A method and system for predicting quality of abrasive jet cold cutting
Through radial basis neural network and genetic algorithm optimization, the problem of poor model complexity and generalization ability in abrasive jet cold cutting quality prediction is solved, and efficient and accurate cutting quality prediction is achieved, improving prediction accuracy and stability.
Patent Information
- Application Number
- CN202410209518.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-26
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-02-26
AI Technical Summary
The existing abrasive jet cold cutting quality prediction method has the problems of complex model establishment, difficult to determine parameters, and time-consuming, based on physical models; as well as poor generalization ability, low interpretability and poor stability of the model, and data-driven methods .
The radial-based neural network model is used in combination with genetic algorithms, and the cutting experimental data is collected and normalized to optimize neural network parameters to improve prediction accuracy and model stability.
Effectively fit nonlinear and high-dimensional data relationships, improve the prediction accuracy of cutting quality, avoid local optimal solutions, enhance the generalization ability and stability of the model, eliminate the influence of data dimensions, and improve the reliability and robustness of the model.
Smart Images

Figure CN118246527B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of abrasive jet prediction, and in particular relates to an abrasive jet cold cutting quality prediction method and system. Background Art
[0002] Abrasive jet cold cutting is a technology that uses a liquid-solid two-phase jet mixed with abrasive particles in a high-pressure water jet to cut various materials. Abrasive jet cold cutting has the advantages of low cutting temperature, high cutting accuracy, wide cutting range, and low environmental pollution. It has been widely used in industrial fields, marine engineering, aerospace and other fields. However, the quality of abrasive jet cold cutting is affected by many factors, such as cutting process parameters, abrasive particle characteristics, and the properties of the material being cut.
[0003] At present, there are two main methods for predicting the quality of abrasive jet cold cutting: methods based on physical models and methods based on data-driven. The method based on physical models is to establish a mathematical model based on the physical mechanism of abrasive jet cold cutting, and obtain the predicted value of cutting quality by solving the equations of the model. The advantage of this method is that it can reflect the essential laws of abrasive jet cold cutting, but there are also some disadvantages, such as the complex process of model establishment, the difficulty in determining the parameters of the model, and the long time it takes to solve the model. The method based on data-driven is to establish a data model based on the experimental data of abrasive jet cold cutting, using machine learning, artificial intelligence and other technologies, and obtain the predicted value of cutting quality by training the parameters of the model. The advantage of this method is that it can quickly and accurately predict the cutting quality, but there are also some disadvantages, such as poor generalization ability of the model, low interpretability of the model, and poor stability of the model. Therefore, it is urgent to develop an abrasive jet cold cutting prediction method that combines physical mechanism and data-driven. Summary of the invention
[0004] Based on the above technical problems, the present invention provides an abrasive jet cold cutting quality prediction method and system, aiming to achieve accurate prediction and optimization of cutting quality by collecting cutting experimental data based on a neural network model and a genetic algorithm.
[0005] The present invention provides a method for predicting the quality of abrasive jet cold cutting, the method comprising:
[0006] Step S1: Collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators;
[0007] Step S2: normalizing the experimental data and dividing them into a training set and a test set;
[0008] Step S3: constructing a radial basis function neural network model; the radial basis function neural network model has an input layer of cutting process parameters, an output layer of cutting process quality indicators, an activation function of a Gaussian function, and a loss function of a mean square error;
[0009] Step S4: using a genetic algorithm to optimize the parameters of the radial basis neural network model; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias;
[0010] Step S5: Use the test set to evaluate the prediction performance of the radial basis neural network model, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network model.
[0011] Optionally, the collecting cutting experimental data specifically includes:
[0012] The cross-sectional area A of the abrasive jet is expressed as:
[0013]
[0014] Where, d is the nozzle diameter; α is the semi-apex angle of the jet; L is the target distance;
[0015] The velocity V of the abrasive jet is expressed as:
[0016]
[0017] Where, Q is the abrasive jet flow rate; V is the jet velocity;
[0018] The kinetic energy of the jet E k It is expressed as:
[0019]
[0020] In the formula, E k is the kinetic energy of the abrasive jet; ρ is the density of the abrasive jet;
[0021] The pressure P of the jet is expressed as:
[0022]
[0023] Where, P is the abrasive jet pressure;
[0024] The impact force F of the jet is expressed as:
[0025]
[0026] Where F is the impact force of the abrasive jet; θ is the angle between the axis and the surface of the material being cut;
[0027] The loss level LR is expressed as:
[0028]
[0029] Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; m n is the mass of the nozzle worn during the cutting process; ρ a is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n ;
[0030] Energy consumption EC is expressed as:
[0031]
[0032] Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water;
[0033] Cutting accuracy CP is expressed as:
[0034]
[0035] Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a It is the surface roughness of the cut after cutting.
[0036] Optionally, the normalizing the experimental data specifically includes:
[0037]
[0038] Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix;
[0039] Data normalization is expressed as:
[0040]
[0041] In the formula, x ij andij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x ij ′ and y ij ′ are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j ) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
[0042] Optionally, the constructing of a radial basis neural network model specifically includes:
[0043] The input layer of the radial basis neural network model is the cutting process parameters; the output layer is the cutting process quality index; the number of neurons in the hidden layer is M;
[0044] The activation function of the hidden layer is expressed as:
[0045]
[0046] Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm;
[0047] The output function of the output layer is expressed as:
[0048]
[0049] In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron;
[0050] The loss function is expressed as:
[0051]
[0052] In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, the number of partial experimental data.
