A method for predicting straightening rate of metal matrix composite plate

Abstract

This invention belongs to the technical field of roller straightening, specifically relating to a method for predicting the straightening rate of metal matrix composite plates. The method includes collecting experimental data during the straightening process of the metal matrix composite plate, as well as the unevenness and maximum residual stress values ​​of the straightened plate; generating training and testing set data for a BP neural network; constructing the topology of two BP neural networks, determining the number of neurons in the input, hidden, and output layers, and selecting the activation function of the neural network; and initializing the weights and thresholds of the BP neural network. This invention uses the unevenness and maximum residual stress of the straightened metal matrix composite plate as the objective function, and the straightening force, roller diameter, roller spacing, roller speed, roller length, and the yield strength and thickness of each material as independent variables, performing data analysis and normalization. Then, based on different objective functions, it calculates individual optimal solutions and group optimal solutions, thereby completing multi-objective PSO-BP training to obtain the straightening rate of the composite plate.

Description

Technical Field

[0001] This invention belongs to the technical field of roller straightening, and specifically relates to a method for predicting the straightening rate of metal-based composite plates. Background Technology

[0002] In modern industry, metal matrix composites (MMCs) have been widely used due to their superior properties as multi-metal materials and relatively lower production costs. As the service range and operating conditions of MMCs become increasingly broad and demanding, the requirements for their forming processes are also becoming more stringent. Straightening processes can effectively reduce shape defects and stress concentration in MMCs after rolling, thereby improving their application range and lifespan. Therefore, straightening has become an essential step in the processing of MMCs.

[0003] In the straightening process of metal matrix composite (MMC) panels, the success of straightening is often verified by measuring the panel's flatness and residual stress. Currently, flatness can be measured in real time using laser scanning, while residual stress measurement is divided into destructive testing (i.e., drilling holes) and non-destructive testing (measuring only the surface). In actual production, it is not possible to effectively and clearly determine whether the maximum residual stress of the panel meets the service conditions. Summary of the Invention

[0004] To address the aforementioned issues, this invention discloses a method for predicting the straightening rate of metal-based composite plates. This method, based on big data, predicts the straightening success rate without damaging the metal-based composite plate, thereby meeting the subsequent service requirements of the plate and significantly improving straightening accuracy and product quality.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for predicting the straightening rate of a metal-based composite plate, the method comprising the following steps:

[0007] S1: Collect experimental data during the straightening process of the metal matrix composite plate, as well as the flatness and maximum residual stress value of the plate after straightening, and generate training set data and test set data for the BP network.

[0008] S2: Construct the topology of two BP neural networks based on the training and test data of the BP network, determine the number of neurons in the input layer, hidden layer, and output layer, and select the activation function of the neural network;

[0009] S3: Initialize the weights and thresholds of the BP neural network, evaluate them using the prediction function, and determine the multi-objective function as the basic function of PSO;

[0010] S4: Initialize the dimensions of the PSO particle position and velocity vectors, and set the basic parameters of the multi-objective PSO algorithm, including the number of particles, particle length, particle range, maximum velocity, learning factor, termination condition, and inertia weight.

[0011] S5: Calculate the individual optimal value of a particle. Randomly input a particle, calculate the initial fitness value, and obtain the individual optimal solution for each particle.

[0012] S6: Calculate the optimal value of the particle swarm, find the Pareto optimal solution based on the dominance relationship, and select the one with the fewest Pareto front points as the leader, which is the global optimal solution.

[0013] S7: Update the particle position and velocity update formulas, update the position and velocity of each particle, and consider whether the updated velocity and position are within the specified range; then update the individual optimal value and the population optimal value of the particles based on the recalculated particle fitness value.

[0014] S8: Determine if the iteration stopping condition is met. If not, re-detect the dominance relationship and iterate the calculation of the particle's fitness value until the condition is met.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] This invention uses the unevenness and maximum residual stress of the straightened metal matrix composite plate as the objective function, and the straightening force, roller diameter, roller spacing, roller speed, roller length of the straightener, and the yield strength and thickness of each material as independent variables. Data analysis and normalization are performed on these variables. Then, based on different objective functions, individual optimal solutions and group optimal solutions are calculated, thereby completing multi-objective PSO-BP training to obtain the straightening rate of the composite plate.

