Neural network model training and process parameter optimization method for electron beam welding process

Abstract

The invention relates to the technical field of electron beam welding, in particular to a neural network model training and process parameter optimization method for an electron beam welding process, and the method comprises the steps: obtaining the historical experiment data of the electron beam welding process; a data set is generated according to historical experimental data, a pre-constructed neural network model is trained through the data set, model parameters of the neural network model are updated based on error back propagation in the training process, and the input of the neural network model is welding parameters and the thickness of a base plate; the output of the neural network model is weld quality, welding energy consumption and machining allowance; and the trained neural network model is used for monitoring the weld quality, the welding energy consumption and the machining allowance of the electron beam welding process, and welding parameters and the thickness of the base plate are optimized according to the weld quality, the welding energy consumption and the machining allowance. Therefore, the problems of unstable welding quality, high welding energy consumption, high subsequent machining amount, high production cost, low production efficiency and the like in the prior art are solved.

Description

Technical Field

[0001] This application relates to the technical field of electron beam welding, and particularly relates to a method for training a neural network model of an electron beam welding process and optimizing process parameters. Background Art

[0002] Electron beam welding is a high-precision and high-efficiency welding technology, which is widely used in fields such as aerospace, automotive manufacturing, precision instruments, and nuclear industry. However, nail-shaped defects are likely to occur during the electron beam welding process, and usually pads are required to lead out these defects.

[0003] In the related art, the determination of welding parameters mainly relies on experiments and experience, lacking a systematic multi-parameter collaborative optimization method, which easily leads to a too thick pad increasing the subsequent machining amount or a too thin pad resulting in ineffective defect leading-out. This not only causes unstable welding quality, but also increases the subsequent machining amount and cost, reduces production efficiency, and due to the unoptimized welding parameters, the voltage and current of the electron beam are often too high, resulting in increased energy consumption. Summary of the Invention

[0004] This application provides a method for training a neural network model of an electron beam welding process and optimizing process parameters to solve the problems of unstable welding quality, high welding energy consumption, subsequent machining amount and production cost, and low production efficiency in the related art.

[0005] The first aspect of the embodiments of this application provides a method for training a neural network model of an electron beam welding process, including the following steps: obtaining historical experimental data of the electron beam welding process, where the historical experimental data includes welding parameters, weld quality, welding energy consumption, pad thickness, and machining allowance of the pad thickness of the electron beam welding process; generating a data set according to the historical experimental data, and using the data set to train a pre-constructed neural network model. During the training process, the model parameters of the neural network model are updated based on error backpropagation, where the input of the neural network model is the welding parameter and the pad thickness, and the output of the neural network model is the weld quality, welding energy consumption, and machining allowance; using the trained neural network model to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimizing the welding parameters and the pad thickness according to the weld quality, welding energy consumption, and machining allowance.

[0006] Optionally, the neural network model includes a first propagation layer, a second propagation layer, and a loss function. The first propagation layer generates a first prediction result based on the input data of the neural network model. The input data includes multiple welding parameters, backing plate thicknesses, and training errors in the dataset. The second propagation layer generates a second prediction result based on the first prediction result. The loss function calculates the error based on the second prediction result and the true value in the dataset, and reversely transmits the error to the input of the neural network model, and updates the model parameters of the neural network model according to the error until the loss function converges. The true value in the dataset includes weld quality, welding energy consumption, and machining allowance.

[0007] Optionally, the propagation layer structures of the first propagation layer and the second propagation layer are the same. The propagation layer structure includes: a hidden layer, a normalization layer, a non-linear activation function, and a regularization function. The hidden layer is used to extract and transform the input data of the neural network model. The normalization layer is used to normalize the output data of the hidden layer. The non-linear activation function is used to introduce non-linear characteristics to the normalized data. The regularization function is used to randomly lose some neurons during the training process.

[0008] Optionally, training the pre-constructed neural network model using the dataset includes: dividing the data in the dataset into a training set, a test set, and a validation set according to a target ratio; training the neural network model using the training set and the loss function, and validating the loss value or prediction accuracy of the trained neural network model using the validation set until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and then stop training the neural network model; testing the trained neural network model using the test set.