[0053] Optionally, the use of a genetic algorithm to optimize radial basis neural network model parameters specifically includes:
[0054] Initialize the parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is P m , the number of elite individuals is E, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1];
[0055] Generate the initial population, randomly generate S individuals, each of which has M hidden neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as:
[0056] X s =[M,C,∑,W,B],s=1,2,…,S
[0057] Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1;
[0058] Calculate the fitness, for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as:
[0059]
[0060] In the formula, ι is a very small positive number, which is used to avoid the situation where the denominator is zero;
[0061] Selection operation: Use the roulette method to select S individuals to enter the next generation, and retain E elite individuals, that is, the E individuals with the highest fitness, which do not participate in crossover and mutation operations and are directly copied to the next generation;
[0062] The crossover operation, using adaptive crossover probability, is expressed as:
[0063]
[0064] Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation;
[0065] For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including:
[0066] The number of hidden neurons M is crossed, a crossover point p is randomly selected, and then the first p bits of binary codes of M of the two individuals are exchanged to generate two new M values, which are expressed as:
[0067] M′1=M1[1:p]+M2[p+1:M max ]
[0068] M′2=M2[1:p]+M1[p+1:M max ]
[0069] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1;
[0070] The intersection of the center vector C randomly selects an intersection point a, and then exchanges the first a elements of C of the two individuals to generate two new C vectors, expressed as:
[0071]
[0072]
[0073] Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the Bth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M mm =min(M′1,M′2);
[0074] Crossover of width parameter ∑ randomly selects a crossover point r, and then exchanges the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as:
[0075]
[0076]
[0077] Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min ;
[0078] Connect the cross of weight W, randomly select a crossover point f, and then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as:
[0079]
[0080]
[0081] Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min ;
[0082] Biased crossover of B randomly selects a crossover point h, and then exchanges the first h elements of B of the two individuals to generate two new B vectors, expressed as:
[0083] B′1=[b 11 , b 12 , …, b 1h , b 2h+1 , b 2h+2 , …, b 23 ]
[0084] B′2=[b 21 , b 22 , …, b 2h , b 1h+1 , b 1h+2 , …, b 13 ]
[0085] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ is the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3;
[0086] The mutation operation uses adaptive mutation probability, expressed as:
[0087]
[0088] Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation;
[0089] For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including:
[0090] The variation of the number of hidden layer neurons M randomly selects a crossover point o, and then flips the binary code of the oth bit to generate a new M value, which is expressed as:
[0091] M′=M[1:o-1]+M[o]+M[o+1:M max ]
[0092] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max ;
[0093] The mutation of the center vector C randomly selects an intersection point u, and then adds a random number that follows a normal distribution to the u-th element to generate a new C vector, which is expressed as:
[0094] C′=[c1,c2,…,c u +δ,…,c M′ ]
[0095] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤o≤M′;
[0096] The variation of the width parameter ∑ randomly selects a crossover point r, and then adds a random number that follows a normal distribution to the rth element to generate a new ∑ vector, which is expressed as:
[0097] ∑′=[σ1,σ2,…,σ r +δ,…,σ M′ ]
[0098] Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤r≤M′;
[0099] To mutate the connection weight W, randomly select a mutation point v, and then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as:
[0100] W′=[w 11 , w 12 ,…,w v +δ,…,w 3M′]
[0101] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤v≤3M′;
[0102] The variation of bias B randomly selects a variation point z, and then adds a random number that follows a normal distribution to the zth element to generate a new B vector, which is expressed as:
[0103] B′=[b1,b2,…,b z +δ,…,b3]
[0104] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3;
[0105] Termination condition: if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, stop evolution, output the parameters and fitness of the optimal individual, and the corresponding radial basis neural network model; otherwise, let g = g + 1, return to calculate the fitness, and continue evolution.
[0106] The present invention also provides an abrasive jet cold cutting quality prediction system, the system comprising:
[0107] An experimental data collection module is used to collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators;
[0108] A normalization processing and division module is used to normalize the experimental data and divide it into a training set and a test set;
[0109] A radial basis network construction module is used to construct a radial basis neural network model; the input layer of the radial basis neural network model is the cutting process parameters, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error;
[0110] A network parameter optimization module, used to optimize the parameters of the radial basis neural network model using a genetic algorithm; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias;
[0111] The model performance evaluation module is used to evaluate the prediction performance of the radial basis neural network model using a test set, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network model.
[0112] Optionally, the experimental data collection module specifically includes:
[0113] The cross-sectional area A of the abrasive jet is expressed as:
[0114]
[0115] Where, d is the nozzle diameter; α is the semi-apex angle of the jet; L is the target distance;
[0116] The velocity V of the abrasive jet is expressed as:
[0117]
[0118] Where, Q is the abrasive jet flow rate; V is the jet velocity;
[0119] The kinetic energy of the jet E k It is expressed as:
[0120]
[0121] In the formula, E k is the kinetic energy of the abrasive jet; ρ is the density of the abrasive jet;
[0122] The pressure P of the jet is expressed as:
[0123]
[0124] Where, P is the abrasive jet pressure;
[0125] The impact force F of the jet is expressed as:
[0126]
[0127] Where F is the impact force of the abrasive jet; θ is the angle between the axis and the surface of the material being cut;
[0128] The loss level LR is expressed as:
[0129]
[0130] Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; m n is the mass of the nozzle worn during the cutting process; ρa is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n ;
[0131] Energy consumption EC is expressed as:
[0132]
[0133] Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water;
[0134] Cutting accuracy CP is expressed as:
[0135]
[0136] Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a It is the surface roughness of the cut after cutting.
[0137] Optionally, the normalization processing division module specifically includes:
[0138]
[0139] Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix;
[0140] Data normalization is expressed as:
[0141]
[0142] In the formula, x ij and ij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x′ ij and y′ ij are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
[0143] Optionally, the radial basis network construction module specifically includes:
[0144] The input layer of the radial basis neural network model is the cutting process parameters; the output layer is the cutting process quality index; the number of neurons in the hidden layer is M;
[0145] The activation function of the hidden layer is expressed as:
[0146]
[0147] Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm;
[0148] The output function of the output layer is expressed as:
[0149]
[0150] In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron;
[0151] The loss function is expressed as:
[0152]
[0153] In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, the number of partial experimental data.
