Skin longitudinal stretching process parameter optimization method based on neural network and genetic algorithm

Abstract

The invention provides a skin longitudinal stretching process parameter optimization method based on a neural network and a genetic algorithm, and designs a skin forming quality prediction model based on a BP neural network according to a quality index to be reached by a skin finished product and related material and stretcher process parameters. Meanwhile, the weight and the threshold value of the neural network are updated based on a genetic algorithm. By utilizing the trained and updated prediction model, the real-time efficient optimization of the process parameters of the stretcher can be realized aiming at the material selected by the skin and the expected quality of the design, and if the early-stage correlation analysis and screening of the parameters to be optimized are matched, each link in the longitudinal stretching process can be more accurately and comprehensively improved; therefore, the large-scale and intelligent manufacturing level of skin products can be improved by combining industrial big data.

Description

Technical Field

[0001] The invention belongs to the technical field of aircraft skin manufacturing and processing, and in particular relates to a method for optimizing skin longitudinal drawing process parameters based on a neural network and a genetic algorithm. Background Art

[0002] At present, for the skin, a key component in the aircraft structure, many existing technologies have adopted the advanced process of metal material stretching. When executing this process, it is necessary to make complex and precise adjustments to process parameters such as mold clamping force, stretching trajectory and speed. In order to ensure that the product quality and processing accuracy in actual production meet the design target requirements, there is a high reliance on trial and error verification in the early stage of production and the accumulation of a large amount of data and experience. With the emergence of various new skin component designs, the above verification and data collection work has become increasingly arduous, thus limiting the improvement of the longitudinal stretching process and the efficiency of large-scale production of skins. Summary of the invention

[0003] In view of this, in order to solve the technical problems existing in this field, the present invention provides a method for optimizing the process parameters of skin longitudinal drawing based on neural network and genetic algorithm, which specifically includes the following steps:

[0004] Step 1: Collect the historical process parameters of the stretching machine and the corresponding historical parameters of the skin material and quality during the longitudinal stretching process of the skin, and establish a corresponding data sample set after performing data preprocessing such as data cleaning;

[0005] Step 2: Taking the stretching machine process parameters and skin material parameters as input and the skin quality parameters as output, a skin forming quality prediction model based on BP neural network is established;

[0006] Step 3: Divide the data sample set into a training set and a validation set, use the skin forming quality prediction model established by the training set for training, and use the validation set to verify the training effect;

[0007] Step 4: Output the weights and threshold parameters of the trained BP neural network, construct corresponding codes for the weights and threshold parameters as population individuals of the genetic algorithm, and use various population individuals to establish a population matrix;

[0008] Step 5: Execute the genetic algorithm based on the established population matrix, and update the optimal weight and threshold parameters through the iterative process of gene selection, crossover and mutation, and fitness calculation;

[0009] Step six: based on the optimal weight and threshold parameters, update the skin forming quality prediction model and apply it online, so that it can predict the skin forming quality results according to the real-time collected stretching machine process parameters and skin material parameters; feedback and adjust the stretching machine process parameters according to the prediction results, and finally optimize the stretching machine process parameters.

[0010] Furthermore, step one also includes screening the collected stretching machine process parameters by performing correlation analysis, and using the process parameters retained after screening as the process parameters to be optimized and used to construct the training sample set.

[0011] Furthermore, the process parameters to be optimized obtained through correlation analysis and screening include at least stretching rate and stretching speed, which are specifically used together with the skin material parameters as the input of the skin forming quality prediction model in step 2, and the model output includes at least the thickness uniformity and springback parameters of the skin; the corresponding BP neural network is designed to include an input layer, 8 hidden layers and an output layer, wherein the input layer contains 3 neurons, each hidden layer contains 10 neurons, and the output layer contains 2 neurons.

[0012] Furthermore, the weights and threshold coefficients output by the BP neural network in step 4 specifically include 610 weight coefficients and 5 thresholds; they are encoded using 8-bit signed integers so that each parameter takes a value in the range of [-128, 127], thereby obtaining each population individual with a length of G = 4920 bits.

[0013] The method for optimizing the process parameters of the longitudinal drawing of the skin based on neural networks and genetic algorithms provided by the present invention designs a skin forming quality prediction model based on a BP neural network for the quality indicators to be achieved by the finished skin product and the related materials and stretching machine process parameters, and updates the weights and thresholds of the neural network based on the genetic algorithm. The trained and updated prediction model can realize real-time and efficient optimization of the process parameters of the stretching machine for the materials selected for the skin and the expected quality of the design. If combined with the preliminary correlation analysis and screening of the parameters to be optimized, more accurate and comprehensive improvements can be achieved in each link of the longitudinal drawing process, which is conducive to combining industrial big data to improve the scale and intelligent manufacturing level of skin products. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flow chart of the method provided by the present invention;

[0015] Figure 2 Optional structure diagram for the established BP neural network;

[0016] Figure 3 Schematic diagram of the optimization of process parameters based on skin forming quality. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The method for optimizing the process parameters of the longitudinal drawing of the skin based on the neural network and the genetic algorithm provided by the present invention is as follows: Figure 1 As shown, the specific steps include:

[0019] Step 1: Collect the historical process parameters of the stretching machine and the corresponding historical parameters of the skin material and quality during the longitudinal stretching process of the skin, and establish a corresponding data sample set after performing data preprocessing such as data cleaning;

[0020] Step 2: Taking the stretching machine process parameters and skin material parameters as input and the skin quality parameters as output, a skin forming quality prediction model based on BP neural network is established;

[0021] Step 3: Divide the data sample set into a training set and a validation set, use the skin forming quality prediction model established by the training set for training, and use the validation set to verify the training effect;

[0022] Step 4: Output the weights and threshold parameters of the trained BP neural network, construct corresponding codes for the weights and threshold parameters as population individuals of the genetic algorithm, and use various population individuals to establish a population matrix;

[0023] Step 5: Execute the genetic algorithm based on the established population matrix, and update the optimal weight and threshold parameters through the iterative process of gene selection, crossover and mutation, and fitness calculation;

[0024] Step six: based on the optimal weight and threshold parameters, update the skin forming quality prediction model and apply it online, so that it can predict the skin forming quality results according to the real-time collected stretching machine process parameters and skin material parameters; feedback and adjust the stretching machine process parameters according to the prediction results, and finally optimize the stretching machine process parameters.

