Magnesium alloy forging process optimization method based on simulation and neural network

Abstract

The invention discloses a magnesium alloy forging process optimization method based on simulation and a neural network, and relates to the field of magnesium alloy plastic forming, and the method specifically comprises the following steps: step 1, establishing a magnesium alloy forging finite element model, and performing calculation through a deformation-heat transfer three-dimensional finite element numerical simulation system; 2, establishing a multi-input and multi-output BP neural network model, and determining parameters of an input layer and an output layer; 3, selecting an orthogonal experiment method to obtain training and test sample data of the neural network; 4, performing normalization processing on sample data obtained by the orthogonal test, and importing the sample data into a neural network model for training and prediction; 5, comparing and analyzing a neural network prediction value and an actual simulation value, performing relative error evaluation, and determining the prediction accuracy of the constructed BP neural network model; and step 6, predicting through the constructed BP neural network model to obtain an optimal forging process optimization parameter combination. According to the method, the reduction of the maximum load value and the maximum wear value of the die in the forging process is taken as an optimization target, the optimal parameter combination of the forging process of the magnesium alloy forge piece is quickly and efficiently obtained through the constructed BP neural network model, the product development period is shortened, and the process production efficiency is improved.

Description

Technical Field

[0001] The invention belongs to the field of magnesium alloy plastic forming, and in particular relates to a magnesium alloy forging process optimization method based on simulation and neural network. Background Art

[0002] Magnesium alloys are widely used in aerospace, automotive, rail transportation and other fields due to their light weight, high strength and good processing properties. Their low density allows the product to reduce weight while maintaining sufficient strength. However, magnesium alloys are prone to wear on the mold during high-temperature forging, especially under high load and high-speed conditions. Wear on the mold surface will occur, shortening the mold life.

[0003] When using forging technology to produce products in actual engineering, the service life of the mold needs to be considered. In the process of forging process development of magnesium alloy forgings, the cost of the mold accounts for about 20% of the total cost. The degree of mold wear is an important indicator to measure the service life of the mold. Appropriate adjustment of forging process parameters, such as temperature, load and lubrication conditions, can help reduce the wear of the mold during the forging process of magnesium alloys. However, in current related research, there is no means to efficiently find the best process parameters while improving the service life of the mold. Usually, a large number of finite element simulations are required to determine the degree of mold wear, so as to find the best combination of process parameters, which greatly prolongs the product development cycle. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and to provide a magnesium alloy forging process optimization method based on simulation and neural network. The method takes the minimum value of the maximum forging load and the maximum die wear as the optimization goal, and uses the constructed BP neural network for prediction to optimize a large number of cumbersome finite element simulation processes, so as to quickly find the optimal combination of forging process parameters, thereby greatly reducing the product development cycle and improving efficiency.

[0005] The technical solution adopted by the present invention is: A method for optimizing magnesium alloy forging process based on simulation and neural network comprises the following steps: Step 1: Establish a finite element model of the magnesium alloy forging, perform finite element simulation calculations through a deformation-heat transfer three-dimensional finite element numerical simulation system, and obtain the temperature field, stress field, strain field, maximum forging load, and die wear degree of the forging, wherein the minimum value of the maximum forging load and the maximum die wear value are taken as the optimization goal; Step 2: Establish a multi-input and multi-output BP neural network model, with forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient as input parameters of the neural network model, and maximum forging load and die wear degree as output parameters of the neural network model; Step 3: Use the orthogonal experimental design method to design and allocate the five parameters of forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient, and obtain the maximum forging load and die wear degree under each parameter combination through finite element simulation; Step 4: normalize the sample data obtained from the finite element simulation and import it into the neural network model for training and prediction; Step 5: Compare and analyze the predicted value of the neural network and the actual simulated value and conduct relative error evaluation to determine the accuracy of the prediction of the constructed BP neural network model; Step 6: Predict the best forging process optimization parameter combination through the constructed BP neural network model.

[0006] Further options are: The structural parameters of the BP neural network described in step 2 are set as follows: the number of input layer nodes is 5, the number of output layer nodes is 2, and the number of hidden layer nodes is 1. The calculation formula of the number of hidden layer nodes s is as follows: Where w is the number of input neurons, v is the number of output neurons, and a is a constant between [1, 10]. Its specific value can be adjusted during neural network training to obtain the optimal number of hidden layer neurons.

[0007] Further options are: The orthogonal experimental design method described in step 3 uses the expression To express, where n is the number of experiments, k is the experimental factor, and m is the number of levels of each factor; in this experiment, the five input parameters of the BP neural network are used as experimental factors, namely forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient, and four levels are selected for each process parameter, namely .

