Reliability-based titanium alloy medium plate welding optimization control system and control method for new energy vehicle

Abstract

The invention discloses a reliability-based titanium alloy medium plate welding optimization control system and method for a new energy vehicle, and the system comprises a data obtaining and preprocessing module, a performance prediction module, a reliability evaluation module and a process parameter optimization module, welding process parameters and the tensile strength of a welding joint, the welding width of the bottom surface of a welding seam and angular deformation under the corresponding welding process parameters are recorded; the performance prediction module is used for training and learning the sample data acquired in the data acquisition module, the trained performance prediction module is used for prediction, and the weight of an RBF neural network and the initial value of a Gaussian function parameter are optimized based on an improved genetic algorithm in the module; the reliability evaluation module is used for establishing a welding non-probabilistic reliability model based on an ellipsoid convex model and considering the tensile strength of a welding joint, the weld penetration depth and deformation, and evaluating the rationality of the current process parameter group by comparing the reliability obtained through calculation with a reliability threshold value; the process parameter optimization module is used for constructing a fitness function about the tensile strength of a welding joint, the weld width of the bottom surface of a welding seam and angular deformation by adopting a weight method, and optimizing an optimal process parameter group based on a particle swarm optimization algorithm of simulated annealing; according to the invention, efficient and reliable welding process optimization can be realized, the reliability and the performance stability of a welding joint are ensured, and the experiment trial and error cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of welding titanium alloy medium and thick plates for new energy vehicles, and particularly relates to an optimization control method for the welding process of titanium alloy thin plates. Background Art

[0002] Titanium alloys are widely used in aerospace, marine engineering, medical devices, and other fields due to their excellent strength, corrosion resistance, and low density. Welding titanium alloy thin plates is particularly important in these fields because thin plate structures offer significant advantages in reducing structural weight and improving structural efficiency. However, the welding process of titanium alloy thin plates is complex, and the welding quality is affected by a variety of process parameters, such as welding speed, wire feed speed, pulse current, and pulse frequency. Slight changes in these parameters can lead to significant changes in the mechanical properties, weld morphology, and degree of deformation of the welded joint, thereby affecting the reliability and service life of the welded structure.

[0003] Traditional welding process optimization methods rely primarily on empirical experience or extensive trial and error, which is not only time-consuming and labor-intensive, but also makes it difficult to ensure the stability and reliability of welding quality. Furthermore, existing methods often lack a systematic assessment of welding process reliability, making it difficult to effectively predict the performance of welded joints in actual use. With the increasing demand for welding quality in modern industry, the development of an efficient, accurate, and reliable optimization control method for titanium alloy thin plate welding is becoming increasingly important.

[0004] In recent years, with the development of computing technology, welding process parameter optimization methods based on artificial intelligence and optimization algorithms have gradually attracted attention. Among them, the particle swarm optimization (PSO) algorithm has been widely used in welding process parameter optimization due to its advantages such as simplicity, ease of implementation and fast convergence speed. However, the PSO algorithm is prone to fall into local optimal solutions in complex optimization problems. In order to solve this problem, researchers have proposed an improved PSO algorithm combined with the simulated annealing algorithm (SA). Through the global search capability of SA, PSO is prevented from falling into local optimal solutions. In addition, in order to improve the accuracy and stability of welding performance prediction, a prediction model based on the radial basis function (RBF) neural network has also been introduced into welding process optimization. However, the traditional RBF neural network is easily affected by the initial parameter selection and local optimal solution during the training process, resulting in insufficient prediction accuracy. To this end, the present invention proposes an RBF neural network model based on an improved genetic algorithm (GA), which improves the training effect and prediction accuracy of the model by dynamically adjusting the number of intersections and the adaptive mutation probability.

[0005] In terms of reliability assessment of welding process, traditional probabilistic reliability models require a large amount of experimental data to estimate the probability distribution of parameters, which is often difficult to achieve in practical applications. For this reason, non-probabilistic reliability models have emerged. Among them, the non-probabilistic reliability model based on the ellipsoid convex model is widely used in the reliability assessment of engineering structures because it does not require the probability distribution of assumed parameters and can effectively process uncertainty information. The present invention introduces the non-probabilistic reliability model based on the ellipsoid convex model into the reliability assessment of the titanium alloy thin plate welding process. By constructing the limit state equation, the tensile strength of the weld joint, the weld width at the bottom of the weld and the angular deformation are reliably assessed, providing a theoretical basis for the optimization of welding process parameters.

[0006] In response to the above problems, the present invention proposes a reliability-based optimization control method for titanium alloy thin plate welding. By combining an improved PSO algorithm, an improved GA-RBF neural network model and a welding non-probabilistic reliability model based on an ellipsoid convex model, efficient optimization of the welding process parameters of titanium alloy thin plates is achieved, and the reliability and performance stability of the welded joints are ensured. Summary of the Invention

[0007] In order to solve the above technical problems, the present invention provides a reliability-based titanium alloy medium and thick plate welding optimization control system and control method for new energy vehicles with high efficiency optimization, high reliability, multi-objective optimization and high prediction accuracy, including: a data acquisition module obtains data samples through experiments, a performance prediction module uses welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as input variables, and adopts an improved genetic algorithm to predict the tensile strength, the weld width at the bottom of the weld, and the angular deformation of the RBF neural network model; a reliability evaluation module calculates the reliability of the welding process through the non-probabilistic reliability model and limit state equation of the ellipsoid convex model, and evaluates the rationality of the process parameter group; a process parameter optimization module uses welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as optimization variables, and tensile strength, the weld width at the bottom of the weld, and angular deformation as optimization targets, defines a fitness function based on a weighted weight method, and adopts a particle swarm algorithm based on simulated annealing for optimization, thereby achieving efficient and reliable welding process optimization and ensuring the reliability and performance stability of the welded joint.

[0008] The present invention provides a reliability-based optimization and control system for the welding of titanium alloy medium and thick plates for new energy vehicles, which is applied to the welding of titanium alloy medium and thick plates for new energy vehicles. The control system includes a data acquisition and preprocessing module, a performance prediction module, a reliability evaluation module and a process parameter optimization module.