[0009] Optionally, updating the model parameters of the neural network model according to the error includes: obtaining the initial learning rate of the neural network model; calculating the model gradient of the neural network model based on the error, updating the initial learning rate based on the model gradient, and updating the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

[0010] An embodiment of the second aspect of the present application provides a method for optimizing electron beam welding process parameters, including the following steps: obtaining the current monitoring data of the electron beam welding process; identifying the welding parameters and backing plate thickness in the current monitoring data; inputting the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model, and the output of the neural network model is weld quality, welding energy consumption, and machining allowance, where the neural network model is trained according to the neural network model training method for the electron beam welding process in the above embodiment; optimizing the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0011] An embodiment of the third aspect of the present application provides a neural network model training device for an electron beam welding process, including: a first acquisition module, configured to acquire historical experimental data of the electron beam welding process, where the historical experimental data includes welding parameters, weld quality, welding energy consumption, backing plate thickness, and machining allowance of the backing plate thickness; a training module, configured to generate a data set based on the historical experimental data, and use the data set to train a pre-constructed neural network model. During the training process, the model parameters of the neural network model are updated based on error backpropagation, where the input of the neural network model is the welding parameter and the backing plate thickness, and the output of the neural network model is the weld quality, welding energy consumption, and machining allowance; a monitoring module, configured to use the trained neural network model to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimize the welding parameters and the backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0012] Optionally, the neural network model includes a first propagation layer, a second propagation layer, and a loss function. The first propagation layer generates a first prediction result based on the input data of the neural network model, where the input data includes multiple welding parameters, backing plate thickness, and training errors in the data set; the second propagation layer generates a second prediction result based on the first prediction result; the loss function calculates an error based on the second prediction result and the true value in the data set, and reversely transmits the error to the input of the neural network model, and updates the model parameters of the neural network model according to the error until the loss function converges, where the true value in the data set includes the weld quality, welding energy consumption, and machining allowance.

[0013] Optionally, the propagation layer structures of the first propagation layer and the second propagation layer are the same, where the propagation layer structure includes: a hidden layer, a normalization layer, a non-linear activation function, and a regularization function. The hidden layer is configured to perform feature extraction and transformation on the input data of the neural network model; the normalization layer is configured to perform normalization processing on the output data of the hidden layer; the non-linear activation function is configured to introduce non-linear characteristics to the normalized data; the regularization function is configured to randomly lose some neurons during the training process.

[0014] Optionally, the training module is further configured to: divide the data in the data set into a training set, a test set, and a validation set according to a target ratio; use the training set and the loss function to train the neural network model, and use the validation set to verify the loss value or prediction accuracy of the trained neural network model until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and stop training the neural network model; use the test set to test the trained neural network model.

[0015] Optionally, update the model parameters of the neural network model according to the error, including: obtaining the initial learning rate of the neural network model; calculating the model gradient of the neural network model based on the error, updating the initial learning rate based on the model gradient, and updating the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

[0016] An embodiment of the fourth aspect of the present application provides an electronic welding process parameter optimization device, including: a second acquisition module, configured to acquire current monitoring data of an electron beam welding process; an identification module, configured to identify welding parameters and backing plate thickness in the current monitoring data; an input module, configured to input the welding parameters and backing plate thickness in the current monitoring data into a trained neural network model, and the output of the neural network model is weld quality, welding energy consumption, and machining allowance, where the neural network model is trained by the neural network model training device for the electron beam welding process in the above embodiment; an optimization module, configured to optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0017] An embodiment of the fifth aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to execute the neural network model training method for the electron beam welding process in the above embodiment, or the electronic welding process parameter optimization method.

[0018] An embodiment of the fourth aspect of the present application provides a computer-readable storage medium, on which a computer program or instruction is stored, and the computer program or instruction is executed by a processor to execute the neural network model training method for the electron beam welding process in the above embodiment, or the electronic welding process parameter optimization method.

[0019] Therefore, the present application has at least the following beneficial effects:

[0020] Embodiments of the present application can generate a data set according to historical experimental data of an electron beam welding process, use the data set to train a pre-constructed neural network model, and update the model parameters of the neural network model based on error backpropagation during the training process to improve the accuracy of the model output. Furthermore, use the trained neural network model to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance to improve the welding quality, reduce the welding energy consumption, reasonably control the backing plate thickness, reduce the subsequent machining amount, and reduce the production cost. By automatically adjusting the welding process parameters and backing plate thickness, the production efficiency is improved, thereby meeting the requirements of large-scale production. Thus, the technical problems of unstable welding quality, high welding energy consumption, subsequent machining amount, and production cost, and low production efficiency in the related art are solved.

