A method and device for predicting and controlling the welding angular distortion of a butt joint

Abstract

The present invention relates to a method and device for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm. The method includes: establishing a finite element model of a MAG welding butt joint and correcting the model through experiments; predicting the welding angular deformation; controlling the welding angular deformation. Among them, the prediction and control of the welding angular deformation respectively include the following steps: selecting design variables and statistically determining the magnitude of the angular deformation based on orthogonal experiments; determining the neural network model structure based on the design variables and the magnitude of the angular deformation and training it; based on the GA genetic algorithm, using the magnitude of the angular deformation as the individual fitness to optimize the neural network model; predicting or controlling the deformation based on the optimized model. Compared with the prior art, the present invention has the advantages of being able to quickly obtain the deformation amount of the butt joint and controlling the deformation, etc.

Description

Technical Field

[0001] The present invention relates to the field of welding deformation control, and in particular, to a method and device for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm. Background Technique

[0002] Due to the non-uniformity of the welding heat input, welding deformation will occur after welding. Welding deformation has a great impact on the use safety and installation configuration of workpieces. Therefore, the prediction and control of welding deformation have great engineering significance. However, welding deformation is related to multiple factors, and usually, experienced engineers need to adjust different welding process parameters and continuously optimize the relevant process parameters. This method relying on personnel usually leads to an increase in production costs and a decrease in the efficiency of process design.

[0003] Using finite element simulation can solve the problems existing in relying on empirical judgment to a certain extent. However, finite element simulation has problems such as long calculation time for large or complex workpieces, so it is not conducive to the situation of large-scale welding production of different types.

[0004] An artificial neural network is a computational model that uses engineering techniques to imitate the human brain neural network. It has a powerful self-adaptive learning ability and can establish a highly non-linear relationship between input and output to form a mapping. Currently, no technology has introduced an artificial neural network into the control and prediction of welding deformation. Therefore, proposing a fast and accurate prediction and control of welding deformation based on an artificial neural network is one of the urgent problems to be solved in current engineering. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm, predict the welding angular deformation of the butt joint under different process parameters, and thus provide a fast and accurate guidance for the deformation control of the butt joint.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm includes the following steps:

[0008] S1: Establish a finite element model of a MAG welding butt joint and correct the model through experiments;

[0009] S2: Welding angular deformation prediction:

[0010] S21: Select process parameters as the first design variable, select several levels within the value range of the first design variable, design an orthogonal test table, implement the orthogonal test plan, perform finite element simulations according to the parameters of each test group, and count the magnitude of the first angular deformation on the central section of the model after the finite element simulation;

[0011] S22: Use the first design variable as the input parameter of the prediction neural network model and the magnitude of the first angular deformation as the output parameter to construct a training set and a test set;

[0012] S23: Determine the structure of the prediction neural network model based on the input and output parameters of the prediction neural network, and perform training and testing until the mean square error between the neural network predicted output value and the actual value of the test samples is within the pre-configured allowable range;

[0013] S24: Use the first design variable to form population individuals, use the magnitude of the first angular deformation as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculations, obtain the optimal solution of the fitness, optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S23, reconstruct the prediction neural network model and perform the optimization solution of the GA genetic algorithm;

[0014] S25: Perform welding angular deformation prediction based on the optimized prediction neural network model;

[0015] S3: Welding angular deformation control:

[0016] S31: Use the process parameters selected in S21 and multiples of the magnitude of the first angular deformation as the second design variable, select several levels within the value range of the second design variable, design an orthogonal test table, implement the orthogonal test plan, perform finite element simulations according to the parameters of each experimental group, and count the magnitude of the second angular deformation after applying reverse deformation on the central section of the model after the finite element simulation;

[0017] S32: Use the second design variable as the input parameter of the control neural network and the magnitude of the second angular deformation as the output parameter to construct a training set and a test set;

[0018] S33: Determine the structure of the control neural network model based on the input and output parameters of the control neural network, and perform training and testing until the mean square error between the neural network predicted output value and the actual value of the test samples is within the pre-configured allowable range;

[0019] S34: Use the second set of design variables to form population individuals, use the second angular deformation magnitude as the individual fitness, and employ the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculations to obtain the optimal solution of the fitness, thereby optimizing the neural network. Then, compare the optimized prediction results with the experimental results. If the difference exceeds the threshold, return to S33, reconstruct the control neural network model, and perform optimization and solution using the GA genetic algorithm;

[0020] S35: Implement welding angular deformation control based on the optimized control neural network model.