[0154] Optionally, the network parameter optimization module specifically includes:
[0155] Parameter initialization submodule, used to initialize parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is Pm , the number of elite individuals is E, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1];
[0156] The initial population generation submodule is used to generate the initial population and randomly generate S individuals. Each individual consists of M hidden layer neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as:
[0157] X s =[M,C,∑,W,B],s=1,2,…,S
[0158] Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1;
[0159] The fitness calculation submodule is used to calculate the fitness for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as:
[0160]
[0161] In the formula, ι is a very small positive number, which is used to avoid the situation where the denominator is zero;
[0162] The selection submodule is used to perform selection operations. It uses the roulette method to select S individuals to enter the next generation, while retaining E elite individuals, that is, the E individuals with the highest fitness, which do not participate in crossover and mutation operations and are directly copied to the next generation;
[0163] The crossover submodule is used to perform a crossover operation using an adaptive crossover probability, expressed as:
[0164]
[0165] Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation;
[0166] For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including:
[0167] The number of hidden neurons M is crossed, a crossover point p is randomly selected, and then the first p bits of binary codes of M of the two individuals are exchanged to generate two new M values, which are expressed as:
[0168] M′1=M1[1:p]+M2[p+1:M max ]
[0169] M′2=M2[1:p]+M1[p+1:M max ]
[0170] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1;
[0171] The intersection of the center vector C randomly selects an intersection point a, and then exchanges the first a elements of C of the two individuals to generate two new C vectors, expressed as:
[0172]
[0173]
[0174] Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the Bth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M min =min(M′1,M′2);
[0175] Crossover of width parameter ∑ randomly selects a crossover point r, and then exchanges the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as:
[0176]
[0177]
[0178] Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min ;
[0179] Connect the cross of weight W, randomly select a crossover point f, and then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as:
[0180]
[0181]
[0182] Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min ;
[0183] Biased crossover of B randomly selects a crossover point h, and then exchanges the first h elements of B of the two individuals to generate two new B vectors, expressed as:
[0184] B′1=[b 11 , b 12 , …, b 1h , b 2h+1 , b 2h+2 , …, b 23 ]
[0185] B′2=[b 21 , b 22 , …, b 2h , b 1h+1 , b 1h+2 , …, b 13 ]
[0186] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ is the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3;
[0187] The mutation submodule is used to perform mutation operations and uses an adaptive mutation probability, expressed as:
[0188]
[0189] Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation;
[0190] For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including:
[0191] The variation of the number of hidden layer neurons M randomly selects a crossover point o, and then flips the binary code of the oth bit to generate a new M value, which is expressed as:
[0192] M′=M[1:o-1]+M[o]+M[o+1:M max ]
[0193] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max ;
[0194] The mutation of the center vector C randomly selects an intersection point u, and then adds a random number that follows a normal distribution to the u-th element to generate a new C vector, which is expressed as:
[0195] C′=[c1,c2,…,c u +δ,…,c M′ ]
[0196] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤o≤M′;
[0197] The variation of the width parameter ∑ randomly selects a crossover point r, and then adds a random number that follows a normal distribution to the rth element to generate a new ∑ vector, which is expressed as:
[0198] ∑′=[σ1,σ2,…,σ r +δ,…,σ M′ ]
[0199] Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤r≤M′;
[0200] To mutate the connection weight W, randomly select a mutation point v, and then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as:
[0201] W′=[w 11 , w 12 ,…,w v +δ,…,w 3M′ ]
[0202] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤v≤3M′;
[0203] The variation of bias B randomly selects a variation point z, and then adds a random number that follows a normal distribution to the zth element to generate a new B vector, which is expressed as:
[0204] B′=[b1,b2,…,b z +δ,…,b3]
[0205] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3;
[0206] Conditional termination submodule, used for termination conditions, if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, stop evolution, output the parameters and fitness of the optimal individual, and the corresponding radial basis neural network model; otherwise, let g = g + 1, return to calculate the fitness, and continue evolution.
[0207] Compared with the prior art, the present invention has the following beneficial effects:
[0208] The present invention adopts radial basis function (RBF) neural network as data model, which can effectively fit nonlinear, high-dimensional and complex data relationships and improve the prediction accuracy of cutting quality; adopts genetic algorithm (GA) as optimization algorithm, which can effectively avoid local optimal solution, find global optimal solution and improve the generalization ability and stability of the model; adopts normalization processing and error range judgment, which can effectively eliminate the dimension and scale influence of data and improve the reliability and robustness of the model; BRIEF DESCRIPTION OF THE DRAWINGS
[0209] Figure 1 A flow chart of a method for predicting abrasive jet cold cutting quality according to the present invention;
[0210] Figure 2 This is a structural diagram of an abrasive jet cold cutting quality prediction system of the present invention. DETAILED DESCRIPTION
[0211] The present invention is further described below in conjunction with specific implementation cases and drawings, but the present invention is not limited to these embodiments.
[0212] Example 1
[0213] like Figure 1 As shown, the present invention discloses a method for predicting the quality of abrasive jet cold cutting, the method comprising:
[0214] Step S1: Collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators.
[0215] Step S2: Normalize the experimental data and divide them into a training set and a test set.
[0216] Step S3: construct a radial basis function neural network model; the input layer of the radial basis function neural network model is the cutting process parameters, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error.
[0217] Step S4: Use genetic algorithm to optimize radial basis neural network parameters; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias.
[0218] Step S5: Use the test set to evaluate the prediction performance of the radial basis neural network, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network.
[0219] The following is a detailed discussion of each step:
[0220] Step S1: Collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators.
[0221] Step S1 specifically includes:
[0222] In the abrasive jet structure, the jet is conical, with its apex at the nozzle outlet, the semi-apex angle of the jet is α, and the angle between its axis and the surface of the material being cut is θ.
[0223] The cross-sectional area A of the abrasive jet is expressed as:
[0224]
[0225] Where d is the nozzle diameter; this formula indicates that the cross-sectional area of the jet increases with the increase of the target distance, but the rate of increase decreases with the increase of the target distance; when the target distance is zero, the cross-sectional area of the jet is equal to the cross-sectional area of the nozzle, that is, A = πd 2 / 4; when the target distance tends to infinity, the cross-sectional area of the jet tends to infinity, that is, A→∞.