[0021] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present application. Description of the Drawings

[0022] The above-mentioned and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of embodiments in conjunction with the drawings, where:

[0023] Figure 1 It is a flowchart of a method for training a neural network model for an electron beam welding process according to an embodiment of the present application;

[0024] Figure 2 It is a schematic structural diagram of a neural network model according to an embodiment of the present application;

[0025] Figure 3 It is a diagram of the training process of a neural network model according to an embodiment of the present application;

[0026] Figure 4 It is a flowchart of a method for optimizing electron beam welding process parameters according to an embodiment of the present application;

[0027] Figure 5 It is a schematic diagram of a device for training a neural network model of an electron beam welding process according to an embodiment of the present application;

[0028] Figure 6 It is a schematic diagram of a device for optimizing electron beam welding process parameters according to an embodiment of the present application;

[0029] Figure 7 It is a schematic structural diagram of an electronic device according to an embodiment of the present application. Detailed Description of the Embodiments

[0030] Embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application and should not be construed as limiting the present application.

[0031] The following describes the neural network model training and process parameter optimization method for the electron beam welding process of the embodiments of the present application. In view of the problems in the related art mentioned in the above background technology, the determination of welding parameters mainly relies on experiments and experience, lacking a systematic multi-parameter collaborative optimization method, which easily leads to too thick or too thin backing plates, unstable welding quality, increased energy consumption and subsequent machining volume, and increased production costs. The present application provides a neural network model training method for the electron beam welding process. In this method, a data set can be generated according to the historical experimental data of the electron beam welding process, and the pre-constructed neural network model is trained using the data set. During the training process, the model parameters of the neural network model are updated based on error backpropagation to improve the accuracy of the model output. Furthermore, the trained neural network model is used to monitor the weld quality, welding energy consumption and machining allowance of the electron beam welding process, and the welding parameters and backing plate thickness are optimized according to the weld quality, welding energy consumption and machining allowance, so as to improve the welding quality, reduce the welding energy consumption, reasonably control the backing plate thickness, reduce the subsequent machining volume, and reduce the production cost. By automatically adjusting the welding process parameters and backing plate thickness, the production efficiency is improved, thereby meeting the requirements of large-scale production. Thus, the problems of unstable welding quality, high welding energy consumption, subsequent machining volume and production cost, and low production efficiency in the related art are solved.

[0032] Specifically, Figure 1 FIG. is a schematic flow chart of a neural network model training method for an electron beam welding process provided by an embodiment of the present application.

[0033] As Figure 1 shown, the neural network model training method for the electron beam welding process includes the following steps:

[0034] In step S101, historical experimental data of the electron beam welding process is obtained, where the historical experimental data includes welding parameters, weld quality, welding energy consumption, backing plate thickness and machining allowance of the backing plate thickness of the electron beam welding process.

[0035] Among them, the weld parameters include acceleration voltage, electron beam current, welding speed, focus depth, etc., and the weld quality mainly refers to the presence or absence of nail tip defects or the size of nail tip defects; the welding energy consumption is the energy consumption per unit length.

[0036] It can be understood that the embodiments of the present application can obtain the historical experimental data of the electron beam welding process for subsequent generation of a data set to train the pre-constructed neural network model.

[0037] It should be noted that after obtaining the historical experimental data of the electron beam welding process, the data needs to be preprocessed to remove outliers.

[0038] In step S102, a data set is generated based on historical experimental data, and a pre-constructed neural network model is trained using the data set. During the training process, the model parameters of the neural network model are updated based on error backpropagation. The input of the neural network model is welding parameters and backing plate thickness, and the output of the neural network model is weld quality, welding energy consumption, and machining allowance.

[0039] It can be understood that the embodiments of the present application can generate a data set according to historical experimental data and train a pre-constructed neural network model using the data set. During the training process, the model parameters of the neural network model are updated based on error backpropagation, thereby improving the accuracy of the output data of the neural network model.

[0040] In the embodiments of the present application, the neural network model includes a first propagation layer, a second propagation layer, and a loss function. The first propagation layer generates a first prediction result based on the input data of the neural network model. The input data includes multiple welding parameters, backing plate thickness, and training errors in the data set. The second propagation layer generates a second prediction result based on the first prediction result. The loss function calculates the error based on the second prediction result and the true value in the data set, and reversely transmits the error to the input of the neural network model. The model parameters of the neural network model are updated according to the error until the loss function converges. The true value in the data set includes weld quality, welding energy consumption, and machining allowance.

[0041] The prediction result is the predicted weld quality and machining allowance output based on the input welding parameters, backing plate thickness, and training error. The loss function can be a mean square error loss function.

[0042] It can be understood that the neural network model of the embodiments of the present application includes a first propagation layer, a second propagation layer, and a loss function, where

[0043] the first propagation layer generates a first prediction result based on the input data of the neural network model;

[0044] the second propagation layer generates a second prediction result based on the first prediction result;

[0045] the loss function is used to calculate the error based on the second prediction result and the true value in the data set, and reversely transmit the error to the input of the neural network model, and update the input of the neural network model according to the error until the loss function converges.