[0021] The process parameters include the length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input.

[0022] The value ranges of the design variables are specifically as follows: the value range of the length L of the welding plate is 50 - 400 mm, the value range of the thickness H of the welding plate is 7 - 16 mm, the value range of the welding speed S is 4 - 16 mm / s, the value range of the preheating temperature T is 20 - 300 °C, the welding heat input is the product of the current and voltage, where the value range of the current I is 240 - 320 A, the value range of the voltage U is 24 - 32 V; the value range of the multiple C of the first angular deformation magnitude A is 0.9 - 1.3.

[0023] The structure of the prediction neural network model is as follows:

[0024] Define the number of neurons in the input layer as n, and the number of neurons in the output layer as m. Here, the number of neurons in the input layer and the output layer are respectively the number of input variables and output variables, that is, n = 6, m = 1;

[0025] Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer as 9, and the number of neurons in the second hidden layer as 12;

[0026] Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

[0027] The structure of the control neural network model is as follows:

[0028] Define the number of neurons in the input layer as p, and the number of neurons in the output layer as q. Here, the number of neurons in the input layer and the output layer are respectively the number of input variables and output variables, that is, p = 7, q = 1;

[0029] Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer as 13, and the number of neurons in the second hidden layer as 15;

[0030] Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

[0031] A butt joint welding angular distortion prediction and control device based on BP neural network and GA genetic algorithm, comprising:

[0032] A butt joint finite element model construction and correction module, used to establish a finite element model of the MAG welded butt joint and correct the model through experiments;

[0033] A welding angular distortion prediction module, used to perform the following steps:

[0034] S21: Select process parameters as the first design variables, select several levels within the value range of the first design variables, design an orthogonal test table, implement the orthogonal test plan, perform finite element simulations according to the parameters of each test group, and count the magnitude of the first angular distortion on the central section of the model after finite element simulation;

[0035] S22: Use the first design variables as the input parameters of the prediction neural network model and the magnitude of the first angular distortion as the output parameters to construct a training set and a test set;

[0036] S23: Determine the structure of the prediction neural network model based on the input parameters and output parameters of the prediction neural network, and perform training and testing until the mean square error between the neural network prediction output value and the actual value of the test samples is within the pre-configured allowable range;

[0037] S24: Use the first design variables to form population individuals, use the magnitude of the first angular distortion as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculation, obtain the optimal solution of the fitness, optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S23, reconstruct the prediction neural network model and perform the optimization solution of the GA genetic algorithm;

[0038] S25: Perform welding angular distortion prediction based on the optimized prediction neural network model;

[0039] A welding angular distortion control module, used to perform the following steps:

[0040] S31: Use the process parameters selected in S21 and multiples of the magnitude of the first angular distortion as the second design variables, select several levels within the value range of the second design variables, design an orthogonal test table, implement the orthogonal test plan, perform finite element simulations according to the parameters of each experimental group, and count the magnitude of the second angular distortion after applying anti-deformation on the central section of the model after finite element simulation;

[0041] S32: Use the second design variable as the input parameter for controlling the neural network, and use the second angular deformation magnitude as the output parameter to construct a training set and a test set;

[0042] S33: Determine the structure of the control neural network model based on the input and output parameters of the control neural network, and perform training and testing until the mean square error between the neural network predicted output value and the actual value of the test samples is within the pre-configured allowable range;

[0043] S34: Use the second design variable to form population individuals, use the second angular deformation magnitude as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculation to obtain the optimal solution of the fitness, thereby optimizing the neural network, and compare the predicted results after optimization with the experimental results. If the result difference exceeds the threshold, return to S33, reconstruct the control neural network model and perform the optimization solution of the GA genetic algorithm;

[0044] S35: Implement welding angular deformation control based on the optimized control neural network model.

[0045] The process parameters include the length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input.

[0046] The value ranges of the design variables are specifically as follows: the value range of the length L of the welding plate is 50 - 400 mm, the value range of the thickness H of the welding plate is 7 - 16 mm, the value range of the welding speed S is 4 - 16 mm / s, the value range of the preheating temperature T is 20 - 300 °C, the welding heat input is the product of current and voltage, where the value range of the current I is 240 - 320 A, the value range of the voltage U is 24 - 32 V; the value range of the multiple C of the first angular deformation magnitude A is 0.9 - 1.3.