[0009] The data acquisition module obtains data samples through experiments, systematically changes the welding speed, wire feed speed, pulse frequency and pulse current, conducts multiple welding tests, records the welding process parameters, and measures the tensile strength of the weld joint, the weld bottom width and the angular deformation under the corresponding welding process parameters based on an optical microscope, a three-dimensional scanner and an electronic tensile testing machine.

[0010] The performance prediction module is used to train and learn the sample data obtained in the data acquisition module, and the trained performance prediction module is used for prediction; the performance prediction module includes data normalization processing, training set / test set division, creation of RBF neural network, training of RBF neural network, simulation testing, and performance evaluation. By using welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as input variables, it is used to predict the tensile strength of the weld joint, the weld width and angular deformation of the weld bottom surface, and optimizes the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters based on the improved genetic algorithm.

[0011] The reliability assessment module establishes a non-probabilistic welding reliability model based on the ellipsoid convex model that takes into account the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation. According to the tensile strength of the base material, the minimum weld width at the bottom of the weld required for full penetration, and the maximum angular deformation allowed by the service conditions, the corresponding limit state equations are defined respectively. The reliability based on the ellipsoid convex model is defined as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. The calculated reliability is compared with the reliability threshold to evaluate the rationality of the current process parameter group.

[0012] The process parameter optimization module uses a weighted method to construct a fitness function for the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation. It then searches for the optimal process parameter group based on a particle swarm algorithm based on simulated annealing.

[0013] The performance prediction module is used to train and learn the sample data obtained in the data acquisition module, and the trained performance prediction module is used to perform predictions; the performance prediction module includes data normalization processing, training set / test set division, creation of RBF neural network, training of RBF neural network, simulation testing, and performance evaluation; the creation of RBF neural network consists of three parts: input layer, hidden layer and output layer, the signal source of the input layer includes welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I), and the output layer outputs tensile strength, weld width and angular deformation of the bottom surface of the weld; and, based on the improved genetic algorithm, the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters are optimized, binary encoding is used, the individual length is 10, and an elite retention strategy, a strategy of dynamically adjusting the number of intersections according to population diversity, and an adaptive mutation probability are adopted.

[0014] It should be noted that the optimization of the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters is performed based on the improved genetic algorithm. The specific steps are as follows:

[0015] (1) Randomly generate an initial population with a population size of M. Use a 10-bit binary code string to represent the network weight w, the Gaussian function base width vector b, and the Gaussian function center vector c. Set the crossover probability Pc, the mutation probability Pm, and the termination iteration number t of the genetic algorithm. max ;

[0016] (2) Fitness calculation and evaluation;

[0017] (3) genetic operations, including selection, crossover, and mutation;

[0018] (4) population renewal;

[0019] (5) Termination condition judgment: check whether the maximum number of iterations has been reached or the fitness value has converged. If so, output the optimal solution; otherwise, return to step (2) to continue iteration.

[0020] It should be noted that to prevent the optimal solution from being lost during the evolution of the genetic algorithm, an elite retention strategy is adopted. During each generation of evolution, the fitness of the individuals in the current population is first evaluated. Then, 1 / 3 of the individuals in the population are selected in descending fitness order as elite individuals. These elite individuals replace the worst 1 / 3 of the individuals in the population, while retaining the historical optimal value remembered by each individual. The middle 1 / 3 of the individuals in the population are generated through selection, crossover, and mutation operations.

[0021] It should be noted that in order to take into account the diversity of the population and the computational complexity, a strategy of dynamically adjusting the number of intersections according to the population diversity is adopted; when the population diversity D(t) is lower than the threshold θ l When , it indicates that the population diversity is low, and the number of crossover points can be increased to introduce new gene combinations; when the population diversity D(t) is higher than the threshold θ h When , it indicates that the population diversity is high. In this case, the number of crossover points can be reduced to retain excellent gene combinations. The specific steps for calculating the number of crossover points are as follows:

[0022] (1) Set the number of basic intersections n b , diversity threshold θ l and θ h , adjustment coefficients α and β;

[0023] (2) Calculate the diversity of the population, the formula is:

[0024]

[0025] Where D(t) represents the diversity of the population when the number of iteration steps in the genetic algorithm is t; M represents the population size in the genetic algorithm; U represents the length of the chromosome; represents the kth gene of individual i.

[0026] (3) Calculating the number of intersections

[0027]

[0028] It should be noted that in order to balance the global search ability and local improvement ability of the genetic algorithm, an adaptive mutation probability is adopted, and the mutation probability is taken as:

[0029]

[0030] Among them, f represents the current fitness value of the individual; f avg and f min They represent the average fitness value and minimum fitness value of the current population respectively; for individuals whose current fitness value is better than the average fitness value of the population, the corresponding mutation probability is small, so the individual is retained; conversely, for particles whose current fitness value is worse than the average fitness value of the population, the corresponding mutation probability is large, which makes the individual move towards a better genetic direction.

[0031] The reliability assessment module is composed of a welding non-probabilistic reliability model based on an ellipsoid convex model, a limit state equation, and a rationality judgment of reliability. The welding non-probabilistic reliability model based on an ellipsoid convex model takes welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as uncertain parameters, which are recorded as a vector X = [V, W, f, I] T , and calculate the mean vector μ corresponding to the above uncertain parameters x and covariance matrix ∑ x , according to the mean vector μ x and covariance matrix ∑ x , and after decomposition and transformation, the welding non-probabilistic reliability model based on the ellipsoid convex model is constructed as The limit state equation uses the RBF neural network prediction model based on the improved genetic algorithm in the performance prediction module as the function g(X), and obtains the tensile strength g1(X) of the weld joint, the weld width g2(X) at the bottom of the weld, and the angular deformation g3(X), and the tensile strength R b 80% of the minimum weld width d required for penetration min , the maximum welding angle deformation S allowed under service conditions max As the three safety allowable values ​​K1, K2, and K3, the three limit state equations of welding are obtained: G1(X)=g1(X)-0.8·R b, G2(X)=g2(X)-d min , G3(X)=S max -g3(X); The rationality judgment of the reliability is combined with the non-probabilistic reliability model based on the ellipsoid convex model and the limit state equation, and the reliability is defined as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. If G1≥0, G2≥0 and G3≥0, the process parameters are reasonable.