[0008] Further options are: The normalization calculation formula described in step 4 is: In the formula, is the data value after normalization, and X is the data value without normalization. and are the corresponding maximum and minimum values ​​in this group of sample data respectively.

[0009] Further options are: The neural network training described in step 4, during the training process, when the training error is less than the target value or reaches the maximum number of training times, the training is terminated automatically; in addition, the BP neural network training adopts the trainbfg algorithm with faster convergence speed and smaller step size, the input layer function adopts the nonlinear logsig function, the output layer adopts the sigmoid function, the hidden layer adopts the tanh function, and the target error of the grid performance is , the rate is , the maximum number of training times is 1000.

[0010] Further options are: The calculation formula for the relative error assessment described in step 5 is: Where Z is the relative error, Y is the predicted value of the BP neural network, and X is the actual value of the finite element simulation.

[0011] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: The present invention combines finite element simulation software with BP neural network to optimize a large number of cumbersome finite element simulation processes, quickly find the optimal combination of forging process parameters, and obtain the effect of minimizing the maximum forging load value and the maximum die wear value as the optimization target, which greatly reduces the product development cycle and improves efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of the magnesium alloy forging process optimization method based on simulation and neural network of the present invention; Figure 2 This is the structural diagram of the constructed BP neural network model; Figure 3 The initial structural finite element model of the magnesium alloy steering pedal; Figure 4 The maximum forging load is actually simulated for the optimal process parameters; Figure 5 The actual simulated die wear distribution diagram for the optimal process parameters. Specific implementation plan

[0013] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0014] This embodiment takes the forging process of a steering pedal for a rail transit seat as an example, and the present invention provides a forging process optimization method for a magnesium alloy steering pedal, comprising the following steps: Step 1: The initial structure of the magnesium alloy steering pedal is as follows: Figure 3As shown in the figure, a finite element model of the steering pedal is established, and finite element simulation calculation is performed through a deformation-heat transfer three-dimensional finite element numerical simulation system to obtain the forging temperature field, stress field, strain field, maximum forging load and die wear degree, wherein the process takes the minimum value of the maximum forging load and the maximum die wear as the optimization goal; the forming process includes three steps: blank making, pre-forging and final forging, wherein the pre-forging process includes two stages: early upsetting deformation and later extrusion forming. The required forming load is small, the die wear is small, and the die shape is simple, so only the final forging process needs to be considered.

[0015] Step 2: Establish a multi-input and multi-output BP neural network model, with forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient as input parameters of the neural network model, and maximum forging load and die wear degree as output parameters of the neural network model. Therefore, the number of input layer nodes is 5, the number of output layer nodes is 2, and the number of hidden layer layers is 1. Figure 2 It is a 3-layer BP neural network structure, in which the calculation formula of the number of nodes s in the hidden layer is: Where w is the number of input neurons, v is the number of output neurons, and a is a constant between [1, 10]. Its specific value can be adjusted during neural network training to obtain the optimal number of hidden layer neurons.

[0016] Step 3: Use the orthogonal experimental design method to design and allocate the five parameters of forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient, and obtain the maximum forging load value and die wear degree under each parameter combination through finite element simulation; In this experiment, five input parameters of the BP neural network are used as experimental factors, namely forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient. Four levels are selected for each process parameter, namely , used for fitting and verification of BP neural network training.

[0017]

[0018] Step 4: Normalize the sample data obtained from the finite element simulation so that the training and test samples fall between [-1, 1]. This can effectively ensure the intuitiveness of the data distribution range and enable the training samples to converge quickly to meet the accuracy requirements. The processed data is imported into the neural network model for training and testing.

[0019]

[0020] During the training process, when the training error is less than the target value or reaches the maximum number of training times, the training is terminated automatically. In addition, the BP neural network training uses the trainbfg algorithm with faster convergence speed and smaller step size. The input layer function uses the nonlinear logsig function, the output layer uses the sigmoid function, and the hidden layer uses the tanh function. The target error of the grid performance is , the rate is , the maximum number of training times is 1000.

[0021] After the BP neural network training is completed, certain data is usually required for testing. Within the selected data range, 4 groups of data are randomly selected for simulation as test samples. The test results are shown in Table 3.

[0022]

[0023] Step 5: Compare and analyze the predicted values ​​of the neural network and the actual simulated values ​​and conduct relative error evaluation to determine the accuracy of the prediction of the constructed BP neural network model, as shown in Table 4.