[0047] The structure of the prediction neural network model is:

[0048] Define the number of neurons in the input layer as n, and the number of neurons in the output layer as m. Among them, the number of neurons in the input layer and the output layer are respectively the number of input variables and output variables, that is, n = 6, m = 1;

[0049] Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer is 9, and the number of neurons in the second hidden layer is 12;

[0050] Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

[0051] The structure of the control neural network model is:

[0052] Define the number of neurons in the input layer as p and the number of neurons in the output layer as q. Here, the numbers of neurons in the input layer and the output layer are respectively the numbers of input variables and output variables, that is, p = 7 and q = 1.

[0053] Define that the number of hidden layers in the neural network model is two. The number of neurons in the first hidden layer is 13, and the number of neurons in the second hidden layer is 15.

[0054] Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer. Select the adam function as the training function for backpropagating the error.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] Through the finite element simulation of the welding process of the butt joint by the BP neural network model and the genetic algorithm in the present invention, the magnitude of the angular deformation of the butt joint is obtained. The length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input, which have a greater influence on the angular deformation, are selected as the input quantities. The orthogonal test analysis is used to combine the welding processes and the magnitude of the angular deformation is used as the output quantity. The BP neural network is trained and the neural network model is optimized in combination with the genetic algorithm. Finally, a neural network with a higher accuracy of predicting the angular deformation by finite element simulation is obtained. The angular deformation multiple, the length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input are used as the input quantities again. The orthogonal test analysis is used to combine the welding processes again to obtain the magnitude of the angular deformation after applying the elastic reverse deformation. The orthogonal test analysis is used to combine the welding processes and the magnitude of the angular deformation after applying the elastic reverse deformation is used as the output quantity. The BP neural network is trained and the neural network model is optimized in combination with the genetic algorithm. Finally, a neural network with a higher accuracy of predicting the angular deformation after applying the elastic reverse deformation by finite element simulation is obtained. By using this algorithm model, the magnitude of the angular deformation of the butt joint can be quickly obtained according to different welding process parameters, meeting the changing welding scenarios, greatly simplifying the adjustment process of welding parameters, and improving the welding production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is the flowchart of the method of the present invention;

[0058] Figure 2 It is the schematic diagram of the butt joint model in an embodiment;

[0059] Figure 3 It is the schematic diagram of the prediction neural network;

[0060] Figure 4 It is the schematic diagram of the butt joint model after applying the elastic reverse deformation in an embodiment;

[0061] Figure 5Schematic diagram of a control neural network for applying elastic reverse deformation. Specific implementation manner

[0062] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and provides a detailed implementation manner and a specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0063] This embodiment provides a method for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm, as Figure 1 shown, including the following steps:

[0064] S1: Establish a finite element model of a MAG welding butt joint and correct the model through experiments.

[0065] The welding butt joint material used in this embodiment is Q960E high-strength steel. The welding process parameters are shown in Table 1, and the butt joint model is shown in Figure 2 , and the finite element model is corrected through welding experiments under the same process parameters.

[0066] Table 1 Welding process parameters

[0067]

[0068] S2: Welding angular deformation prediction:

[0069] S21: Select process parameters: the length L of the welding plate, the thickness H of the welding plate, the width B of the welding plate, the welding speed S, the preheating temperature T, and the welding heat input E as the first design variables. Select several levels within the value ranges of the first design variables, design an orthogonal test table, implement the orthogonal test plan, perform finite element simulations according to the parameters of each test group, and count the magnitude A of the first angular deformation on the central cross-section of the model after the finite element simulation.

[0070] In this embodiment, the value ranges of the first design variables are as follows: the value range of the length L of the welding plate is 50 - 400 mm, the value range of the thickness H of the welding plate is 7 - 16 mm, the value range of the welding speed S is 4 - 16 mm / s, the value range of the preheating temperature T is 20 - 300 °C, the welding heat input is the product of the current and voltage, where the value range of the current I is 240 - 320 A, and the value range of the voltage U is 24 - 32 V.

[0071] S22: Use the first design variables as the input parameters of the prediction neural network model, and the magnitude A of the first angular deformation as the output parameter. Normalize the input parameters and output parameters and randomly divide them into a training set and a test set.

[0072] To eliminate the differences in the order of magnitude between data in each dimension, the data of each input and output parameter is normalized to the interval (0, 1) using Equation 1-1.

[0073]

[0074] In the formula: x k is the data to be normalized; x max is the maximum value in the data; x min is the minimum value in the data.

