels, and removes the corresponding edge pixel image according to the pixel defects of the edge image during shooting to reduce the generalization of pixel features. The optical image features of the welding rod obtained after processing are more in line with the selection of the welding optimization process method of the deep learning model in the present invention, thereby improving the model's processing quality of the input features and improving the accuracy of welding method selection.

[0080] In addition, the beneficial effect of the present invention is to construct a multiple neural network model based on the three-party characteristics of welding rods, welding mother parts and welding process parameters, so as to process the relevant input features that affect welding performance in a high-dimensional and efficient manner. By processing the three-party influencing feature vectors, the welding process optimization characteristics suitable for 7Ni steel processing and welding are obtained. The constructed process optimization model is more robust and has an optimized output of the welding process method under the 7Ni steel welding scenario. Under the processing of optical image features of the input image features and the use of the dual neural network model, the obtained welding process is more reliable and reduces welding defects.

[0081] Of course, it can be understood that each embodiment of the present invention can achieve one of the effects alone, and a combination of multiple embodiments of the present invention can achieve all of the above effects, but it is not required that each embodiment of the present invention achieve all of the above advantages and effects, because each embodiment of the present invention can constitute a separate technical solution and make one or more contributions to the existing technology. Each embodiment of the present invention is not affected, and the solution can still be implemented even if one of the embodiments is deleted.

[0082] The prior art mentioned in the above background technology section and specific embodiments section of the present invention can be regarded as part of the present invention and used to understand the meaning of some technical features or parameters.

Claims

1. A welding process optimization method for 7Ni steel processing, characterized in that: The method comprises: P1. Detect and process the electrode feature vector, collect the welding base material characteristic parameters, obtain the welding process parameters, and obtain the corresponding welding process method and welding process performance composed of weld defect characteristics; P2. Based on a three-way feature vector fusion method, high-dimensional fusion of the welding rod feature vector, the parent material characteristic parameters, and the welding process parameters is used to obtain welding process optimization features; P3. Constructing a welding process optimization model based on the welding process optimization characteristics and the welding process performance; the welding process optimization model is constructed by inputting the welding process optimization characteristics and the welding process performance into a multiple activation function neural network model improved based on welding process parameters, and the multiple activation function neural network model improved based on welding process parameters adopts a dual activation function input method for model construction, and the dual activation function calculation formula is: ； Where, is the dual activation function value, is the input welding process optimization feature, V is the welding speed, T is the welding environment temperature, and H is the welding environment humidity; P4. Output new welding process performance for the 7Ni steel to be subsequently welded according to the welding process optimization model.

2. The welding process optimization method for 7Ni steel processing according to claim 1, characterized in that: The welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters. The welding rod optical image feature is obtained by acquiring and processing the feature with a CCD camera.

3. The welding process optimization method for 7Ni steel processing according to claim 2, characterized in that: The basic parameters of the welding rod include the welding rod diameter, the coating radius and the welding rod eccentricity, wherein the welding rod eccentricity measurement formula is: ； Where E is the eccentricity of the welding rod, The distance from the center of the welding rod core to the outside of the welding rod, is the distance between the center of the electrode coating and the outside of the electrode, D is the coating radius, and the electrode feature vector is constructed by combining the optical image features of the electrode with the electrode diameter, coating radius, and electrode eccentricity.

4. The welding process optimization method for 7Ni steel processing according to claim 3, characterized in that: The parent material characteristic parameters include the content, carbon content and density of the reversed austenite of 7Ni steel, and the welding process parameters include welding speed, welding environment temperature and welding environment humidity.

5. The welding process optimization method for 7Ni steel processing according to claim 4, characterized in that: The high-dimensional fusion method based on the tripartite feature vector fusion method is obtained by longitudinally splicing the feature vectors of the welding rod feature vector, the parent material characteristic parameters and the welding process parameters and then supplementing the matrix vector values.

6. A welding process optimization system for 7Ni steel processing, the system implementing the method according to claim 5, comprising a welding rod feature processing generation module, a welding base material characteristic parameter acquisition module, a welding process parameter acquisition module, a welding process optimization feature generation module, a welding process optimization model construction module, and a welding process and performance control module, characterized in that: The welding rod feature processing and generating module is used to detect and process the welding rod feature vector; The welding base material characteristic parameter acquisition module is used to collect the welding base material characteristic parameters; The welding process parameter acquisition module is used to obtain welding process parameters; The welding process optimization feature generation module is configured to obtain the welding process optimization feature through high-dimensional fusion of the welding rod feature vector, the base material characteristic parameters and the welding process parameters through a three-way feature vector fusion method; The welding process optimization model construction module is used to construct a welding process optimization model based on the welding process optimization characteristics and welding process performance; The welding process performance is composed of the corresponding welding process method and weld defect characteristics; the welding process optimization model is constructed by inputting the welding process optimization characteristics and the welding process performance into a multiple activation function neural network model based on welding process parameter improvement. The multiple activation function neural network model based on welding process parameter improvement adopts a dual activation function input method for model construction, and the dual activation function calculation formula is: ； Where, is the dual activation function value, is the input welding process optimization feature, V is the welding speed, T is the welding environment temperature, and H is the welding environment humidity; The welding process and performance control module is configured to obtain welding process performance and obtain new welding process performance of the 7Ni steel to be welded according to the welding process optimization model.

7. A welding process optimization system for 7Ni steel processing according to claim 6, characterized in that: The welding rod feature vector is obtained by fusing the welding rod optical image feature and the welding rod basic parameters. The welding rod optical image feature is obtained by acquiring and processing the feature with a CCD camera.

Images