[0032] It should be noted that the specific steps of constructing the ellipsoidal convex model in the reliability assessment module include:

[0033] (1) The welding speed (V), wire feed speed (W), pulse frequency (f), and pulse current (I) are taken as uncertain parameters and expressed as vector X = [V, W, f, I] T , and for the uncertain parameter X j,i , (j=1,2,...,N;i=1,2,3,4), its value range is X I The possible upper and lower bounds are x R 、x L , whose expression is:

[0034]

[0035] (2) Calculate the mean vector μ of the four uncertain parameters: welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) x and covariance matrix ∑ x , the calculation formula is as follows:

[0036]

[0037] Where N represents the number of samples; X j represents the parameter vector of the jth sample; X j,i represents the i-th uncertain parameter of the j-th sample;

[0038] (3) According to the mean vector μ x and covariance matrix ∑ x , the welding non-probabilistic reliability model based on the ellipsoid convex model is constructed as follows:

[0039]

[0040] (4) For the matrix Perform eigenvalue decomposition:

[0041] Σ X =H T ΛH H T H=I

[0042] Where Λ=diag(λ i ), λ i ,i=1,2,...,4,is the matrix The eigenvalues ​​of ; I is the unit matrix; introduce the vector and make the following mathematical transformation:

[0043]

[0044] y i =HX i i=1,2,3,4

[0045] y o =Hμ X

[0046] Then the ellipsoid model in step (3) can be transformed into:

[0047]

[0048] Furthermore, the ellipsoid model is expanded by combining the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I), and the final welding non-probabilistic reliability model based on the ellipsoid convex model is obtained:

[0049]

[0050] Among them, V o =(V L +V R ) / 2;W o =(W L +W R ) / 2;f o =(f L +f R ) / 2;I o =(I L +I R ) / 2;e V =(V R -V L ) / 2;e W =(W R -W L ) / 2;e f =(f L +f R ) / 2;e I =(I R -I L ) / 2.

[0051] It should be noted that the rationality judgment of reliability in the reliability assessment module defines reliability as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. The specific steps for solving it are as follows:

[0052] (1) The four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) are sampled using the full factor method. Divide it into L equal parts, then the total number of initial sample points falling into the rectangle is L 4 ;

[0053] (2) The sample X obtained in step (1) j =[V j , W j , f j , I j ] T ,j=1,2,...,L 4 , substitute the non-probabilistic reliability model of welding one by one, count the number of samples that meet the conditions, and record it as Q1;

[0054] (3) The sample X obtained in step (2) j =[V j , W j , f j , I j ] T , i=1,2,...,Q1, substitute the three limit state equations of welding one by one: G1(X)=g1(X)-0.8·R b , G2(X)=g2(X)-d min , G3(X)=S max -g3(X), count the number of samples that make G1≥0, G2≥0 and G3≥0, and record it as Q2;

[0055] (4) Calculate the reliability of the welding process β = Q1 / Q2.

[0056] The process parameter optimization module uses a particle swarm algorithm based on simulated annealing to perform target optimization. The algorithm parameters are set as follows: the number of individuals in the initialization group N, the learning factor C1, the learning factor C2, the annealing constant inertia weight λ, the maximum number of iterations M max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T The welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) are used as optimization variables. According to the tensile strength, weld width at the bottom of the weld, and angular deformation obtained by the performance prediction module, the reliability of the current process parameter group is evaluated through the reliability evaluation module. If the current process parameter group is reasonable, the tensile strength, weld width at the bottom of the weld, and angular deformation are used as optimization targets. The fitness function is defined based on the weighted weight method, and the particle swarm algorithm is used for optimization. In the optimization process, in order to avoid falling into a local solution, the ability of the simulated annealing algorithm to jump during the search process is combined.

[0057] It should be noted that the process parameter optimization module performs target optimization through a particle swarm algorithm based on simulated annealing. The specific steps are as follows:

[0058] (1) Randomly set the speed and position of each particle, where the number of individuals in the initial group is N = L 4 The spatial dimension of each particle is 4, which corresponds to the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I), and the upper and lower boundary values ​​corresponding to the four dimensions of particle position are (j=1,2,...,L 4 ; i=1,2,3,4);

[0059] (2) Based on the position of each particle, the performance prediction module is used to predict the tensile strength, weld width at the bottom of the weld, and angular deformation of the corresponding sample;

[0060] (3) Combining the current particle swarm and the predicted value obtained by the performance prediction module, the reliability β of the current process parameter group is obtained through the reliability evaluation module, and compared with the reliability threshold. If the reliability is satisfied, step (4) is performed, otherwise, return to step (1);

[0061] (4) According to the three predicted values ​​obtained by the performance prediction module, the fitness function is defined as the weighted sum of the three predicted values, with weights w1, w2, and w3 respectively. The fitness function formula is:

[0062] P j =w1×g1(X)+w2×g2(X)-w3×g3(X)

[0063] (5) Store the particle's position and fitness value in the particle's individual extreme value P g In the g The individual position and fitness of the best fitness value in the global extreme value P Z middle;

[0064] (6) Determine the initial temperature T0 = f(P g ) / In5;

[0065] (7) Determine the current temperature of each particle P j The fitness value is calculated as follows:

[0066]

[0067] (8) From all P j Determine the global optimal replacement value P Z , and update the position and velocity of each particle according to the following two formulas:

[0068] X j,i (m+1)=X j,i (m)+v j,i (m+1)

[0069]

[0070]

[0071] Among them, m represents the current iteration number; j represents the number of particles; i represents the dimension of the particle.

[0072] (9) Calculate the particle fitness value and update P g and P Z , and then perform the cooling operation as follows:

[0073] T m+1 =λT m

[0074] (10) When the number of iterations reaches the maximum number of iterations, the search stops and the results are output. Otherwise, the search returns to step (3) to continue.

