CN-116127797-B - Method for predicting elbow cross-section deformation by reinforced QPSO-BPNN with informed initialization

CN116127797BCN 116127797 BCN116127797 BCN 116127797BCN-116127797-B

Abstract

The invention discloses a method for predicting the cross-section deformation of an elbow by reinforced QPSO-BPNN with informed initialization. The method comprises the steps of sampling the determined bent pipe process parameter variables by using Latin hypercube uniform sampling technology, establishing a finite element numerical simulation test, calculating bent pipe section deformation indexes corresponding to each group of sample input so as to construct a sample data set, establishing a BPNN prediction model, optimizing the initial weight and the threshold value of the BPNN by using an informed enhancement QPSO, introducing Gaussian random vector and a self-adaptive parameter adjustment strategy to enhance the QPSO optimization performance, carrying out informed initialization of the population based on bent pipe priori knowledge, accelerating the algorithm optimizing process, training the optimized BPNN prediction model, and predicting the bent pipe section deformation by using the model which is completely trained. The invention realizes flexible, reliable, accurate and efficient prediction of the cross section deformation of the bent pipe.

Inventors

ZHANG SHUYOU
WANG CAICHENG
WANG ZILI
LI RUISEN
TAN JIANRONG

Assignees

浙江大学

Dates

Publication Date: 20260508
Application Date: 20221208

Claims (6)

1. A method for predicting elbow cross-sectional deformation by reinforced QPSO-BPNN with informed initialization, comprising the steps of: step 1, latin hypercube sampling is carried out on process parameter variables in a metal pipe bending forming process, the value range of each process parameter variable is taken as input, and a plurality of groups of sampled process parameter samples are obtained by outputting; Step 2, according to the sampling condition of the technological parameters, establishing a plurality of groups of finite element numerical simulation tests, calculating bent pipe section deformation indexes corresponding to each group of inputs, constructing a sample data set and normalizing; Step3, constructing a BPNN prediction model, taking the technological parameter variable in the step 1 as an input variable and the corresponding bent pipe section deformation index as an output variable, and determining the hidden layer structure of the BPNN by setting the number of hidden layer nodes and the number of layers; Step 4, adopting reinforced QPSO with informed initialization to optimize the initial weight and threshold of the BPNN prediction model, introducing Gaussian random vector and self-adaptive parameter adjustment strategy to reinforce the optimization performance of the QPSO, carrying out informed initialization on the population based on the prior knowledge of the bent pipe, and accelerating the algorithm optimizing process; Training the optimized BPNN prediction model by using the normalized sample data set in the step 2 to generate a bend section deformation prediction model, and inputting each group of process parameter samples of the to-be-tested bend into the bend section deformation prediction model after training so as to complete the deformation prediction of the bend section; In the step 1, the process parameter variables include pipe diameter Thickness ratio Ratio of boosting distance of pressure block to bending arc length Angle of bend Radius of curvature of relative tube Coefficient of friction of pressure block Coefficient of friction of anti-wrinkling block Coefficient of friction of rotary bending die Gap ratio of pressure block Gap ratio of rotary bending die ; The step 2 specifically comprises the following steps: 2.1 According to each group of process parameter samples in the step 1, establishing a finite element numerical model corresponding to each group of samples; 2.2 Calculating bent pipe section deformation indexes corresponding to each set of process parameter samples based on a plurality of sets of finite element numerical models, thereby constructing a sample data set and normalizing the sample data set; in the step 2.2), the specific steps for calculating the deformation index of the cross section of the bent pipe are as follows: 1) Starting from the initial surface of the bent section of tube at the same bending angle Uniformly cutting the bent section of the pipe as intervals to obtain N bent pipe cross sections, wherein, = N, N is the set interception number, Is the bending angle of the pipe; 2) Calculating the short axis change rate of each bent pipe cross section : Wherein, the Is the original diameter of the pipe before bending, Is the length of the short axis of the cross section of the pipe after bending; 3) For all cross sections taken out, the bending angles of the cross sections are respectively equal to that of the cross sections As independent variable, short axis change rate of corresponding section As dependent variable, all points are processed by a cubic spline interpolation method , ) Fitting to obtain a section deformation curve of the pipe bending section, ; 4) And calculating the average value of the short axis change rate on the section deformation curve to obtain the size of the deformation index of the elbow section.
2. The method for predicting elbow cross-sectional deformation with informed-initialization enhanced QPSO-BPNN as recited in claim 1, wherein the specific steps of step 4 include: 4.1 Taking the initial weight and the threshold value of the BPNN as the searching dimension d of the particles, and setting the number of the particles in the population; 4.2 Initializing the particle population in an informed manner; 4.3 Taking the mean square error of the BPNN as the fitness function of the QPSO, and calculating the fitness value of each particle according to the fitness function f (x); 4.4 Updating individual optimal positions of particles And global optimal position of population , And The calculation formula of (2) is as follows: Wherein, the Means to make At minimum T is the number of iterations, t is, Is the position of the ith particle at the t-th iteration; Initial individual optimal position Namely primary particles ; 4.5 Updating the current position of the particle; 4.6 Judging whether the iteration number t reaches the set maximum iteration number : If yes, iteration is ended, the global optimal position of the population is output, and the global optimal position of the population is used as an initial weight and a threshold value after the BPNN prediction model is optimized; Otherwise, returning to the step 4.3).
3. The method for predicting elbow cross-sectional deformation with informed-initialization of the reinforced QPSO-BPNN according to claim 2, wherein the step 4.2) is specifically: the initialization of the knowledge means that the d dimension of the primary population adopts Gaussian distribution Initializing, wherein D is more than or equal to 1 and less than or equal to D, D is the total search dimension of the particles, and C (D) and S (D) are manually set parameters; C (d) is the center of Gaussian distribution initialization, the sensitivity of the bent pipe process parameters to the bent pipe section deformation index is analyzed based on a Sobol' sensitivity analysis method, the parameters with the sensitivity value larger than 0.3 are regarded as parameters with larger influence on the bent pipe section deformation, C (d) in the dimension related to the parameters is set to 0.1, and C (d) in the other dimensions is set to 0.
4. The method for predicting elbow cross-sectional deformation with informed-initialization and reinforced QPSO-BPNN as claimed in claim 3, wherein said step 4.5) is specifically: The current location of the particle updates the formula: Wherein, the Is the position of the ith particle at the t+1st iteration; is the local attractor of the ith particle at the t iteration; For shrinking expansion coefficient, adopting a linear decreasing strategy to carry out self-adaptive adjustment; Is a gaussian random vector introduced; is a random number uniformly generated between [0, 1); the calculation formula of the local attractor is as follows: Wherein, the Is a random number uniformly generated between [0, 1 ].
5. The method for predicting elbow cross-sectional deformation with informed-initialization of enhanced QPSO-BPNN of claim 4, wherein Gaussian random vector The d-th dimension of (c) is calculated by: Wherein M is the total number of particles in the population, The d-th search dimension, which is the individual optimal position of the i-th particle.
6. The method for predicting elbow cross-sectional deformation with informed-initialization of the enhanced QPSO-BPNN of claim 5, The adaptive adjustment formula of (2) is: Wherein, the Representing the number of iterations as t Is a function of the number of (c), 、 Respectively represent The upper and lower limits of the value are taken, Representing the maximum number of iterations.