[0226] The velocity V of the abrasive jet is expressed as:
[0227]
[0228] Where Q is the abrasive jet flow rate; V is the jet velocity.
[0229] The kinetic energy of the jet E k It is expressed as:
[0230]
[0231] In the formula, E k is the kinetic energy of the abrasive jet; m is the mass of the abrasive jet; ρ is the density of the abrasive jet.
[0232] The pressure P of the jet is expressed as:
[0233]
[0234] Where P is the abrasive jet pressure.
[0235] The impact force F of the jet is expressed as:
[0236]
[0237] Where F is the abrasive jet impact force.
[0238] The loss level LR is expressed as:
[0239]
[0240] Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; m n is the mass of the nozzle worn during the cutting process; ρ a is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n .
[0241] Energy consumption EC is expressed as:
[0242]
[0243] Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water.
[0244] Cutting accuracy CP is expressed as:
[0245]
[0246] Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a is the surface roughness of the cut after cutting; F is the impact force of the abrasive jet; α is the semi-vertex angle of the jet; D is the abrasive particle size; d is the nozzle diameter; V is the jet velocity; and L is the target distance.
[0247] In this embodiment, the cutting process parameters include jet pressure P, abrasive flow rate Q, abrasive particle size D, nozzle diameter d, jet velocity V, target distance L, abrasive jet impact force F, jet semi-apex angle α, cutting angle θ, etc.; the cutting processing quality indicators include loss degree, energy consumption and cutting accuracy.
[0248] Step S2: Normalize the experimental data and divide them into a training set and a test set.
[0249] Step S2 specifically includes:
[0250]
[0251] Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix.
[0252] Data normalization is expressed as:
[0253]
[0254] In the formula, x ij and ij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x′ ij and y′ ij are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j ) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
[0255] The purpose of dividing the data into training set and test set is to allow the radial basis function (RBF) neural network and genetic (GA) algorithm to learn and optimize on a part of the data and evaluate and verify on another part of the data. Generally speaking, the training set occupies the majority of the data, and the test set occupies a small part of the data; the specific proportion is determined based on the amount of data and requirements.
[0256] Step S3: construct a radial basis function neural network model; the input layer of the radial basis function neural network model is the cutting process parameters, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error.
[0257] Step S3 specifically includes:
[0258] A radial basis function neural network model (RBF) is constructed. The input layer of the radial basis function neural network model is the cutting process parameters, which has nine neurons, including jet pressure P, abrasive flow rate Q, abrasive particle size D, nozzle diameter d, jet velocity V, target distance L, abrasive jet impact force F, jet semi-vertex angle a, cutting angle θ, etc. The neurons of the input layer are set according to the actual situation; the output layer is the cutting process quality index, which has three neurons, corresponding to three cutting process quality indexes respectively; the number of neurons in the hidden layer is M, which can be optimized and determined by the GA algorithm.
[0259] The activation function of the hidden layer is a Gaussian function, expressed as:
[0260]
[0261] Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm.
[0262] The output function of the output layer is a linear function, expressed as:
[0263]
[0264] In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron.
[0265] The loss function is the mean square error, expressed as:
[0266]
[0267] In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, part of the experimental data; the goal is to minimize the loss function Loss by optimizing the parameters of the RBF neural network.
[0268] Step S4: using an improved genetic algorithm to optimize the parameters of the radial basis neural network; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias.
[0269] Step S4 specifically includes:
[0270] Initialize the parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is P m , the number of elite individuals is E, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1].
[0271] Generate the initial population, randomly generate S individuals, each of which has M hidden neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as:
[0272] X s =[M,C,∑,W,B],s=1,2,…,S
[0273] Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1.
[0274] Calculate the fitness, for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as:
[0275]
[0276] In the formula, ι is a very small positive number, which is used to avoid the situation where the denominator is zero; the larger the fitness function, the higher the quality of the individual.
[0277] The selection operation uses the roulette method to select S individuals to enter the next generation, while retaining E elite individuals, that is, the E individuals with the highest fitness, which do not participate in the crossover and mutation operations and are directly copied to the next generation.
[0278] The crossover operation uses an adaptive crossover probability, expressed as:
[0279]
[0280] Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation.
[0281] The adaptive crossover probability decreases with the increase of evolutionary generations to maintain the diversity of the population. For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including:
[0282] I. Crossover of the number of hidden neurons M: Randomly select a crossover point p, then exchange the first p bits of binary codes of M of the two individuals to generate two new M values, expressed as:
[0283] M′1=M1[1:p]+M2[p+1:M max ]
[0284] M′2=M2[1:p]+M1[p+1:M max ]
[0285] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1.
[0286] II. Crossover of the center vector C: Randomly select a crossover point a, then exchange the first a elements of C of the two individuals to generate two new C vectors, expressed as:
[0287]
[0288]
[0289] Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the βth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M min =min(M′1, M′2).
[0290] III. Crossover of width parameter ∑: Randomly select a crossover point r, then exchange the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as:
[0291]
[0292]
[0293] Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min .
[0294] IV. Crossover of connection weight W: Randomly select a crossover point f, then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as:
[0295]
[0296]
[0297] Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min .
[0298] V. Crossover of biased B: Randomly select a crossover point h, then exchange the first h elements of B of the two individuals to generate two new B vectors, expressed as:
[0299] B′1=[b 11 , b 12 , …, b 1h , b 2h+1 , b 2h+2 , …, b 23 ]
[0300] B′2=[b 21 , b 22 , …, b 2h , b 1h+1 , b 1h+2 , …, b 13 ]
[0301] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ represents the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3.
[0302] The mutation operation uses adaptive mutation probability, expressed as:
[0303]
[0304] Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation.
[0305] The probability of adaptive mutation decreases with the increase of evolutionary generations to maintain the diversity of the population. For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including:
[0306] (1) Variation of the number of hidden layer neurons M: Randomly select a crossover point o, then flip the binary code of the oth bit to generate a new M value, expressed as:
[0307] M′=M[1:o-1]+M[o]+M[o+1:M max ]
[0308] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max .