[0046] In the embodiments of the present application, the propagation layer structures of the first propagation layer and the second propagation layer are the same. The propagation layer structure includes: a hidden layer, a normalization layer, a non-linear activation function, and a regularization function. The hidden layer is used to extract and transform the input data of the neural network model; the normalization layer is used to normalize the output data of the hidden layer; the non-linear activation function is used to introduce non-linear characteristics into the normalized data; the regularization function is used to randomly discard some neurons during the training process.

[0047] Among them, the non-linear activation function can be preset according to specific circumstances, such as the Sigmoid activation function, ReLU function, Tanh function, and Softmax function, etc.; the regularization function can be implemented by Dropout.

[0048] It can be understood that the propagation layer structures of the first propagation layer and the second propagation layer of the neural network model in the embodiments of the present application are the same, and both include a hidden layer, a normalization layer, a non-linear activation function, and a regularization function. Among them,

[0049] the hidden layer is used to extract and transform the input data of the neural network model;

[0050] the normalization layer is used to normalize the output data of the hidden layer, making the distribution of the output data more stable and easier to process, which can improve the convergence speed of the neural network model, and then improve the training speed of the neural network model, and to a certain extent, can reduce the risk of overfitting of the neural network model;

[0051] the non-linear activation function is used to introduce non-linear characteristics into the normalized data to improve the processing ability of the neural network model for complex data;

[0052] the regularization function is used to randomly discard some neurons during the training process, so that the neural network model does not overly rely on certain specific neurons to prevent overfitting of the neural network model and improve the generalization ability of the neural network model. During the training process, each neuron is randomly assigned a value of 0 (i.e., discarded) with a probability P, and the output of the neurons that are not discarded will be scaled by 1 / (1 - p), so that the expected value remains unchanged.

[0053] Specifically, the structure of the neural network model in the embodiments of the present application is as Figure 2As shown, taking welding parameters and backing plate thickness as inputs, and welding quality (size of nail tip defect), backing plate machining allowance, and energy consumption as outputs, an error backpropagation neural network model is constructed. In the first step, initial weights are randomly given first. The input layer receives the original data and transmits it to the hidden layer. The input is nonlinearly transformed through the nonlinear activation functions of the two hidden layers, and then the prediction result is output through the regression linear activation function. So far, the forward propagation is completed, and the output of each layer is calculated. In the second step, the mean square error loss function is used to calculate the error between the predicted value and the true value, and then the error is reversely transmitted from the output layer to the input layer to calculate the gradient of each layer, and then the weights and biases of the network are updated and optimized through the gradient descent method. In the third step, the forward propagation, error calculation, reverse propagation, and weight update of the previous two steps are repeated, and the iteration is continuously carried out until the loss function converges.

[0054] In the embodiment of the present application, training a pre-constructed neural network model using a data set includes: dividing the data in the data set into a training set, a test set, and a validation set according to a target ratio; training the neural network model using the training set and a loss function, and using the validation set to verify the loss value or prediction accuracy of the trained neural network model until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and then stopping training the neural network model; using the test set to test the trained neural network model.

[0055] Among them, the preset period, preset number, change threshold, and preset threshold can all be set according to specific circumstances, and no specific limitation is made here; the target ratio can be set according to the specific situation, such as the training set: test set: validation set is 8:1:1.

[0056] It can be understood that in the embodiment of the present application, the data in the data set can be divided into a training set, a test set, and a validation set according to a target ratio. The training set is used to train the neural network model, the validation set is used to verify the performance of the neural network model, and the test set is used to test the true performance of the trained neural network model. The loss value or accuracy of the trained neural network model is verified using the validation set until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and then stopping training the neural network model, that is, stopping training in advance when the training performance of the validation set cannot be improved to prevent problems such as gradient disappearance or explosion and model overfitting.

[0057] In the embodiment of the present application, updating the model parameters of the neural network model according to the error includes: obtaining the initial learning rate of the neural network model; calculating the model gradient of the neural network model based on the error, updating the initial learning rate based on the model gradient, and updating the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

[0058] Among them, the model parameters mainly refer to the weights and biases of the model.

[0059] It can be understood that the embodiments of the present application can calculate the model gradient of the neural network model based on the error, and can refer to the historical model gradient during the training process. Based on the model gradient, the initial learning rate is updated, which can automatically reduce the learning rate of parameters with frequent updates and large amplitude changes, increase the learning rate of parameters with a lower update frequency, use different learning rates for different parameters, accelerate the convergence speed of the model, and then update the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

[0060] Among them, the weights in the model parameters can dynamically adjust the weights through Adam adaptive learning rate or by using Weight_Decay (weight decay) to achieve L2 regularization to realize amplitude update, and the weight adjustment is carried out after the backpropagation of the neural network model and before the new round of weight update.