[0075] A reliability-based optimization control system for welding titanium alloy medium and thick plates for new energy vehicles using the above reliability-based optimization control system mainly includes the following steps:

[0076] S1: Initialize the position and velocity of the particles, where the number of individuals in the initial group N = L 4 The spatial dimension of each particle is 4, corresponding to the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I). The learning factor C1, learning factor C2, annealing constant inertia weight λ, and maximum number of iterations M are set. max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T ;

[0077] S2: The tensile strength of the weld joint, the weld width at the bottom, and the angular deformation are predicted through the RBF neural network based on the improved genetic algorithm in the performance prediction module.

[0078] In order to apply the performance prediction module to predict the tensile strength of welded joints, weld width at the bottom of the weld, and angular deformation, it is first necessary to train and learn the sample points. To improve the accuracy and stability of the prediction, the present invention uses a BRF neural network model based on an improved genetic algorithm to train and predict samples. The genetic algorithm adopts an elite retention strategy, a strategy for dynamically adjusting the number of intersections based on population diversity, and an adaptive mutation probability. The specific steps for calculating the number of intersections in the genetic algorithm are as follows:

[0079] ① Set the number of basic intersections n b , diversity threshold θ l and θ h , adjustment coefficients α and β;

[0080] ②Calculate the diversity of the population, the formula is:

[0081]

[0082] Where D(t) represents the diversity of the population when the number of iteration steps in the genetic algorithm is t; M represents the population size in the genetic algorithm; U represents the length of the chromosome; represents the kth gene of individual i.

[0083] ③Calculation of the number of intersections:

[0084]

[0085] S3: Based on the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation obtained in step S2, the reliability is calculated using a non-probabilistic reliability model based on the convex ellipsoid model. This reliability is then compared with the reliability threshold to determine the rationality of the current set of process parameters. If the reliability threshold is met, proceed to step S4; otherwise, return to step S1.

[0086] In order to use the reliability assessment module to judge the rationality of the current process parameter set, it is necessary to first construct a non-probabilistic reliability model based on the ellipsoidal convex model and the limit state equation, and then combine the two to calculate the reliability. The specific steps are as follows:

[0087] ① Sample X j =[V j , W j , f j , I j ] T ,j=1,2,...,L 4 , one by one into the non-probabilistic reliability model of welding, and statistically analyze the samples that meet the conditions

[0088] Quantity, and record it as Q1;

[0089] ② The sample X obtained in step ① j =[V j , W j , f j , I j ] T , j=1,2,...,Q1, and substitute the three limit state equations of welding one by one: G1(X)=g1(X)-0.8·R b , G2(X)=g2(X)-d min, G3(X)=S max -g3(X), count the number of samples that make G1≥0, G2≥0 and G3≥0, and record it as Q2;

[0090] ③Calculate the reliability of the welding process β=Q1 / Q2.

[0091] S4: The process parameter optimization module is used to perform target optimization, with welding speed (V), wire feed speed (W), pulse frequency (f), and pulse current (I) as optimization variables, and tensile strength, weld width at the bottom of the weld, and angular deformation as optimization targets. Based on the three predicted values ​​obtained by the performance prediction module, the fitness function is defined as the weighted sum of these three predicted values. In the optimization process, to avoid falling into local solutions, the simulated annealing algorithm is combined with its ability to jump during the search process.

[0092] S5: Store the particle's position and fitness value in the particle's individual extreme value P g In the g The individual position and fitness of the best fitness value in the global extreme value P Z middle;

[0093] S6: Determine the initial temperature T0 = f(Pg) / In5;

[0094] S7: Determine the current temperature of each particle P j The fitness value of

[0095] S8: From all P j Determine the global optimal replacement value P Z , and update the position and velocity of each particle according to the following two formulas;

[0096] S9: Calculate the particle target value and update P g and P Z , and then perform cooling operation;

[0097] S10: When the number of iterations reaches the maximum number of iterations, the search stops and the result is output, otherwise the search returns to step S3 to continue.

[0098] Compared with the prior art, the present invention has the following beneficial effects:

[0099] The present invention dynamically adjusts the number of intersections and the adaptive mutation probability according to the population diversity, and can better adapt to the evolutionary state of the population. When the population diversity is high, increasing the number of intersections can promote gene recombination and diversity, and avoid premature convergence; when the population is close to the optimal solution, adaptively reducing the mutation probability can improve the convergence accuracy of the algorithm. This dynamic adjustment strategy significantly improves the training efficiency and prediction accuracy of the model, thereby accelerating the execution speed of the entire welding optimization control method. The present invention introduces a non-probabilistic reliability model based on an ellipsoidal convex model in welding. The model does not require the probability distribution of the assumed parameters and can effectively handle uncertainty only through interval information. By constructing the limit state equation, the reliability of the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation can be accurately evaluated. This non-probabilistic reliability model can also provide reliable evaluation results when there is insufficient data, thereby enhancing the reliability evaluation capability of the welding process. The present invention sets a reliability threshold (β T ), the calculated reliability (β) is compared with the threshold. Only when the reliability meets the threshold is the current process parameter set considered reasonable, and the next optimization step is initiated. This reliability judgment mechanism ensures the reliability and stability of the optimized welding process parameters in practical applications, effectively reducing the risk of weld joint failure. This invention utilizes an improved particle swarm optimization algorithm, combined with the global search capabilities of a simulated annealing algorithm, to effectively prevent the particle swarm algorithm from falling into local optimal solutions.

[0100] The present invention considers multiple performance indicators, defines a fitness function for multi-objective optimization, and constructs a particle swarm optimization model based on simulated annealing to efficiently optimize welding process parameters such as welding speed, wire feed speed, pulse current, and pulse frequency. At the same time, an RBE neural network model based on an improved genetic algorithm is used to predict the tensile strength of welded joints, the weld width at the bottom of the weld, and angular deformation. To improve prediction accuracy, an elite retention strategy, a dynamic adjustment strategy for the number of intersections, and an adaptive mutation probability are adopted. Furthermore, a non-probabilistic reliability model based on an ellipsoidal convex model is constructed, combined with a reliability threshold judgment mechanism to ensure the reliability of the welding process. These innovations can achieve efficient and reliable welding process optimization, ensure the reliability and performance stability of welded joints, and reduce experimental trial and error costs, thus having important practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 Schematic diagram of a reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles according to the present invention;

[0102] Figure 2 The present invention is a flow chart of a reliability-based optimization control system method for welding medium and thick titanium alloy plates for new energy vehicles. DETAILED DESCRIPTION

[0103] The following is combined with Figure 1 and Figure 2 The present invention will be further described.