[0309] (2) Mutation of the center vector C: Randomly select an intersection point u, then add a random number that follows a normal distribution to the u-th element to generate a new C vector, expressed as:
[0310] C′=[c1,c2,…,c u +δ,…,c M′ ]
[0311] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c It is a smaller standard deviation used to control the amplitude of variation; it satisfies 1≤o≤M′.
[0312] (3) Variation of the width parameter ∑: Randomly select a crossover point r, then add a random number that follows a normal distribution to the rth element to generate a new ∑ vector, expressed as:
[0313] ∑′=[σ1,σ2,…,σ r +δ,…,σ M′ ]
[0314] Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤r≤M′.
[0315] (4) Mutation of connection weight W: Randomly select a mutation point v, then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as:
[0316] W′=[w 11 , w 12 ,…,w v +δ,…,w 3M′ ]
[0317] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤v≤3M′.
[0318] (5) Variation of bias B: Randomly select a mutation point z, then add a random number that follows a normal distribution to the zth element to generate a new B vector, expressed as:
[0319] B′=[b1,b2,…,b z +δ,…,b3]
[0320] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3.
[0321] Termination condition: if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, or the program is terminated early, then the evolution is stopped, and the parameters and fitness of the optimal individual, as well as the corresponding RBF neural network model are output; otherwise, let g = g + 1, return to calculate the fitness, and continue evolution.
[0322] Step S5: Use the test set to evaluate the prediction performance of the radial basis neural network, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network.
[0323] Step S5 specifically includes:
[0324] Use the data of the test set to substitute the parameters and fitness of the optimal individual and the corresponding RBF neural network model to obtain the predicted value; denormalize the predicted value to obtain the original predicted value, including:
[0325]
[0326] In the formula, min(y j ) and max(y j ) are the minimum and maximum values of the cutting quality index of the jth column respectively; is the predicted value; is the original predicted value.
[0327] Calculate predicted values and the actual value y J The mean absolute error, root mean square error and mean absolute percentage error between (x) are expressed as:
[0328]
[0329]
[0330]
[0331] In the formula, MAE J is the mean absolute error of the Jth cutting quality index; RMSE J is the root mean square error of the Jth cutting quality index; MAPE J is the mean absolute percentage error of the Jth cutting quality index; is the test set sample; and JA (x) represent the predicted value and actual value of the Jth cutting quality index of the Ath sample, respectively.
[0332] Use the weighted comprehensive evaluation method to determine whether the error meets the requirements, including:
[0333] First, a threshold is set for each error type (mean absolute error, root mean square error, and mean absolute percentage error), which represents an "acceptable" error; then, for each error type of each indicator, its normalized error relative to the corresponding threshold is calculated, expressed as:
[0334]
[0335] In the formula, E J,et is the normalized error of the et-th error type of the J-th cutting quality index; error J,et is the actual error of the et-th error type of the J-th cutting quality index, error J,et Including MAE J 、RMSE J and MAPE J ;threshold et is the threshold of the et-th error type.
[0336] Then the weighted error is calculated for each indicator, expressed as:
[0337]
[0338] In the formula, Γ J is the weighted error of the Jth cutting quality index; is the weight of the et-th error type.
[0339] Finally, analyze the error Γ of each indicator J If the size of Γ J The expected accuracy requirement is achieved, indicating that the RBF neural network has good prediction performance and is used for predictive control of cutting quality. The RBF neural network model is output; if Γ J If it is too large, it means that the generalization ability of the RBF neural network is insufficient, and it is necessary to adjust the parameters or structure of the network, or increase the amount of data in the training set and retrain the RBF neural network.
[0340] In this embodiment, an abrasive jet cold cutting robot is subsequently used for cutting. The cutting robot is an intelligent cutting device that combines robot technology and cutting technology. It can automatically complete cutting processes of various shapes and sizes according to different cutting requirements, improve cutting efficiency and quality, and reduce cutting costs and human resources. The cutting process parameters are sent to the control system of the cutting robot. The control system controls the movement of the cutting robot and the work of the cutting equipment according to the parameter instructions to complete the cutting process. During the cutting process, the cutting state is monitored in real time, and the cutting process parameters are input into the RBF neural network to obtain a real-time prediction value, which is compared with the expected target value to calculate the real-time prediction error. If the prediction error is within the allowable error range, it means that the cutting quality meets the requirements and continues cutting; if the prediction error exceeds the error range, it means that the cutting quality is unqualified, and the cutting process parameters need to be adjusted and sent to the control system of the cutting robot again to adjust the movement of the cutting robot and the work of the cutting equipment until the prediction error is within the error range.
[0341] Example 2
[0342] like Figure 2 As shown, the present invention discloses an abrasive jet cold cutting quality prediction system, the system comprising:
[0343] The experimental data collection module 10 is used to collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators.
[0344] The normalization processing and division module 20 is used to perform normalization processing on the experimental data and divide it into a training set and a test set.
[0345] The radial basis network construction module 30 is used to construct a radial basis neural network model; the input layer of the radial basis neural network model is the cutting process parameters, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error.
[0346] The network parameter optimization module 40 is used to optimize the radial basis neural network model parameters using a genetic algorithm; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias.
[0347] The model performance evaluation module 50 is used to evaluate the prediction performance of the radial basis neural network model using a test set, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network model.
[0348] As an optional implementation, the experimental data collection module 10 of the present invention specifically includes:
[0349] The cross-sectional area A of the abrasive jet is expressed as:
[0350]
[0351] Where d is the nozzle diameter; α is the semi-apex angle of the jet; and L is the target distance.
[0352] The velocity V of the abrasive jet is expressed as:
[0353]
[0354] Where Q is the abrasive jet flow rate; V is the jet velocity.
[0355] The kinetic energy of the jet E k It is expressed as:
[0356]
[0357] In the formula, E k is the kinetic energy of the abrasive jet; ρ is the density of the abrasive jet.