[0061] Specifically, the training of the neural network model in the embodiments of the present application is as Figure 3 shown.

[0062] Data collection and preprocessing: Collect a large amount of electron beam welding experimental data during the production and manufacturing process, including welding parameters (acceleration voltage, electron beam current, welding speed, focus depth, etc.), weld quality (presence or absence of nail tip defects or the size of nail tip defects), backing plate thickness, and subsequent backing plate machining allowance, etc., and preprocess the data to remove outliers.

[0063] 2. Use the preprocessed data to train the neural network model. This neural network model has 2 hidden layers. The first layer is set with 12 neurons, the second layer is set with 5 neurons, the number of samples is 475, and the data set is divided into a training set, a test set, and a validation set according to the ratio of 8:1:1. The divided training set is iterated 300 rounds in the model, and the learning rate is set to 0.1.

[0064] In the first step, randomly give the initial weights first. The input layer receives the original data and passes it to the hidden layer. The input is non-linearly transformed through the Sigmoid activation function of the 2 hidden layers, and then the prediction result is output through the regression linear activation function. So far, the forward propagation is completed, and the output of each layer is calculated.

[0065] In the second step, use the mean square error loss function to calculate the error between the predicted value and the true value, then reverse transmit the error from the output layer to the input layer, calculate the gradient of each layer, and then update and optimize the weights and biases of the network through the gradient descent method.

[0066] In the third step, repeat the forward propagation, error calculation and backpropagation, and weight update in the previous two steps, and continuously iterate until the loss function converges.

[0067] 3. Evaluate the trained neural network model using the data in the validation set and stop training when the conditions are met.

[0068] In step S103, monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process using the trained neural network model, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0069] It can be understood that the embodiments of the present application can monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process using the trained neural network model, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance, thereby improving the welding quality, reducing the welding energy consumption, reasonably controlling the backing plate thickness, reducing the subsequent machining amount, reducing the cost, and improving the production efficiency by automatically adjusting the welding process parameters and backing plate thickness to meet the requirements of mass production.

[0070] In short, during the actual welding process, monitor data such as welding parameters, weld quality, welding energy consumption, and machining allowance in real time, and continuously adjust and optimize the welding parameters and backing plate thickness through a feedback mechanism to ensure the stability and consistency of the welding quality.

[0071] In summary, the embodiments of the present application improve the neural network technology to automatically optimize the electron beam welding process parameters, reduce energy consumption, ensure nail tip defects in the weld, reduce rework and scrap rates; by reasonably controlling the backing plate thickness, reduce the subsequent machining amount, reduce the material processing production cost, and improve the production efficiency; by improving a suitable neural network model according to the training characteristics of the electron beam welding process parameter data set, the neural network model has a fast convergence speed, less human intervention, is not easy to overfit, has strong generalization ability, and can adapt to the actual electron beam welding production needs.

[0072] According to the neural network model training method for the electron beam welding process proposed by the embodiments of the present application, a data set can be generated based on the historical experimental data of the electron beam welding process, and the pre-constructed neural network model can be trained using the data set. During the training process, update the model parameters of the neural network model based on error backpropagation to improve the accuracy of the model output. Furthermore, use the trained neural network model to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance to improve the welding quality, reduce the welding energy consumption, reasonably control the backing plate thickness, reduce the subsequent machining amount, reduce the production cost, and improve the production efficiency by automatically adjusting the welding process parameters and backing plate thickness, thereby meeting the requirements of mass production.

[0073] The above embodiments focus on the description of the neural network model training method for the electron beam welding process. The following embodiments focus on using the neural network model trained in the above embodiments from an application perspective to optimize the electron beam welding process parameters.

[0074] An embodiment of the present application also provides a method for optimizing electron beam welding process parameters.

[0075] As Figure 4 shown, the method for optimizing electron beam welding process parameters includes the following steps:

[0076] In step S201, obtain the current monitoring data of the electron beam welding process.

[0077] Among them, the current monitoring data includes welding parameters, welding quality, welding energy consumption, backing plate thickness, and backing plate machining allowance.

[0078] In step S202, identify the welding parameters and backing plate thickness in the current monitoring data.

[0079] It can be understood that the embodiment of the present application can identify the welding parameters and backing plate thickness in the current monitoring data for subsequent optimization of the welding parameters and backing plate thickness.

[0080] In step S203, input the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model. The output of the neural network model is weld quality, welding energy consumption, and machining allowance, where the neural network model is trained based on the above-mentioned neural network model training method for the electron beam welding process.