[0104] like Figure 1 As shown, the reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles of the present invention includes a data acquisition and preprocessing module, a performance prediction module, a reliability evaluation module and a process parameter optimization module.

[0105] The data acquisition module obtains data samples through experiments, systematically changes the welding speed, wire feed speed, pulse frequency and pulse current, conducts multiple welding tests, records the welding process parameters, and measures the tensile strength of the weld joint, the weld bottom width and the angular deformation under the corresponding welding process parameters based on an optical microscope, a three-dimensional scanner and an electronic tensile testing machine.

[0106] The performance prediction module is used to train and learn the sample data obtained in the data acquisition module, and the trained performance prediction module is used for prediction; the performance prediction module includes data normalization processing, training set / test set division, creation of RBF neural network, training of RBF neural network, simulation testing, and performance evaluation. By using welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as input variables, it is used to predict the tensile strength of the weld joint, the weld width and angular deformation of the weld bottom surface, and optimizes the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters based on the improved genetic algorithm.

[0107] The reliability assessment module establishes a non-probabilistic welding reliability model based on the ellipsoid convex model that takes into account the tensile strength, weld penetration and deformation of the weld joint. According to the tensile strength of the base material, the minimum weld bottom width required for full penetration, and the maximum welding deformation allowed by the service conditions, the corresponding limit state equations are defined respectively. The reliability based on the ellipsoid convex model is defined as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. The calculated reliability is compared with the reliability threshold to evaluate the rationality of the current process parameter group.

[0108] The process parameter optimization module uses a weighted method to construct a fitness function for the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation. It then searches for the optimal process parameter group based on a particle swarm algorithm based on simulated annealing.

[0109] The performance prediction module is used to train and learn the sample data obtained in the data acquisition module. The trained performance prediction module is used to perform prediction. The performance prediction module includes data normalization processing, training set / test set division, creation of RBF neural network, training of RBF neural network, simulation testing, and performance evaluation; the creation of RBF neural network consists of three parts: input layer, hidden layer and output layer. The signal source of the input layer includes welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I); the output layer outputs tensile strength, weld width and angular deformation of the bottom surface of the weld; based on the improved genetic algorithm, the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters are optimized, binary encoding is used, the individual length is 10, and an elite retention strategy, a strategy of dynamically adjusting the number of intersections according to population diversity, and an adaptive mutation probability are adopted.

[0110] It should be noted that the optimization of the weights of the RBF neural network and the initial values ​​of the Gaussian function parameters is performed based on the improved genetic algorithm. The specific steps are as follows:

[0111] (1) Randomly generate an initial population with a population size of M. Use a 10-bit binary code string to represent the network weight w, the Gaussian function base width vector b, and the Gaussian function center vector c. Set the crossover probability Pc, the mutation probability Pm, and the termination iteration number t of the genetic algorithm. max ;

[0112] (2) Fitness calculation and evaluation;

[0113] (3) genetic operations, including selection, crossover, and mutation;

[0114] (4) population renewal;

[0115] (5) Termination condition judgment: check whether the maximum number of iterations has been reached or the fitness value has converged. If so, output the optimal solution; otherwise, return to step (2) to continue iteration.

[0116] It should be noted that to prevent the optimal solution from being lost during the evolution of the genetic algorithm, an elite retention strategy is adopted. During each generation of evolution, the fitness of the individuals in the current population is first evaluated. Then, 1 / 3 of the individuals in the population are selected in descending fitness order as elite individuals. These elite individuals replace the worst 1 / 3 of the individuals in the population, while retaining the historical optimal value remembered by each individual. The middle 1 / 3 of the individuals in the population are generated through selection, crossover, and mutation operations.

[0117] It should be noted that in order to take into account the diversity of the population and the computational complexity, a strategy of dynamically adjusting the number of intersections according to the population diversity is adopted; when the population diversity D(t) is lower than the threshold θ lWhen , it indicates that the population diversity is low, and the number of crossover points can be increased to introduce new gene combinations; when the population diversity D(t) is higher than the threshold θ h When , it indicates that the population diversity is high. In this case, the number of crossover points can be reduced to retain excellent gene combinations. The specific steps for calculating the number of crossover points are as follows:

[0118] (1) Set the number of basic intersections n b , diversity threshold θ l and θ h , adjustment coefficients α and β;

[0119] (2) Calculate the diversity of the population, the formula is:

[0120]

[0121] Where D(t) represents the diversity of the population when the number of iteration steps in the genetic algorithm is t; M represents the population size in the genetic algorithm; U represents the length of the chromosome; represents the kth gene of individual i.

[0122] (3) Solving the number of intersections nc

[0123]

[0124] It should be noted that in order to balance the global search ability and local improvement ability of the genetic algorithm, an adaptive mutation probability is adopted, and the mutation probability is taken as:

[0125]

[0126] Among them, f represents the current fitness value of the individual; f avg and f min They represent the average fitness value and minimum fitness value of the current population respectively; for individuals whose current fitness value is better than the average fitness value of the population, the corresponding mutation probability is small, so the individual is retained; conversely, for particles whose current fitness value is worse than the average fitness value of the population, the corresponding mutation probability is large, which makes the individual move towards a better genetic direction.