[0075] S23: Determine the structure of the prediction neural network model based on the input parameters and output parameters of the prediction neural network, and perform training and testing until the mean square error between the neural network prediction output value and the actual value of the test sample is within the pre-configured allowable range.

[0076] In this embodiment, the structure of the prediction neural network model is as Figure 3 shown, specifically:

[0077] Define the number of neurons in the input layer as n, and the number of neurons in the output layer as m. Among them, the number of neurons in the input layer and the output layer are the numbers of input variables and output variables respectively, that is, n = 6, m = 1;

[0078] Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer is 9, and the number of neurons in the second hidden layer is 12;

[0079] Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating errors.

[0080] Among them, the relu transfer function is:

[0081] f(x) = max(0, x) (1-2)

[0082] S24: Use the welding plate length L, welding plate thickness H, welding plate width B, welding speed S, preheating temperature T, and welding heat input E to form population individuals, use the first angular deformation magnitude A as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculation, obtain the optimal solution of the fitness, optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S23, reconstruct the prediction neural network model and perform the optimization solution of the GA genetic algorithm.

[0083] In this embodiment, the GA genetic algorithm is encoded in binary real number form, uses the angular deformation magnitude A as the individual fitness, calculates the optimal fitness value in the population, and obtains better weights and thresholds by performing selection, crossover, and mutation on the population within the maximum genetic range.

[0084] When using the genetic algorithm for optimization, the population size is 70, the number of evolutions is 100, the crossover probability is 0.65, and the mutation probability is 0.1.

[0085] S25: Predict the welding angular deformation based on the optimized prediction neural network model.

[0086] Use the training set to train the prediction neural network optimized by the genetic algorithm. When the mean square error MSE of the test set is less than 0.05, the model training is completed; otherwise, continue to modify the network model parameters until the training standard is met.

[0087] S3: Welding angular deformation control:

[0088] S31: Apply elastic back-deformation to the welded joint, as Figure 4 shown. Take the process parameters selected in S21: the length L of the welding plate, the thickness H of the welding plate, the width B of the welding plate, the welding speed S, the preheating temperature T, and the welding heat input E, as well as the multiple C of the first angular deformation size A as the second design variables. Select several levels within the value range of the second design variables, design an orthogonal test table, implement the orthogonal test scheme, conduct finite element simulations according to the parameters of each experimental group, and count the second angular deformation size F after applying back-deformation on the central cross-section of the model after finite element simulation.

[0089] In this embodiment, the value range of the process parameters is the same as that in S21, and the value range of the multiple C of the first angular deformation size A is 0.9 - 1.3.

[0090] S32: Use the second design variables as the input parameters of the control neural network, and the second angular deformation size F as the output parameter. Normalize the input parameters and output parameters and randomly divide them into a training set and a test set.

[0091] S33: Determine the structure of the control neural network model based on the input parameters and output parameters of the control neural network, and conduct training and testing until the mean square error between the neural network prediction output value and the actual value of the test samples is within the pre-configured allowable range.

[0092] In this embodiment, the structure of the control neural network model is as Figure 5 shown, specifically:

[0093] Define the number of neurons in the input layer as p, and the number of neurons in the output layer as q. Among them, the number of neurons in the input layer and the output layer are respectively the number of input variables and output variables, that is, p = 7, q = 1;

[0094] Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer is 13, and the number of neurons in the second hidden layer is 15;

[0095] The relu function is selected as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and the adam function is selected as the training function for backpropagating the error.

[0096] S34: Use the multiple C of the first angular deformation size A, the length L of the welding plate, the thickness H of the welding plate, the width B of the welding plate, the welding speed S, the preheating temperature T, and the welding heat input E to form population individuals. Use the second angular deformation size F as the individual fitness. Use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculation to obtain the optimal solution of the fitness, so as to optimize the neural network. Then compare the predicted result after optimization with the experimental result. If the result difference exceeds the threshold, return to S33, reconstruct the control neural network model and perform the optimization solution of the GA genetic algorithm.

[0097] In this embodiment, the GA genetic algorithm is encoded in the form of binary real numbers, uses the angular deformation size F as the individual fitness, calculates the optimal fitness value in the population, and obtains better weights and thresholds by performing selection, crossover, and mutation on the population within the maximum genetic range.

[0098] When using the genetic algorithm for optimization, the population size is 80, the number of evolutions is 120, the crossover probability is 0.65, and the mutation probability is 0.1.