[0081] Among them, the structure and training process of the neural network model have been described in the above embodiments and will not be elaborated here.

[0082] It can be understood that the embodiment of the present application can input the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model, and the output of the neural network model is weld quality, welding energy consumption, and machining allowance.

[0083] In step S204, optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0084] It can be understood that the embodiment of the present application can optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance, continuously adjust and optimize the welding parameters and backing plate thickness through the feedback mechanism of the neural network model to ensure the stability and consistency of welding quality, reduce welding energy consumption, reasonably control the backing plate thickness, reduce subsequent machining volume, reduce costs, and improve production efficiency by automatically adjusting the welding process parameters and backing plate thickness to meet the needs of large-scale production.

[0085] In addition, it should be noted that the optimization of the welding parameters and backing plate thickness in the embodiments of the present application is achieved through a neural network model. The neural network model in the present application is an optimal implementation method, and traditional optimization algorithms, expert systems, or hybrid methods can also be used. Among them,

[0086] Traditional optimization algorithms: Traditional optimization algorithms such as genetic algorithms and particle swarm optimization can be used to replace the deep learning model for optimizing welding parameters and backing plate thickness;

[0087] Expert system: An expert system can be constructed to optimize welding parameters and backing plate thickness through a rule base and an inference engine;

[0088] Hybrid method: The traditional optimization algorithm and the deep learning model can be combined. The traditional optimization algorithm is used for preliminary optimization, and then the deep learning model is used for fine optimization.

[0089] According to the method for optimizing electron beam welding process parameters proposed in the embodiments of the present application, the welding parameters and backing plate thickness in the monitoring data of the current electron beam welding process can be identified. By inputting the welding parameters and backing plate thickness into the trained neural network model, the welding parameters and backing plate thickness are continuously adjusted and optimized through the feedback mechanism of the neural network model, ensuring the stability and consistency of welding quality, reducing welding energy consumption, reasonably controlling the backing plate thickness, reducing subsequent machining volume, reducing production costs, and improving production efficiency by automatically adjusting welding process parameters and backing plate thickness to meet the requirements of large-scale production.

[0090] Next, a neural network model training device for the electron beam welding process according to the embodiments of the present application will be described with reference to the accompanying drawings.

[0091] Figure 5 It is a block diagram of a neural network model training device for the electron beam welding process in the embodiments of the present application.

[0092] As Figure 5 shown, the neural network model training device 10 for the electron beam welding process includes: a first acquisition module 101, a training module 102, and a monitoring module 103.

[0093] Among them, the first acquisition module 101 is used to acquire historical experimental data of the electron beam welding process. The historical experimental data includes welding parameters, weld quality, welding energy consumption, backing plate thickness, and machining allowance of the backing plate thickness of the electron beam welding process. The training module 102 is used to generate a data set according to the historical experimental data, and use the data set to train a pre-constructed neural network model. During the training process, the model parameters of the neural network model are updated based on error backpropagation. The input of the neural network model is welding parameters and backing plate thickness, and the output of the neural network model is weld quality, welding energy consumption, and machining allowance. The monitoring module 103 is used to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process by using the trained neural network model, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0094] In the embodiment of the present application, the neural network model includes a first propagation layer, a second propagation layer, and a loss function. Among them, the first propagation layer generates a first prediction result according to the input data of the neural network model. The input data includes multiple welding parameters, backing plate thickness, and training errors in the data set. The second propagation layer generates a second prediction result according to the first prediction result. The loss function calculates the error according to the second prediction result and the true value in the data set, and reversely transmits the error to the input of the neural network model, and updates the model parameters of the neural network model according to the error until the loss function converges. The true value in the data set includes weld quality, welding energy consumption, and machining allowance.

[0095] In the embodiment of the present application, the propagation layer structures of the first propagation layer and the second propagation layer are the same. The propagation layer structure includes: a hidden layer, a normalization layer, a non-linear activation function, and a regularization function. The hidden layer is used to perform feature extraction and transformation on the input data of the neural network model. The normalization layer is used to perform normalization processing on the output data of the hidden layer. The non-linear activation function is used to introduce non-linear characteristics into the normalized data. The regularization function is used to randomly lose some neurons during the training process.

[0096] In the embodiment of the present application, the training module 102 is further used to: divide the data in the data set into a training set, a test set, and a validation set according to a target ratio; use the training set and the loss function to train the neural network model, and use the validation set to verify the loss value or prediction accuracy of the trained neural network model until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and stop training the neural network model; use the test set to test the trained neural network model.