[0127] The reliability assessment module is composed of a welding non-probabilistic reliability model based on an ellipsoid convex model, a limit state equation, and a rationality judgment of reliability. The welding non-probabilistic reliability model based on an ellipsoid convex model takes welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as uncertain parameters, which are recorded as a vector X = [V, W, f, I] T , and calculate the mean vector μ corresponding to the above uncertain parameters x and covariance matrix ∑ x , according to the mean vector μx and covariance matrix ∑ x , and after decomposition and transformation, the welding non-probabilistic reliability model based on the ellipsoid convex model is constructed as The limit state equation uses the RBF neural network prediction model based on the improved genetic algorithm in the performance prediction module as the function g(X), and obtains the tensile strength g1(X) of the weld joint, the weld width g2(X) at the bottom of the weld, and the angular deformation g3(X), and the tensile strength R b 80% of the minimum weld bottom surface required for penetration min , the maximum welding angle deformation S allowed under service conditions max As the three safety allowable values ​​K1, K2, and K3, the three limit state equations of welding are obtained: G1(X)=g1(X)-0.8·R b , G2(X)=g2(X)-d min , G3(X)=S max -g3(X); The rationality judgment of the reliability is combined with the non-probabilistic reliability model based on the ellipsoid convex model and the limit state equation, and the reliability is defined as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. If G1≥0, G2≥0 and G3≥0, the process parameters are reasonable.

[0128] It should be noted that the specific steps of constructing the ellipsoidal convex model in the reliability assessment module include:

[0129] (1) The welding speed (V), wire feed speed (W), pulse frequency (f), and pulse current (I) are taken as uncertain parameters and expressed as vector X = [V, W, f, I] T , and for the uncertain parameter X j，i , (j=1,2,...,N;i=1,2,3,4), its value range is X I The possible upper and lower bounds are x R 、x L , whose expression is:

[0130] X∈X I =[X L ,X R ]

[0131] (2) Calculate the mean vector μ of the four uncertain parameters: welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) x and covariance matrix ∑ x , the calculation formula is as follows:

[0132]

[0133] Where N represents the number of samples; X j represents the parameter vector of the jth sample; X j,i represents the i-th uncertain parameter of the j-th sample;

[0134] (3) According to the mean vector μ x and covariance matrix ∑ x , the welding non-probabilistic reliability model based on the ellipsoid convex model is constructed as follows:

[0135]

[0136] (4) For the matrix Perform eigenvalue decomposition:

[0137] Σ X =H T ΛH H T H=I

[0138] Where Λ=diag(λ i ), λ i ,i=1,2,...,4,is the matrix The eigenvalues ​​of ; I is the unit matrix; introduce the vector and make the following mathematical transformation:

[0139]

[0140] y i =HX i i=1,2,3,4

[0141] y o =Hμ X

[0142] Then the ellipsoid model in step (3) can be transformed into:

[0143]

[0144] Furthermore, the ellipsoid model is expanded by combining the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I), and the final welding non-probabilistic reliability model based on the ellipsoid convex model is obtained:

[0145]

[0146] Among them, V o =(V L +V R ) / 2;W o =(W L +W R ) / 2;f o =(f L +fR ) / 2;I o =(I L +I R ) / 2;

[0147] e V =(V R -V L ) / 2;e W =(W R -W L ) / 2;e f =(f L +f R ) / 2;e I =(I R -I L ) / 2.

[0148] It should be noted that the rationality judgment of reliability in the reliability assessment module defines reliability as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area, and the steps for solving it are as follows:

[0149] (1) The four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) are sampled using the full factor method. The interval [X L ,X R ] is divided into L equal parts, then the total number of initial sample points falling into the rectangle is L 4 ;

[0150] (2) The sample X obtained in step (1) j =[V j , W j , f j , I j ] T ,j=1,2,...,L 4 , substitute the non-probabilistic reliability model of welding one by one, count the number of samples that meet the conditions, and record it as Q1;

[0151] (3) The sample X obtained in step (2) j =[V j , W j , f j , I j ] T , j=1,2,...,Q1, and substitute the three limit state equations of welding one by one: G1(X)=g1(X)-0.8·R b , G2(X)=g2(X)-d min , G3(X)=S max -g3(X), count the number of samples that make G1≥0, G2≥0 and G3≥0, and record it as Q2;

[0152] (4) Calculate the reliability of the welding process β = Q1 / Q2.

[0153] The process parameter optimization module uses a particle swarm algorithm based on simulated annealing to perform target optimization. The algorithm parameters are set as follows: the number of individuals in the initialization group N, the learning factor C1, the learning factor C2, the annealing constant inertia weight λ, the maximum number of iterations M max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T The welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) are used as optimization variables. According to the tensile strength, weld bottom penetration, and angular deformation obtained by the performance prediction module, the reliability of the current process parameter group is evaluated through the reliability evaluation module. If the current process parameter group is reasonable, the tensile strength, weld bottom penetration, and weld angular deformation are used as optimization targets. The fitness function is defined based on the weighted weight method, and the particle swarm algorithm is used for optimization. In the optimization process, in order to avoid falling into a local solution, the ability of the simulated annealing algorithm to jump during the search process is combined.

[0154] It should be noted that the process parameter optimization module performs target optimization through a particle swarm algorithm based on simulated annealing. The specific steps are as follows:

[0155] (1) Randomly set the speed and position of each particle, where the number of individuals in the initial group is N = L 4 The spatial dimension of each particle is 4, which corresponds to the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I), and the upper and lower boundary values ​​of the four dimensions of particle position are (j=1,2,...,L 4 ; i=1,2,3,4);

[0156] (2) Based on the position of each particle, the performance prediction module is used to predict the tensile strength, weld width and angular deformation of the corresponding sample;

[0157] (3) Combining the current particle swarm and the predicted value obtained by the performance prediction module, the reliability β of the current process parameter group is obtained through the reliability evaluation module, and compared with the reliability threshold. If the reliability is satisfied, step (4) is performed, otherwise, return to step (1);

[0158] (4) According to the three predicted values ​​obtained by the performance prediction module, the fitness function is defined as the weighted sum of the three predicted values, with weights w1, w2, and w3 respectively. The fitness function formula is:

[0159] P j=w1×g1(X)+w2×g2(X)-w3×g3(X)

[0160] (5) Store the particle's position and fitness value in the particle's individual extreme value P g In the g The individual position and fitness of the best fitness value in the global extreme value P Z middle;

[0161] (6) Determine the initial temperature T0 = f(P g ) / In5;

[0162] (7) Determine the current temperature of each particle P j The fitness value is calculated as follows:

[0163]

[0164] (8) From all P j Determine the global optimal replacement value P Z , and update the position and velocity of each particle according to the following two formulas:

[0165] X j,i (m+1)=X j,i (m)+v j,i (m+1)

[0166]

[0167]

[0168] Among them, m represents the current iteration number; j represents the number of particles; i represents the dimension of the particle.