[0025] In a preferred embodiment of the present invention, step one also includes screening the collected stretching machine process parameters by performing correlation analysis. For example, the process parameters collected in the early stage that may affect the skin quality are screened and eliminated by calculating the correlation coefficient and sorting the results. Only the process parameters retained after screening are used as the process parameters to be optimized and used to construct the training sample set, thereby further improving the pertinence and accuracy of parameter optimization.

[0026] In a preferred embodiment of the present invention, the process parameters to be optimized obtained by correlation analysis and screening include at least stretching rate and stretching speed, which are specifically used together with the skin material parameters as the input of the skin forming quality prediction model in step 2, and the model output includes at least the thickness uniformity and springback parameters of the skin; the corresponding design is as follows: Figure 2 The BP neural network shown includes an input layer, 8 hidden layers and an output layer, wherein the input layer contains 3 neurons, each hidden layer contains 10 neurons, and the output layer contains 2 neurons.

[0027] In a preferred embodiment of the present invention, the weights and threshold coefficients output by the BP neural network in step 4 specifically include 610 weight coefficients and 5 thresholds, with a total of P = 615 network parameters; 8-bit signed integers are used for encoding so that each parameter takes a value in the range of [-128, 127], thereby obtaining a length of G = P*8 = 4920 bits for each population individual, of which the first 4880 bits correspond to each other and the following 40 bits correspond to each other.

[0028] Set the initial population to N, the crossover probability and mutation probability to Pc = 0.7 and Pm = 0.01 respectively, take the stable performance of the genetic population as the standard and the artificial neural network as the objective function, and the population matrix formula is as follows:

[0029] X∈{-128,127} N×G

[0030] Select a suitable genetic generation, perform population inheritance and evolution, and finally obtain an optimal solution set containing n solutions; after the maximum genetic generation Gmax = 100 generations, set a threshold ∈ and a maximum genetic generation Gmax. When the change in the best fitness of consecutive generations is less than ∈ or reaches the maximum generation, the algorithm stops; compared with the data set before optimization, the thickness deviation and springback rate of the skin after stretching are reduced by more than 60%, which shows that the method proposed in the present invention can effectively optimize the longitudinal stretching process parameters; in addition, in the specific implementation, the springback value can be numbered from small to large 1-100, such as Figure 3 As shown in the figure, when the process parameters tend to the springback rate, we can choose the solution with a smaller number, and when the process parameters tend to the thickness uniformity, we can choose the solution with a larger number.

[0031] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0032] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing process parameters of skin longitudinal drawing based on neural network and genetic algorithm, characterized in that: The specific steps include: Step 1: Collect the historical process parameters of the stretching machine and the corresponding historical parameters of the skin material and quality during the longitudinal stretching process of the skin, and establish a corresponding data sample set after performing data preprocessing including data cleaning; Step 2: Taking the stretching machine process parameters and skin material parameters as input and the skin quality parameters as output, a skin forming quality prediction model based on BP neural network is established; Step 3: Divide the data sample set into a training set and a validation set, use the skin forming quality prediction model established by the training set for training, and use the validation set to verify the training effect; Step 4: Output the weights and threshold parameters of the trained BP neural network, construct corresponding codes for the weights and threshold parameters as population individuals of the genetic algorithm, and use various population individuals to establish a population matrix; Step 5: Execute the genetic algorithm based on the established population matrix, and update the optimal weight and threshold parameters through the iterative process of gene selection, crossover and mutation, and fitness calculation; Step six: based on the optimal weight and threshold parameters, update the skin forming quality prediction model and apply it online, so that it can predict the skin forming quality results according to the real-time collected stretching machine process parameters and skin material parameters; feedback and adjust the stretching machine process parameters according to the prediction results, and finally optimize the stretching machine process parameters.

2. The method according to claim 1, characterized in that: Step one also includes screening the collected stretching machine process parameters by performing correlation analysis, and using the process parameters retained after screening as the process parameters to be optimized and used to construct the training sample set.

3. The method according to claim 2, characterized in that: The process parameters to be optimized obtained through correlation analysis and screening include at least stretching rate and stretching speed. In step 2, they are specifically used together with the skin material parameters as the input of the skin forming quality prediction model, and the model output includes at least the thickness uniformity and rebound parameters of the skin; the corresponding BP neural network is designed to include an input layer, 8 hidden layers and an output layer, wherein the input layer contains 3 neurons, each hidden layer contains 10 neurons, and the output layer contains 2 neurons.

4. The method according to claim 3, characterized in that: The weights and threshold coefficients output by the BP neural network in step 4 specifically include 610 weight coefficients and 5 thresholds; they are encoded using 8-bit signed integers so that each parameter takes a value in the range of [-128, 127], thereby obtaining a length of G = 4920 bits for each population individual.

Images