[0097] In the embodiment of the present application, updating the model parameters of the neural network model according to the error includes: obtaining the initial learning rate of the neural network model; calculating the model gradient of the neural network model based on the error, updating the initial learning rate based on the model gradient, and updating the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

[0098] It should be noted that the foregoing explanation of the embodiment of the neural network model training method for the electron beam welding process is also applicable to the neural network model training device for the electron beam welding process in this embodiment, and will not be elaborated here.

[0099] The neural network model training device for the electron beam welding process according to the embodiment of the present application can generate a data set according to the historical experimental data of the electron beam welding process, use the data set to train a pre-constructed neural network model, and update the model parameters of the neural network model based on error backpropagation during the training process to improve the accuracy of the model output. Furthermore, the trained neural network model is used to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance, so as to improve the welding quality, reduce the welding energy consumption, reasonably control the backing plate thickness, reduce the subsequent machining amount, reduce the production cost, improve the production efficiency by automatically adjusting the welding process parameters and backing plate thickness, and thus meet the requirements of mass production.

[0100] Figure 6 It is a block diagram of an electronic welding process parameter optimization device according to another embodiment of the present application.

[0101] As Figure 6 shown, the electronic welding process parameter optimization device 20 includes: a second acquisition module 201, an identification module 202, an input module 203, and an optimization module 204.

[0102] Among them, the second acquisition module 201 is used to acquire the current monitoring data of the electron beam welding process; the identification module 202 is used to identify the welding parameters and backing plate thickness in the current monitoring data; the input module 203 is used to input the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model, and the output of the neural network model is the weld quality, welding energy consumption, and machining allowance, where the neural network model is trained based on the neural network model training device for the electron beam welding process in the foregoing embodiment; the optimization module 204 is used to optimize the welding parameters and backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

[0103] The electronic welding process parameter optimization device proposed according to the embodiments of the present application can identify the welding parameters and backing plate thickness in the monitoring data of the current electron beam welding process. By inputting the welding parameters and backing plate thickness into the trained neural network model, the welding parameters and backing plate thickness are continuously adjusted and optimized through the feedback mechanism of the neural network model, ensuring the stability and consistency of welding quality, reducing welding energy consumption, reasonably controlling the backing plate thickness, reducing subsequent machining volume, reducing production costs, and improving production efficiency by automatically adjusting the welding process parameters and backing plate thickness to meet the needs of large-scale production.

[0104] Figure 7 The structural schematic diagram of the electronic device provided by the embodiments of the present application. The electronic device may include:

[0105] A memory 701, a processor 702, and a computer program stored on the memory 701 and executable on the processor 702.

[0106] When the processor 702 executes the program, it implements the neural network model training method for the electron beam welding process or the electronic beam welding process parameter optimization method provided in the above embodiments.

[0107] Further, the electronic device further includes:

[0108] A communication interface 703 for communication between the memory 701 and the processor 702.

[0109] The memory 701 is used to store a computer program executable on the processor 702.

[0110] The memory 701 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory.

[0111] If the memory 701, the processor 702, and the communication interface 703 are independently implemented, the communication interface 703, the memory 701, and the processor 702 can be interconnected through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 7 only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0112] Optionally, in a specific implementation, if the memory 701, the processor 702, and the communication interface 703 are integrated on a single chip, the memory 701, the processor 702, and the communication interface 703 can communicate with each other through an internal interface.

[0113] The processor 702 may be a central processing unit (CPU for short), or an application specific integrated circuit (ASIC for short), or one or more integrated circuits configured to implement the embodiments of the present application.

[0114] The embodiments of the present application further provide a computer-readable storage medium, on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the neural network model training method for the electron beam welding process, or the electron beam welding process parameter optimization method as described above is implemented.

[0115] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, without conflict, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples.

[0116] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined.

[0117] Any process or method description depicted in the flowchart or described otherwise herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of this application includes additional implementations where functions may be executed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of this application pertain.

[0118] It should be understood that the various parts of this application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays, field programmable gate arrays, and the like.

[0119] Those of ordinary skill in the art can understand that all or part of the steps carried out in the method of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A method for training a neural network model of an electron beam welding process, characterized in that, Including the following steps: Obtain historical experimental data of the electron beam welding process, where the historical experimental data includes welding parameters, weld quality, welding energy consumption, backing plate thickness, and machining allowance of the backing plate thickness of the electron beam welding process; Generate a data set according to the historical experimental data, and use the data set to train a pre-constructed neural network model. During the training process, update the model parameters of the neural network model based on error backpropagation. Wherein, the input of the neural network model is the welding parameter and the backing plate thickness, and the output of the neural network model is the weld quality, the welding energy consumption, and the machining allowance; Use the trained neural network model to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process, and optimize the welding parameters and the backing plate thickness according to the weld quality, the welding energy consumption, and the machining allowance.