[0024]

[0025] From the data analysis in Table 4, it can be concluded that in the prediction of the maximum forging load, only one set of data has a large error, and its relative error exceeds 10%, while the relative errors of the remaining data are small, and the overall relative error average is 2.2507%; in the prediction of the degree of die wear, the relative error values ​​of all data are less than 5%, and the overall relative error average is 2.0311%, indicating that the constructed BP neural network model structure is relatively accurate in the prediction of the maximum forming load and die wear degree of the final forging of the magnesium alloy steering pedal.

[0026]

[0027] By comparing the predicted values ​​of the neural network and the actual values ​​of the finite element simulation in the test group, and calculating their respective relative errors, it can be concluded, as shown in Table 5, that the relative errors between the predicted values ​​of the maximum forging load and die wear degree of each group of data in the test group and the actual simulation data are small, and the overall average error does not exceed 5%, indicating that the constructed BP neural network model has a high prediction accuracy and good prediction performance.

[0028] Step 6. The constructed BP neural network model is used to predict the best forging process optimization parameter combination. In this case, when the forging speed is 20 mm / s, the billet temperature is 400 °C, the die temperature is 350 °C, the heat transfer coefficient is 3, and the friction coefficient is 0.3, the maximum forging load value and the die wear value are the smallest, which are 142 t and 0.000003433 mm respectively. That is, the above parameters are the best forging process optimization parameter combination.

[0029] The above embodiments are only used to illustrate the design concept and features of the present invention, and it should be understood that the protection scope of the present invention is not limited to the above embodiments. Therefore, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should fall within the protection scope of the present invention.

Claims

1. A magnesium alloy forging process optimization method based on simulation and neural network, characterized in that: include: Step 1: Establish a finite element model of the magnesium alloy forging, and calculate through a deformation-heat transfer three-dimensional finite element numerical simulation system to obtain the temperature field, stress field, strain field, maximum forging load, and maximum die wear of the forging, wherein the minimum value of the maximum forging load and the maximum die wear is taken as the optimization goal; Step 2: Establish a multi-input and multi-output BP neural network model, with forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient as input parameters of the neural network model, and maximum forging load and die wear degree as output parameters of the neural network model; Step 3: Use the orthogonal experimental design method to design and allocate the five parameters of forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient, and obtain the maximum forging load and die wear degree under each parameter combination through finite element simulation; Step 4: normalize the sample data obtained from the finite element simulation and import it into the neural network model for training and prediction; Step 5: Compare and analyze the predicted value of the neural network and the actual simulated value and conduct relative error evaluation to determine the accuracy of the prediction of the constructed BP neural network model; Step 6: Predict the best forging process optimization parameter combination through the constructed BP neural network model.

2. The magnesium alloy forging process optimization method based on simulation and neural network according to claim 1, characterized in that: The structural parameters of the BP neural network described in step 2 are set as follows: the number of input layer nodes is 5, the number of output layer nodes is 2, and the number of hidden layer nodes is 1. The calculation formula of the number of hidden layer nodes s is as follows: Where w is the number of input neurons, v is the number of output neurons, and a is a constant between [1, 10]. Its specific value can be adjusted during neural network training to obtain the optimal number of hidden layer neurons.

3. The magnesium alloy forging process optimization method based on simulation and neural network according to claim 1, characterized in that: The orthogonal experimental design method described in step 3 uses the expression To express, where n is the number of experiments, k is the experimental factor, and m is the number of levels of each factor; in this experiment, the five input parameters of the BP neural network are used as experimental factors, namely forging speed, billet temperature, die temperature, heat transfer coefficient and friction coefficient, and four levels are selected for each process parameter, namely .

4. The magnesium alloy forging process optimization method based on simulation and neural network according to claim 1, characterized in that: The neural network training described in step 4, during the training process, when the training error is less than the target value or the maximum number of training times is reached, the training is terminated automatically; In addition, the BP neural network training uses the trainbfg algorithm with faster convergence speed and less step size. The input layer function uses the nonlinear logsig function, the output layer uses the sigmoid function, and the hidden layer uses the tanh function. The target error of the grid performance is , the rate is , the maximum number of training times is 1000.

5. The magnesium alloy forging process optimization method based on simulation and neural network according to claim 1, characterized in that: The normalization calculation formula described in step 4 is: In the formula, is the data value after normalization, and X is the data value without normalization. and are the corresponding maximum and minimum values ​​in this group of sample data respectively.

6. The magnesium alloy forging process optimization method based on simulation and neural network according to claim 1, characterized in that: The calculation formula for the relative error assessment described in step 5 is: Where Z is the relative error, Y is the predicted value of the BP neural network, and X is the actual value of the finite element simulation.

Images