[0017] The method of this invention employs a multi-objective particle swarm optimization-backpropagation (PSO-BP) approach, which not only possesses the advantages of BP in solving nonlinear mappings but also reduces the problems of poor prediction accuracy and instability associated with BP. Using the unevenness and residual stress of the sheet metal as target values, it predicts the straightening success rate of the metal matrix composite plate. Based on big data, this method predicts the straightening success rate without damaging the metal matrix composite plate, thus meeting the subsequent service requirements of the sheet metal and having significant implications for improving straightening accuracy and product quality. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0019] Figure 1 This is a schematic diagram of the straightening process for metal-based composite panels.

[0020] Figure 2 This is a flowchart of the multi-target PSO-BP neural network in this invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0023] Example 1

[0024] A method for predicting the straightening rate of a metal-based composite plate, the method comprising the following steps:

[0025] S1: Collect experimental data during the straightening process of the metal matrix composite plate, as well as the flatness and maximum residual stress value of the plate after straightening, and generate training set data and test set data for the BP network.

[0026] S2: Construct the topology of two BP neural networks based on the training and test data of the BP network, determine the number of neurons in the input layer, hidden layer, and output layer, and select the activation function of the neural network;

[0027] S3: Initialize the weights and thresholds of the BP neural network, evaluate them using the prediction function, and determine the multi-objective function as the basic function of PSO;

[0028] S4: Initialize the dimensions of the PSO particle position and velocity vectors, and set the basic parameters of the multi-objective PSO algorithm, including the number of particles, particle length, particle range, maximum velocity, learning factor, termination condition, and inertia weight.

[0029] S5: Calculate the individual optimal value of a particle. Randomly input a particle, calculate the initial fitness value, and obtain the individual optimal solution for each particle.

[0030] S6: Calculate the optimal value of the particle swarm, find the Pareto optimal solution based on the dominance relationship, and select the one with the fewest Pareto front points as the leader, which is the global optimal solution.

[0031] S7: Update the particle position and velocity update formulas, update the position and velocity of each particle, and consider whether the updated velocity and position are within the specified range; then update the individual optimal value and the population optimal value of the particles based on the recalculated particle fitness value.

[0032] S8: Determine if the iteration stopping condition is met. If not, re-detect the dominance relationship and iterate the calculation of the particle's fitness value until the condition is met.

[0033] Example 2

[0034] like Figure 1-2 As shown, a method for predicting the straightening rate of a metal-based composite plate includes the following steps:

[0035] The initial structure of the BP neural network is determined. The neural network used in this invention has three layers. The input layers contain the straightening force, roller diameter, roller spacing, roller speed, and roller length of the straightening machine, as well as the yield strength and thickness of the base material and the yield strength and thickness of the cladding material. The output layers contain the unevenness and maximum residual stress of the straightened sheet material, with one node. The empirical formula for calculating the number of hidden layer elements is:

[0036]

[0037] Where l is the number of units in the hidden layer, n is the number of units in the input layer, m is the number of units in the output layer, and a is a constant between 1 and 10.

[0038] The neurons in the input and output layers are normalized using the following formula:

[0039]

[0040] Where, x i Let x be the eigenvalues ​​before normalization and x be the eigenvalues ​​after normalization. max x represents the maximum value of the eigenvalues ​​before normalization. min These are the sum and minimum of the eigenvalues ​​before normalization, respectively.

[0041] In this embodiment, the tan-sigmoid activation function is used between the input and hidden layers of the BP network structure, and the purelin function is used for the output layer. The training iterations are set to 50, and the training target is set to 10. -5 The learning rate was set to 0.05, and Equation 1 was used as the evaluation function.

[0042] err=abs((T_sim-T_test). / T_test) (1)

[0043] Where: err is the target error value, T_sim is the experimental value, and T_test is the target value.

[0044] The two objective functions, flatness and maximum residual stress of the straightened plate, are calculated according to Formula 1.