[0099] S35: Implement welding angular deformation control based on the optimized control neural network model.

[0100] Use the training set to learn the control neural network optimized by the genetic algorithm. When the mean square error MSE of the test set is less than 0.05, the model training is completed; otherwise, continue to modify the network model parameters until the training standard is met. The trained control neural network model can be used to guide welding angular deformation control.

[0101] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. A method for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm, characterized in that, It includes the following steps: S1: Establish a finite element model of the MAG welding butt joint and correct the model through experiments; S2: Welding angular distortion prediction: S21: Select process parameters as the first design variables, select several levels within the value range of the first design variables, design an orthogonal test table, implement the orthogonal test plan, conduct finite element simulations according to the parameters of each test group, and count the magnitude of the first angular distortion on the central section of the model after the finite element simulation; S22: Use the first design variables as the input parameters of the prediction neural network model and the magnitude of the first angular distortion as the output parameter to construct a training set and a test set; S23: Determine the structure of the prediction neural network model based on the input and output parameters of the prediction neural network, and conduct training and testing until the mean square error between the neural network prediction output value and the actual value of the test samples is within the pre-configured allowable range; S24: Use the first design variables to form population individuals, use the magnitude of the first angular distortion as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and conduct fitness calculations, obtain the optimal solution of the fitness, optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S23, reconstruct the prediction neural network model and conduct the optimization solution of the GA genetic algorithm; S25: Conduct welding angular distortion prediction based on the optimized prediction neural network model; S3: Welding angular distortion control: S31: Use the process parameters selected in S21 and multiples of the magnitude of the first angular distortion as the second design variables, select several levels within the value range of the second design variables, design an orthogonal test table, implement the orthogonal test plan, conduct finite element simulations according to the parameters of each experimental group, and count the magnitude of the second angular distortion after applying reverse deformation on the central section of the model after the finite element simulation; S32: Use the second design variables as the input parameters of the control neural network and the magnitude of the second angular distortion as the output parameter to construct a training set and a test set; S33: Determine the structure of the control neural network model based on the input and output parameters of the control neural network, and conduct training and testing until the mean square error between the neural network prediction output value and the actual value of the test samples is within the pre-configured allowable range; S34: Use the second design variables to form population individuals, use the magnitude of the second angular distortion as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and conduct fitness calculations, obtain the optimal solution of the fitness, thereby optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S33, reconstruct the control neural network model and conduct the optimization solution of the GA genetic algorithm; S35: Achieve welding angular distortion control based on the optimized control neural network model.

2. A prediction and control method for welding angular distortion of butt joints based on BP neural network and GA genetic algorithm according to claim 1, characterized in that, The process parameters include the length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input.

3. A method for predicting and controlling the welding angular distortion of a butt joint based on a BP neural network and a GA genetic algorithm according to claim 2, characterized in that, The specific value ranges of the design variables are as follows: the value range of the length L of the welding plate is 50 - 400 mm, the value range of the thickness H of the welding plate is 7 - 16 mm, the value range of the welding speed S is 4 - 16 mm / s, the value range of the preheating temperature T is 20 - 300 °C, the welding heat input is the product of current and voltage, where the value range of the current I is 240 - 320 A, and the value range of the voltage U is 24 - 32 V; the value range of the multiple C of the magnitude A of the first angular deformation is 0.9 - 1.

3.

4. A method for predicting and controlling the welding angular deformation of a butt joint based on a BP neural network and a GA genetic algorithm according to claim 2, characterized in that, The structure of the prediction neural network model is as follows: Define the number of neurons in the input layer as n, and the number of neurons in the output layer as m. Among them, the numbers of neurons in the input layer and the output layer are the numbers of input variables and output variables respectively, that is, n = 6, m = 1; Define that the number of hidden layers of the neural network model is two layers, the number of neurons in the first hidden layer is 9, and the number of neurons in the second hidden layer is 12; Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

5. A method for predicting and controlling the welding angular distortion of a butt joint based on a BP neural network and a GA genetic algorithm according to claim 2, characterized in that, The structure of the control neural network model is as follows: Define the number of neurons in the input layer as p, and the number of neurons in the output layer as q. Among them, the numbers of neurons in the input layer and the output layer are the numbers of input variables and output variables respectively, that is, p = 7, q = 1; Define that the number of hidden layers of the neural network model is two layers, the number of neurons in the first hidden layer is 13, and the number of neurons in the second hidden layer is 15; Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