[0169] (9) Calculate the particle fitness value and update P g and P Z , and then perform the cooling operation as follows:

[0170] T m+1 =λT m

[0171] (10) When the number of iterations reaches the maximum number of iterations, the search stops and the results are output. Otherwise, the search returns to step (3) to continue.

[0172] A method for optimizing the control system for welding medium and thick titanium alloy plates for new energy vehicles based on reliability is provided, such as Figure 2 As shown, it mainly includes the following steps:

[0173] S1: Initialize the position and velocity of the particles, where the number of individuals in the initial group N = L 4 The spatial dimension of each particle is 4, corresponding to the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I). The learning factor C1, learning factor C2, annealing constant inertia weight λ, and maximum number of iterations M are set. max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T ;

[0174] S2: The tensile strength of the weld joint, the weld width at the bottom, and the angular deformation are predicted through the RBF neural network based on the improved genetic algorithm in the performance prediction module.

[0175] In order to apply the performance prediction module to predict the tensile strength of welded joints, weld width at the bottom of the weld, and angular deformation, it is first necessary to train and learn the sample points. To improve the accuracy and stability of the prediction, the present invention uses a BRF neural network model based on an improved genetic algorithm to train and predict the samples. The genetic algorithm adopts an elite retention strategy, a strategy for dynamically adjusting the number of intersections based on population diversity, and an adaptive mutation probability. The specific steps for calculating the number of intersections are as follows:

[0176] ① Set the number of basic intersections n b , diversity threshold θ l and θ h , adjustment coefficients α and β;

[0177] ②Calculate the diversity of the population, the formula is:

[0178]

[0179] Where D(t) represents the diversity of the population when the number of iteration steps in the genetic algorithm is t; M represents the population size in the genetic algorithm; U represents the length of the chromosome; represents the kth gene of individual i.

[0180] ③Calculation of the number of intersections:

[0181]

[0182] S3: Based on the tensile strength of the weld joint, weld width at the bottom of the weld, and angular deformation obtained in step S2, the reliability is calculated using a non-probabilistic reliability model based on the convex ellipsoid model. This reliability is then compared with the reliability threshold to determine the rationality of the current set of process parameters. If the reliability threshold is met, proceed to step S4; otherwise, return to step S1.

[0183] In order to use the reliability assessment module to judge the rationality of the current process parameter set, it is necessary to first construct a non-probabilistic reliability model based on the ellipsoidal convex model and the limit state equation, and then combine the two to calculate the reliability. The specific steps are as follows:

[0184] ① Sample X j =[V j , W j , f j , I j ] T ,j=1,2,...,L 4 , one by one into the non-probabilistic reliability model of welding, and statistically analyze the samples that meet the conditions

[0185] Quantity, and record it as Q1;

[0186] ② The sample X obtained in step ① j =[V j , W j , f j , I j ] T , j=1,2,...,Q1, and substitute the three limit state equations of welding one by one: G1(X)=g1(X)-0.8·R b , G2(X)=g2(X)-d min , G3(X)=S max -g3(X), count the number of samples that make G1≥0, G2≥0 and G3≥0, and record it as Q2;

[0187] ③Calculate the reliability of the welding process β=Q1 / Q2.

[0188] S4: The process parameter optimization module is used to perform target optimization, with welding speed (V), wire feed speed (W), pulse frequency (f), and pulse current (I) as optimization variables, and tensile strength, weld bottom penetration, and weld angle deformation as optimization targets. Based on the three predicted values ​​obtained by the performance prediction module, the fitness function is defined as the weighted sum of these three predicted values. In the optimization process, to avoid falling into local solutions, the simulated annealing algorithm is combined with its ability to jump during the search process.

[0189] S5: Store the particle's position and fitness value in the particle's individual extreme value P g In the g The individual position and fitness of the best fitness value in the global extreme value P Z middle;

[0190] S6: Determine the initial temperature T0 = f(Pg) / In5;

[0191] S7: Determine the current temperature of each particle Pj The fitness value of

[0192] S8: From all P j Determine the global optimal replacement value P` Z , and update the position and velocity of each particle according to the following two formulas;

[0193] S9: Calculate the particle target value and update Pg and P Z , and then perform cooling operation;

[0194] S10: When the number of iterations reaches the maximum number of iterations, the search stops and the result is output, otherwise the search returns to step S3 to continue.

Claims

1. A reliability-based optimization control system for welding titanium alloy medium and thick plates for new energy vehicles, characterized by: It includes data acquisition and preprocessing module, performance prediction module, reliability evaluation module and process parameter optimization module.

2. The reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles according to claim 1 is characterized by: The data acquisition module obtains data samples through experiments, systematically changes the welding speed, wire feed speed, pulse frequency and pulse current, conducts multiple welding tests, records welding process parameters, and measures the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation under the corresponding welding process parameters based on an optical microscope, a three-dimensional scanner and an electronic tensile testing machine.

3. The reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles according to claim 1 is characterized by: The performance prediction module is used to train and learn the sample data obtained in the data acquisition module, and the trained performance prediction module is used to perform predictions; the performance prediction module includes data normalization processing, training set / test set division, creation of RBF neural network, training of RBF neural network, simulation testing, and performance evaluation; The RBF neural network is composed of an input layer, a hidden layer, and an output layer. The signal sources of the input layer include welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I). The output layer outputs tensile strength, weld width at the bottom of the weld, and angular deformation. The weights and initial values ​​of the Gaussian function parameters of the RBF neural network are optimized based on an improved genetic algorithm, using binary coding, an individual length of 10, an elite retention strategy, a strategy for dynamically adjusting the number of intersections according to population diversity, and an adaptive mutation probability.