[0045] The steps of optimizing a backpropagation (BP) neural network using the Particle Swarm Optimization (PSO) algorithm include: reading multiple objective functions of the BP algorithm, initializing the particle swarm, updating the velocity and position of each particle, calculating the fitness value of each particle, calculating the global minimum fitness value of the particle swarm, and outputting the network weights and thresholds determined by the location of the global extremum. This involves detecting the dominance relationships between targets, selecting Pareto fronts and mutation operations, and boundary detection.

[0046] Initializing a multi-objective particle swarm includes setting the multi-objective function, the size of the particles and the position and velocity vectors of each particle, the iteration error precision, constant coefficients c1 and c2, the inertia weight ω, the initial velocity, and the maximum number of iterations.

[0047] in:

[0048] ω=1 / (1+1.5*exp(-2.6*h))

[0049] c1 = 5 / (1 + 9*exp(-2.18*h))

[0050] c2 = 5 / (1 + exp(-2.2 * h))

[0051] The formula for iterative update of velocity and position is:

[0052]

[0053]

[0054] in: The velocity of the k+1 generation particle. Let K be the velocity of the particle. The position of the k-th generation particle. Let be the individual optimal value of the particle population in generation k. Let be the population optimum for the k-th generation particle population. Let r1 be the position of the particle in generation k+1. r1 and r2 are random numbers in the interval [0,1].

[0055] The fitness value formula is:

[0056]

[0057] Where b is the number of training samples, O iq The actual output T determined by the position of the training sample q at the i-th particle.iq The desired output is determined for the position of the i-th particle in the training sample q.

[0058] Based on the Pareto dominance principle, the individual optimal solution is calculated, the mesh is upgraded, Pareto front points are detected, the mesh is divided equally, and Pareto points on each mesh are selected to ultimately determine the mesh quality. Finally, the mesh with the fewest Pareto front points is selected, i.e., the mesh with the optimal point for both objectives.

[0059] The node with the fewest Pareto front points is selected as the leader, i.e., the globally optimal solution. The initial value is used as the individual optimal value.

[0060] Secondly, by performing mutation operations on the PSO population, local optima are prevented from being reached. Boundary checks are also performed to prevent the mutation from failing to meet the conditions.

[0061] After training the multi-objective PSO-BP neural network, this invention calculates the post-correction unevenness and maximum residual stress of a set of experimental values. The results are shown in Table 1, and the errors are all less than 10%, indicating that it can be used in practice. Simultaneously, it determines whether the straightening of 10 sets of data was successful, and the results are shown in Table 2.

[0062] Table 1 Comparison of predicted and actual values ​​of maximum residual stress unevenness in this invention

[0063]

[0064] Table 2 Comparison of predicted and actual straightening success rates in this invention (1 represents successful straightening, 0 represents failed straightening).

[0065]

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the straightening rate of a metal-based composite plate, characterized in that, The prediction method includes the following steps: S1: Collect experimental data during the straightening process of the metal matrix composite plate, as well as the flatness and maximum residual stress value of the plate after straightening, and generate training set data and test set data for the BP network. S2: Construct the topology of two BP neural networks based on the training and test data of the BP network, determine the number of neurons in the input layer, hidden layer, and output layer, and select the activation function of the neural network; S3: Initialize the weights and thresholds of the BP neural network, evaluate them using the prediction function, and determine the multi-objective function as the basic function of PSO; S4: Initialize the dimensions of the PSO particle position and velocity vectors, and set the basic parameters of the multi-objective PSO algorithm, including the number of particles, particle length, particle range, maximum velocity, learning factor, termination condition, and inertia weight. S5: Calculate the individual optimal value of a particle. Randomly input a particle, calculate the initial fitness value, and obtain the individual optimal solution for each particle. S6: Calculate the optimal value of the particle swarm, find the Pareto optimal solution based on the dominance relationship, and select the one with the fewest Pareto front points as the leader, which is the global optimal solution. S7: Update the position and velocity update formulas for particles, update the position and velocity of each particle, and consider whether the updated velocity and position are within the specified range; then update the individual optimal value and the population optimal value of the particles based on the recalculated fitness value; determine whether the iteration stopping condition is met. If not, re-detect the dominance relationship and iterate the calculation of the particle fitness value until the condition is met.

Images