6. A docking joint welding angular distortion prediction and control device based on a BP neural network and a GA genetic algorithm, characterized in that, It includes: A butt joint finite element model construction and correction module, which is used to establish a finite element model of the MAG welding butt joint and correct the model through experiments; A welding angular deformation prediction module, which is used to perform the following steps: S21: Select process parameters as the first design variables, select several levels within the value ranges of the first design variables, design an orthogonal test table, implement the orthogonal test scheme, conduct finite element simulations according to the parameters of each test group, and count the magnitude of the first angular deformation on the central section of the model after the finite element simulations; S22: Use the first design variables as the input parameters of the prediction neural network model and the magnitude of the first angular deformation as the output parameter to construct a training set and a test set; S23: Determine the structure of the prediction neural network model based on the input parameters and output parameters of the prediction neural network, and conduct training and testing until the mean square error between the neural network prediction output value and the actual value of the test samples is within the pre-configured allowable range; S24: Use the first design variables to form population individuals, use the magnitude of the first angular deformation as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculations, obtain the optimal solution of the fitness, optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S23, reconstruct the prediction neural network model and perform the optimization solution of the GA genetic algorithm; S25: Perform welding angular deformation prediction based on the optimized prediction neural network model; The welding angular deformation control module is used to perform the following steps: S31: Using the process parameters selected in S21 and the multiple of the first angular deformation magnitude as the second design variable, select several levels within the value range of the second design variable, design an orthogonal experiment table, implement the orthogonal experiment plan, perform finite element simulation according to the parameters of each experimental group, and count the second angular deformation magnitude after applying reverse deformation on the central section of the model after finite element simulation; S32: Using the second design variable as the input parameter of the control neural network and the second angular deformation magnitude as the output parameter, construct a training set and a test set; S33: Determine the structure of the control neural network model based on the input and output parameters of the control neural network, and perform training and testing until the mean square error between the neural network predicted output value and the actual value of the test samples is within the pre-configured allowable range; S34: Using the second design variable to form population individuals, using the second angular deformation magnitude as the individual fitness, use the GA genetic algorithm to obtain the initial weights and thresholds and perform fitness calculation, obtain the optimal solution of the fitness, thereby optimize the neural network, and compare the optimized prediction results with the experimental results. If the result difference exceeds the threshold, return to S33, reconstruct the control neural network model and perform the optimization solution of the GA genetic algorithm; S35: Realize welding angular deformation control based on the optimized control neural network model.

7. A butt joint welding angular distortion prediction and control device based on a BP neural network and a GA genetic algorithm according to claim 6, characterized in that, The process parameters include the length of the welding plate, the thickness of the welding plate, the width of the welding plate, the welding speed, the preheating temperature, and the welding heat input.

8. A butt joint welding angular distortion prediction and control device based on a BP neural network and a GA genetic algorithm according to claim 7, characterized in that, The specific value range of the design variable is as follows: the value range of the length L of the welding plate is 50 - 400 mm, the value range of the thickness H of the welding plate is 7 - 16 mm, the value range of the welding speed S is 4 - 16 mm / s, the value range of the preheating temperature T is 20 - 300 °C, the welding heat input is the product of current and voltage, where the value range of the current I is 240 - 320 A, the value range of the voltage U is 24 - 32 V; the value range of the multiple C of the first angular deformation magnitude A is 0.9 - 1.

3.

9. A butt joint welding angular distortion prediction and control device based on a BP neural network and a GA genetic algorithm according to claim 7, characterized in that, The structure of the prediction neural network model is: Define the number of neurons in the input layer as n, and the number of neurons in the output layer as m. Among them, the number of neurons in the input layer and the output layer are the number of input variables and output variables respectively, that is, n = 6, m = 1; Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer is 9, and the number of neurons in the second hidden layer is 12; Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating the error.

10. A butt joint welding angular distortion prediction and control device based on a BP neural network and a GA genetic algorithm according to claim 7, characterized in that, The structure of the control neural network model is: Define the number of neurons in the input layer as p, and the number of neurons in the output layer as q. Among them, the number of neurons in the input layer and the output layer are the number of input variables and output variables respectively, that is, p = 7, q = 1; Define the number of hidden layers of the neural network model as two layers, the number of neurons in the first hidden layer is 13, and the number of neurons in the second hidden layer is 15; Select the relu function as the transfer function between the input layer and the hidden layer, and between the hidden layer and the output layer, and select the adam function as the training function for backpropagating errors.

Images