[0358] The pressure P of the jet is expressed as:
[0359]
[0360] Where P is the abrasive jet pressure.
[0361] The impact force F of the jet is expressed as:
[0362]
[0363] Where F is the impact force of the abrasive jet; θ is the angle between the axis and the surface of the material being cut.
[0364] The loss level LR is expressed as:
[0365]
[0366] Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; mn is the mass of the nozzle worn during the cutting process; ρ a is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n .
[0367] Energy consumption EC is expressed as:
[0368]
[0369] Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water.
[0370] Cutting accuracy CP is expressed as:
[0371]
[0372] Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a It is the surface roughness of the cut after cutting.
[0373] As an optional implementation mode, the normalization processing division module 20 of the present invention specifically includes:
[0374]
[0375] Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix.
[0376] Data normalization is expressed as:
[0377]
[0378] In the formula, xij and ij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x′ ij and ij ′ are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j ) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
[0379] As an optional implementation, the radial basis network construction module 30 of the present invention specifically includes:
[0380] The input layer of the radial basis neural network model is the cutting process parameters; the output layer is the cutting process quality index; the number of neurons in the hidden layer is M.
[0381] The activation function of the hidden layer is expressed as:
[0382]
[0383] Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm.
[0384] The output function of the output layer is expressed as:
[0385]
[0386] In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron.
[0387] The loss function is expressed as:
[0388]
[0389] In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, the number of partial experimental data.
[0390] As an optional implementation mode, the network parameter optimization module 40 of the present invention specifically includes:
[0391] Parameter initialization submodule, used to initialize parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is P m , the number of elite individuals is B, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1].
[0392] The initial population generation submodule is used to generate the initial population and randomly generate S individuals. Each individual consists of M hidden layer neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as:
[0393] X s =[M,C,∑,W,B],s=1,2,…,S
[0394] Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1.
[0395] The fitness calculation submodule is used to calculate the fitness for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as:
[0396]
[0397] In the formula, ι is a very small positive number, which is used to avoid the situation where the denominator is zero.
[0398] The selection submodule is used to perform selection operations. It uses the roulette method to select S individuals to enter the next generation, while retaining E elite individuals, that is, the E individuals with the highest fitness, which do not participate in crossover and mutation operations and are directly copied to the next generation.
[0399] The crossover submodule is used to perform a crossover operation using an adaptive crossover probability, expressed as:
[0400]
[0401] Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation.
[0402] For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including:
[0403] The number of hidden neurons M is crossed, a crossover point p is randomly selected, and then the first p bits of binary codes of M of the two individuals are exchanged to generate two new M values, which are expressed as:
[0404] M′1=M1[1:p]+M2[p+1:M max ]
[0405] M′2=M2[1:p]+M1[p+1:M max ]
[0406] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1.
[0407] The intersection of the center vector C randomly selects an intersection point a, and then exchanges the first a elements of C of the two individuals to generate two new C vectors, expressed as:
[0408]
[0409]
[0410] Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the Bth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M min =min(M′1, M′2).
[0411] Crossover of width parameter ∑ randomly selects a crossover point r, and then exchanges the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as:
[0412]
[0413]
[0414] Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min .
[0415] Connect the cross of weight W, randomly select a crossover point f, and then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as:
[0416]
[0417]
[0418] Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min .
[0419] Biased crossover of B randomly selects a crossover point h, and then exchanges the first h elements of B of the two individuals to generate two new B vectors, expressed as:
[0420] B′1=[b 11 , b 12 , …, b 1h , b 2h+1 , b 2h+2 , …, b 23 ]
[0421] B′2=[b 21 , b 22 , …, b 2h , b 1h+1 , b 1h+2 , …, b 13 ]
[0422] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ represents the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3.
[0423] The mutation submodule is used to perform mutation operations and uses an adaptive mutation probability, expressed as:
[0424]
[0425] Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation.
[0426] For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including:
[0427] The variation of the number of hidden layer neurons M randomly selects a crossover point o, and then flips the binary code of the oth bit to generate a new M value, which is expressed as:
[0428] M′=M[1:o-1]+M[o]+M[o+1:M max ]
[0429] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max .
[0430] The mutation of the center vector C randomly selects an intersection point u, and then adds a random number that follows a normal distribution to the u-th element to generate a new C vector, which is expressed as:
[0431] C′=[c1,c2,…,c u +δ,…,c M′ ]
[0432] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c It is a smaller standard deviation used to control the amplitude of variation; it satisfies 1≤o≤M′.
[0433] The variation of the width parameter ∑ randomly selects a crossover point r, and then adds a random number that follows a normal distribution to the rth element to generate a new ∑ vector, which is expressed as:
[0434] ∑′=[σ1,σ2,…,σ r +δ,…,σ M′ ]
[0435] Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤r≤M′.
[0436] To mutate the connection weight W, randomly select a mutation point v, and then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as:
[0437] W′=[w 11 , w 12 ,…,w v +δ,…,w 3M′ ]
[0438] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤v≤3M′.
[0439] The variation of bias B randomly selects a variation point z, and then adds a random number that follows a normal distribution to the zth element to generate a new B vector, which is expressed as:
[0440] B′=[b1,b2,…,b z +δ,…,b3]
[0441] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3.
[0442] Conditional termination submodule, used for termination conditions, if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, stop evolution, output the parameters and fitness of the optimal individual, and the corresponding radial basis neural network model; otherwise, let g = g + 1, return to calculate the fitness, and continue evolution.