2. The method for training a neural network model of the electron beam welding process according to claim 1, wherein, The neural network model includes a first propagation layer, a second propagation layer, and a loss function, where The first propagation layer generates a first prediction result according to the input data of the neural network model, where the input data includes multiple welding parameters, backing plate thickness, and training errors in the data set; The second propagation layer generates a second prediction result according to the first prediction result; The loss function calculates an error according to the second prediction result and the true value in the data set, reversely transmits the error to the input of the neural network model, and updates the model parameters of the neural network model according to the error until the loss function converges, where the true value in the data set includes the weld quality, the welding energy consumption, and the machining allowance.

3. The method for training the neural network model of the electron beam welding process according to claim 2, characterized in that, The propagation layer structures of the first propagation layer and the second propagation layer are the same, where the propagation layer structure includes: a hidden layer, a normalization layer, a non-linear activation function, and a regularization function, where The hidden layer is used to extract and transform the input data of the neural network model; The normalization layer is used to perform normalization processing on the output data of the hidden layer; The non-linear activation function is used to introduce non-linear characteristics into the normalized data; The regularization function is used to randomly lose some neurons during the training process.

4. The method for training the neural network model of the electron beam welding process according to claim 2, wherein, The using the data set to train a pre-constructed neural network model includes: Divide the data in the data set into a training set, a test set, and a validation set according to a target ratio; Use the training set and the loss function to train the neural network model, and use the validation set to verify the loss value or prediction accuracy of the trained neural network model until the number of training times reaches a preset number, or the change rate of the loss value is less than a change threshold, or the prediction accuracy is greater than a preset threshold, and stop training the neural network model; Use the test set to test the trained neural network model.

5. The method for training a neural network model of an electron beam welding process according to claim 2, wherein The updating the model parameters of the neural network model according to the error includes: Obtain the initial learning rate of the neural network model; Calculate the model gradient of the neural network model based on the error, update the initial learning rate based on the model gradient, and update the model parameters of the neural network model based on the model gradient and the updated initial learning rate.

6. An optimization method for electron beam welding process parameters, characterized in that, Comprising: Obtain the current monitoring data of the electron beam welding process; Identify the welding parameters and backing plate thickness in the current monitoring data; Input the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model, and the output of the neural network model is the weld quality, welding energy consumption, and machining allowance, wherein the neural network model is trained based on the neural network model training method of the electron beam welding process according to any one of claims 1-5; Optimize the welding parameters and the backing plate thickness according to the weld quality, the welding energy consumption, and the machining allowance.

7. An apparatus for training a neural network model of an electron beam welding process, characterized in that, Comprising: A first acquisition module, configured to acquire historical experimental data of the electron beam welding process, wherein the historical experimental data includes the welding parameters, weld quality, welding energy consumption, backing plate thickness, and machining allowance of the backing plate thickness of the electron beam welding process; A training module, configured to generate a data set according to the historical experimental data, and use the data set to train a pre-constructed neural network model. During the training process, update the model parameters of the neural network model based on error backpropagation, wherein the input of the neural network model is the welding parameters, the welding energy consumption, and the backing plate thickness, and the output of the neural network model is the weld quality, the welding energy consumption, and the machining allowance; A monitoring module, configured to monitor the weld quality, welding energy consumption, and machining allowance of the electron beam welding process by using the trained neural network model, and optimize the welding parameters and the backing plate thickness according to the weld quality, welding energy consumption, and machining allowance.

8. An electronic welding process parameter optimization device, characterized in that, Comprising: A second acquisition module, configured to acquire the current monitoring data of the electron beam welding process; An identification module, configured to identify the welding parameters and backing plate thickness in the current monitoring data; An input module, configured to input the welding parameters and backing plate thickness in the current monitoring data into the trained neural network model, and the output of the neural network model is the weld quality, the welding energy consumption, and the machining allowance, wherein the neural network model is trained based on the neural network model training device of the electron beam welding process according to claim 7; An optimization module, configured to optimize the welding parameters and the backing plate thickness according to the weld quality, the welding energy consumption, and the machining allowance.

9. An electronic device, characterized in that, Comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the program to implement the neural network model training method of the electron beam welding process according to any one of claims 1-5, or the electron beam welding process parameter optimization method according to claim 6.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, The computer program or instruction is executed by the processor to implement the neural network model training method of the electron beam welding process according to any one of claims 1-5, or the electron beam welding process parameter optimization method according to claim 6.

Images