4. The reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles according to claim 1 is characterized by: The reliability assessment module is composed of a welding non-probabilistic reliability model based on an ellipsoid convex model, a limit state equation, and a rationality judgment of reliability. The welding non-probabilistic reliability model based on an ellipsoid convex model takes welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as uncertain parameters, which are recorded as a vector X = [V, W, f, I] T , and calculate the mean vector μ corresponding to the above uncertain parameters x and covariance matrix ∑ x , according to the mean vector μ x and covariance matrix ∑ x , and after decomposition and transformation, the welding non-probabilistic reliability model based on the ellipsoid convex model is constructed as The limit state equation uses the RBF neural network prediction model based on the improved genetic algorithm in the performance prediction module as the function g(X), and obtains the tensile strength g1(X) of the weld joint, the weld width g2(X) at the bottom of the weld, and the angular deformation g3(X), and the tensile strength R b 80% of the minimum weld width d required for penetration min , the maximum welding angle deformation S allowed under service conditions max As the three safety allowable values ​​K1, K2, and K3, the three limit state equations of welding are obtained: G1(X)=g1(X)-0.8·R b ,G2(X)=g2(X)-d min ,G3(X)=S max -g3(X); The rationality judgment of the reliability is combined with the non-probabilistic reliability model based on the ellipsoid convex model and the limit state equation, and the reliability is defined as the ratio of the area where the limit state equation is greater than or equal to 0 to the volume of the entire ellipsoid area. If G1≥0, G2≥0 and G3≥0, the process parameters are reasonable.

5. The reliability-based optimization control system for welding medium and thick titanium alloy plates for new energy vehicles according to claim 1 is characterized by: The process parameter optimization module uses a particle swarm algorithm based on simulated annealing to perform target optimization. The algorithm parameters are set as follows: the number of individuals in the initialization group N, the learning factor C1, the learning factor C2, the annealing constant inertia weight λ, the maximum number of iterations M max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T ; Taking welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I) as optimization variables, the reliability of the current process parameter group is evaluated through the reliability evaluation module based on the tensile strength, weld width at the bottom of the weld, and angular deformation obtained by the performance prediction module; If the current process parameter group is reasonable, the tensile strength, weld width at the bottom of the weld, and angular deformation are used as optimization targets, and the fitness function is defined based on the weighted weight method, and the particle swarm algorithm is used for optimization; In the optimization process, in order to avoid falling into a local solution, the ability of the simulated annealing algorithm to jump during the search process is combined.

6. A control method for a reliability-based optimization control system for titanium alloy medium and thick plate welding for new energy vehicles according to any one of claims 1 to 5, comprising the following steps: S1: Initialize the position and velocity of the particle, where Initialize the number of individuals in the population N = L 4 The spatial dimension of each particle is 4, corresponding to the four uncertain parameters of welding speed (V), wire feeding speed (W), pulse frequency (f), and pulse current (I). The learning factor C1, learning factor C2, annealing constant inertia weight λ, and maximum number of iterations M are set. max , the spatial dimension is 4, the number of interval divisions is L, and the reliability threshold is β T ; S2: The tensile strength of the weld joint, the weld width at the bottom, and the angular deformation are predicted using the RBF neural network based on the improved genetic algorithm in the performance prediction module; In order to apply the performance prediction module to predict the tensile strength of welded joints, the weld width at the bottom of the weld, and the angular deformation, it is first necessary to train and learn the sample points. To improve the accuracy and stability of the prediction, the present invention uses a BRF neural network model based on an improved genetic algorithm to train and predict the samples. In addition, the genetic algorithm adopts an elite retention strategy, a strategy for dynamically adjusting the number of intersections based on population diversity, and an adaptive mutation probability. The specific steps for calculating the number of intersections in the genetic algorithm are as follows: ① Set the number of basic intersections n b , diversity threshold θ l and θ h , adjustment coefficients α and β; ②Calculate the diversity of the population, the formula is: Where D(t) represents the diversity of the population when the number of iteration steps in the genetic algorithm is t; M represents the population size in the genetic algorithm; U represents the length of the chromosome; represents the kth gene of individual i; ③Calculation of the number of intersections: S3: Based on the tensile strength of the weld joint, the weld width at the bottom of the weld, and the angular deformation obtained in step S2, the values ​​are substituted into the non-probabilistic reliability model based on the convex ellipsoid model to calculate the reliability, and the reliability is compared with the reliability threshold to determine the rationality of the current process parameter set. If the reliability is satisfied, step S4 is performed; otherwise, the process returns to step S1. In order to use the reliability assessment module to judge the rationality of the current process parameter set, it is necessary to first construct a non-probabilistic reliability model based on the ellipsoidal convex model and the limit state equation, and then combine the two to calculate the reliability. The specific steps are as follows: ① Sample X j =[V j , W j , f j , I j ] T ,j=1,2,...,L 4 , substitute the non-probabilistic reliability model of welding one by one, count the number of samples that meet the conditions, and record it as Q1; ② The sample X obtained in step ① j =[V j , W j , f j , I j ] T , j=1,2,...,Q1, and substitute the three limit state equations of welding one by one: G1(X)=g1(X)-0.8·R b ,G2(X)=g2(X)-d min ,G3(X)=S max -g3(X), count the number of samples that make G1≥0, G2≥0 and G3≥0, and record it as Q2; ③Calculate the reliability of the welding process β=Q1 / Q2; S4: The process parameter optimization module is used to perform target optimization, with welding speed (V), wire feed speed (W), pulse frequency (f), and pulse current (I) as optimization variables, and tensile strength, weld width at the bottom of the weld, and angular deformation as optimization targets. Based on the three predicted values ​​obtained by the performance prediction module, the fitness function is defined as the weighted sum of these three predicted values. In the optimization process, to avoid falling into local solutions, the simulated annealing algorithm is combined with its ability to jump during the search process. S5: Store the particle's position and fitness value in the particle's individual extreme value P g In the g The individual position and fitness of the best fitness value in the global extreme value P Z middle; S6: Determine the initial temperature T0 = f(Pg) / In5; S7: Determine the current temperature of each particle P j The fitness value of S8: From all P j Determine the global optimal replacement value P Z , and update the position and velocity of each particle according to the following two formulas; S9: Calculate the particle target value and update P g and P Z , and then perform cooling operation; S10: When the number of iterations reaches the maximum number of iterations, the search stops and the result is output; otherwise, the search returns to step S3 to continue.

Images