[0443] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting the quality of abrasive jet cold cutting, characterized in that: The method comprises: Step S1: Collect cutting experimental data; the experimental data includes cutting process parameters and cutting process quality indicators, specifically including: The cross-sectional area A of the abrasive jet is expressed as: Where, d is the nozzle diameter; α is the semi-apex angle of the jet; L is the target distance; The velocity V of the abrasive jet is expressed as: Where, Q is the abrasive jet flow rate; V is the jet velocity; The kinetic energy of the jet E k It is expressed as: In the formula, E k is the kinetic energy of the abrasive jet; ρ is the density of the abrasive jet; The pressure P of the jet is expressed as: Where, P is the abrasive jet pressure; The impact force F of the jet is expressed as: Where F is the impact force of the abrasive jet; θ is the angle between the axis and the surface of the material being cut; The loss level LR is expressed as: Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; m n is the mass of the nozzle worn during the cutting process; ρ a is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n ; Energy consumption EC is expressed as: Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water; Cutting accuracy CP is expressed as: Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a is the surface roughness of the cut after cutting; D is the abrasive particle size; Step S2: normalizing the experimental data and dividing them into a training set and a test set; Step S3: constructing a radial basis function neural network model; the input layer of the radial basis function neural network model is the cutting process parameters, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error; Step S4: Optimizing the radial basis neural network model parameters using a genetic algorithm; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias, specifically including: Initialize the parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is P m , the number of elite individuals is E, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1]; Generate the initial population, randomly generate S individuals, each of which has M hidden neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as: X s =[M,C,∑,W,B],s=1,2,…,S Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1; Calculate the fitness, for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as: In the formula, l is a very small positive number, which is used to avoid the situation where the denominator is zero; Selection operation: Use the roulette method to select S individuals to enter the next generation, and retain E elite individuals, that is, the E individuals with the highest fitness, which do not participate in crossover and mutation operations and are directly copied to the next generation; The crossover operation uses an adaptive crossover probability, expressed as: Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation; For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including: The number of hidden neurons M is crossed, a crossover point p is randomly selected, and then the first p bits of binary codes of M of the two individuals are exchanged to generate two new M values, which are expressed as: M′1=M1[1:p]+M2[p+1:M max ] M′2=M2[1:p]+M1[p+1:M max ] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1; The intersection of the center vector C randomly selects an intersection point a, and then exchanges the first a elements of C of the two individuals to generate two new C vectors, expressed as: Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the βth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M min =min(M′1,M′2); Crossover of width parameter ∑ randomly selects a crossover point r, and then exchanges the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as: Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min ; Connect the cross of weight W, randomly select a crossover point f, and then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as: Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min ; Biased crossover of B randomly selects a crossover point h, and then exchanges the first h elements of B of the two individuals to generate two new B vectors, expressed as: B′1=[b 11 ,b 12 ,…,b 1h ,b 2h+1 ,b 2h+2 ,…,b 23 ] B′2=[b 21 ,b 22 ,…,b 2h ,b 1h+1 ,b 1h+2 ,…,b 13 ] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ is the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3; The mutation operation uses adaptive mutation probability, expressed as: Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation; For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including: The variation of the number of hidden layer neurons M randomly selects a crossover point o, and then flips the binary code of the oth bit to generate a new M value, which is expressed as: M′=M[1:o-1]+M[o]+M[o+1:M max ] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max ; The mutation of the center vector C randomly selects an intersection point u, and then adds a random number that follows a normal distribution to the u-th element to generate a new C vector, which is expressed as: C′=[c1,c2,…,c u +δ,…,c M ,] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤u≤M′; Variation of the width parameter ∑, randomly selecting a crossover point Then for the elements plus a random number that follows a normal distribution to generate a new ∑ vector, expressed as: Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ is a smaller standard deviation, which is used to control the amplitude of variation; To mutate the connection weight W, randomly select a mutation point v, and then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as: W′=[w 11 ,w 12 ,…,w v +δ,…,w 3M′ ] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤v≤3M′; The variation of bias B randomly selects a variation point z, and then adds a random number that follows a normal distribution to the zth element to generate a new B vector, which is expressed as: B′=[b1,b2,…,b z +δ,…,b3] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3; Termination condition: if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, stop the evolution, output the parameters and fitness of the optimal individual, and the corresponding radial basis neural network model; otherwise, set g = g + 1, return to calculate the fitness, and continue the evolution; Step S5: Use the test set to evaluate the prediction performance of the radial basis neural network model, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network model.
2. The abrasive jet cold cutting quality prediction method according to claim 1, characterized in that: The normalizing of the experimental data specifically includes: Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix; Data normalization is expressed as: In the formula, x ij and ij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x ij ′ and y ij ′ are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j ) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
3. The abrasive jet cold cutting quality prediction method according to claim 1, characterized in that: The construction of the radial basis neural network model specifically includes: The input layer of the radial basis neural network model is the cutting process parameters; the output layer is the cutting process quality index; the number of neurons in the hidden layer is M; The activation function of the hidden layer is expressed as: Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm; The output function of the output layer is expressed as: In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron; The loss function is expressed as: In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, the number of partial experimental data.