Description

Method for predicting elbow cross-section deformation by reinforced QPSO-BPNN with informed initialization Technical Field The invention relates to the field of bend forming quality prediction, in particular to a bend section deformation prediction direction, and specifically relates to a method for predicting bend section deformation by reinforced QPSO-BPNN with informed initialization. Background The metal bent pipe is widely used as a carrier for transporting various liquid and gas fuels in the fields of aviation, aerospace, automobiles, ships and the like, and is called as an industrial blood vessel. The forming conditions of the bent pipe are extremely complex, and various forming defects such as cross section deformation, unloading rebound, pipe wall thickening and wrinkling, pipe wall thinning and stretch cracking are caused. These defects affect the product quality and the service performance of the bent pipe member to different degrees, and cause unavoidable engineering loss, and even safety accidents in severe cases. The deformation of the cross section of the bent pipe is a serious defect generated in the pipe bending forming process, when fluid is transmitted, the deformation of the cross section can cause the increase of head loss and pressure drop in the bent pipe, reduce the flow rate and the flow velocity of the fluid in the pipe and influence the service performance of the bent pipe. Therefore, in order to improve the production quality of the bent pipe, ensure the safety and reliability of the bent pipe component in the use process, and accurately predict the deformation of the cross section is an important problem to be solved urgently. At present, three main types of prediction methods for the deformation of the cross section of the bent pipe are theoretical analysis, experimental analysis and finite element numerical analysis. Because the pipe bending and forming process is coupled with various complex factors, a great amount of assumptions and simplification are needed in theoretical analysis, and a great deviation is generated between a theoretical analysis result and an actual result. Experimental analysis can provide more accurate prediction results, but is often accompanied by high cost and material waste issues. Finite element numerical analysis can simulate the actual process of pipe bending, providing accurate predictions of forming defects, but high-precision finite element simulation can create a significant computational burden. Therefore, there is a need to develop a flexible and reliable prediction method to realize rapid and accurate prediction of the deformation of the elbow cross section. Disclosure of Invention In order to solve the defects of the technology, the invention provides a bend section deformation prediction method for enhancing the knowledge QPSO-BPNN so as to realize flexible, rapid and accurate prediction of the bend section deformation. The technical scheme adopted by the invention comprises the following steps: step 1, latin hypercube sampling is carried out on process parameter variables in a metal pipe bending forming process, the value range of each process parameter variable is taken as input, and a plurality of groups of sampled process parameter samples are obtained by outputting; Step 2, according to the sampling condition of the technological parameters, establishing a plurality of groups of finite element numerical simulation tests, calculating bent pipe section deformation indexes corresponding to each group of inputs, constructing a sample data set and normalizing; Step3, constructing a BPNN prediction model, taking the technological parameter variable in the step 1 as an input variable and the corresponding bent pipe section deformation index as an output variable, and determining the hidden layer structure of the BPNN by setting the number of hidden layer nodes and the number of layers; Aiming at the problems of poor diversity and premature convergence of the QPSO, which are easy to occur, a Gaussian random vector and self-adaptive parameter adjustment strategy is introduced to strengthen the optimization performance of the QPSO, and the population is subjected to the informed initialization based on the prior knowledge of the bent pipe so as to accelerate the algorithm optimizing process; And 5, training the optimized BPNN prediction model by using the normalized sample data set in the step 2 to generate a bend section deformation prediction model, and inputting each group of process parameter samples of the to-be-tested bend into the bend section deformation prediction model after training, so as to complete the deformation prediction of the bend section. In the step 1, the process parameter variables include a pipe diameter d 0, a thickness ratio t 0/d0, a ratio L p/Lb of a boosting distance of a pressure block and a bending arc length, a bending angle theta 0, a relative pipe bending radius R 0/d0, a pressure block friction coefficient f