4. An abrasive jet cold cutting quality prediction system, characterized in that: The system comprises: The experimental data collection module is used to collect cutting experimental data; the experimental data includes cutting process parameters and cutting processing quality indicators, specifically including: The cross-sectional area A of the abrasive jet is expressed as: Where, d is the nozzle diameter; α is the semi-apex angle of the jet; L is the target distance; The velocity V of the abrasive jet is expressed as: Where, Q is the abrasive jet flow rate; V is the jet velocity; The kinetic energy of the jet E k It is expressed as: In the formula, E k is the kinetic energy of the abrasive jet; ρ is the density of the abrasive jet; The pressure P of the jet is expressed as: Where, P is the abrasive jet pressure; The impact force F of the jet is expressed as: Where F is the impact force of the abrasive jet; θ is the angle between the axis and the surface of the material being cut; The loss level LR is expressed as: Where LR is the loss level; m s is the mass of the sample after cutting; m a is the mass of abrasive consumed from the nozzle during the cutting process; m n is the mass of the nozzle worn during the cutting process; ρ a is the abrasive density; ρ n is the nozzle density; t is the cutting time; ∈ is the nozzle wear coefficient; m0 is the material mass before cutting, m0 = m s +m a +m n ; Energy consumption EC is expressed as: Where EC is energy consumption; W e is the electric energy consumed in the cutting process; W w is the water energy consumed in the cutting process, η e is the power conversion efficiency; ρ w is the density of water; c w is the specific heat capacity of water; ΔT is the temperature rise of water; Cutting accuracy CP is expressed as: Where CP is the cutting accuracy, which refers to the similarity between the cut after cutting and the expected cut; ΔL is the deviation between the cut length after cutting and the expected length; ΔW is the deviation between the cut width after cutting and the expected width; R a is the surface roughness of the cut after cutting; D is the abrasive particle size; A normalization processing and division module is used to normalize the experimental data and divide it into a training set and a test set; A radial basis network construction module is used to construct a radial basis neural network model; the input layer of the radial basis neural network model is the cutting process parameter, the output layer is the cutting process quality index, the activation function is the Gaussian function and the loss function is the mean square error; The network parameter optimization module is used to optimize the parameters of the radial basis neural network model using a genetic algorithm; the parameters include the number of hidden layer neurons, center vector, width, connection weight and bias, specifically including: Parameter initialization submodule, used to initialize parameters, set the population size to S, and the maximum number of evolution generations to G max , the initial crossover probability is P c , the initial mutation probability is P m , the number of elite individuals is E, the current generation is g = 0, and the upper limit of the number of hidden layer neurons is M max , the value range of the center vector and width parameter is [0, 1], and the value range of the connection weight and bias is [-1, 1]; The initial population generation submodule is used to generate the initial population and randomly generate S individuals. Each individual consists of M hidden layer neurons and a center vector C = [c1, c2, ..., c M ], width parameter ∑ = [σ1, σ2, …, σ M ], connection weight W = [w 11 , w 12 ,…,w M3 ] and bias B = [b1, b2, b3], expressed as: X s =[M,C,∑,W,B],s=1,2,…,S Where, X s is the sth individual; M is an integer, satisfying 1≤M≤M max ; C, ∑, W and B are all real vectors, satisfying 0≤C, ∑≤1, -1≤W, B≤1; The fitness calculation submodule is used to calculate the fitness for each individual X s , construct the RBF neural network according to its parameters, and use the training set to calculate its loss function Loss s , and then transform it into the fitness function F s , expressed as: In the formula, l is a very small positive number, which is used to avoid the situation where the denominator is zero; The selection submodule is used to perform selection operations. It uses the roulette method to select S individuals to enter the next generation, while retaining E elite individuals, that is, the E individuals with the highest fitness, which do not participate in crossover and mutation operations and are directly copied to the next generation; The crossover submodule is used to perform a crossover operation using an adaptive crossover probability, expressed as: Where P c,s is the crossover probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation; For each pair of adjacent individuals, a crossover operation is performed according to their crossover probability, that is, some or all of the parameters are exchanged to generate two new individuals, including: The number of hidden neurons M is crossed, a crossover point p is randomly selected, and then the first p bits of binary codes of M of the two individuals are exchanged to generate two new M values, which are expressed as: M′1=M1[1:p]+M2[p+1:M max ] M′2=M2[1:p]+M1[p+1:M max ] Where M1 and M2 are the original M values of the two individuals; M′1 and M′2 are the new M values of the two individuals; M[l:ζ] represents the binary code from the lth to the ζth bit of M; and 1≤p≤M max -1; The intersection of the center vector C randomly selects an intersection point a, and then exchanges the first a elements of C of the two individuals to generate two new C vectors, expressed as: Where C′1 and C′2 are the new C vectors of the two individuals; sβ represents the βth central vector element of the sth individual; a satisfies 1≤a≤M min ;M min Meet M min =min(M′1,M′2); Crossover of width parameter ∑ randomly selects a crossover point r, and then exchanges the first r elements of ∑ of the two individuals to generate two new ∑ vectors, expressed as: Where ∑′1 and ∑′2 are the new ∑ vectors of the two individuals; σ sγ represents the γth central vector element of the sth individual; r satisfies 1≤r≤M min ; Connect the cross of weight W, randomly select a crossover point f, and then exchange the first f elements of W of the two individuals to generate two new W vectors, expressed as: Where W′1 and W′2 are the new W vectors of the two individuals; sIJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron of the sth individual; f satisfies 1≤f≤3M min ; Biased crossover of B randomly selects a crossover point h, and then exchanges the first h elements of B of the two individuals to generate two new B vectors, expressed as: B′1=[b 11 ,b 12 ,…,b 1h ,b 2h+1 ,b 2h+2 ,…,b 23 ] B′2=[b 21 ,b 22 ,…,b 2h ,b 1h+1 ,b 1h+2 ,…,b 13 ] Where B′1 and B′2 are the new B vectors of the two individuals; b sJ is the bias of the Jth output layer neuron of the sth individual; h satisfies 1≤h≤3; The mutation submodule is used to perform mutation operations and uses an adaptive mutation probability, expressed as: Where P m,s is the mutation probability of the sth individual; g is the current generation; G max is the maximum evolutionary generation; For each individual, a mutation operation is performed according to its mutation probability, that is, a small perturbation is made to some or all of the parameters to generate a new individual, including: The variation of the number of hidden layer neurons M randomly selects a crossover point o, and then flips the binary code of the oth bit to generate a new M value, which is expressed as: M′=M[1:o-1]+M[o]+M[o+1:M max ] Where M′ is the new value of M; M[o] represents the inverse of the binary code of the oth bit of M, that is, 0 becomes 1 and 1 becomes 0; 1≤o≤M max ; The mutation of the center vector C randomly selects an intersection point u, and then adds a random number that follows a normal distribution to the u-th element to generate a new C vector, which is expressed as: C′=[c1,c2,…,c u +δ,…,c M′ ] Where C′ is the new C vector; δ is a vector that follows the normal distribution N(0, σ c ) of random numbers; σ c is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤u≤M′; Variation of the width parameter ∑, randomly selecting a crossover point Then for the elements plus a random number that follows a normal distribution to generate a new ∑ vector, expressed as: Where ∑′ is the new ∑ vector; δ is a vector that follows the normal distribution N(0, σ σ ) of random numbers; σ σ is a smaller standard deviation, which is used to control the amplitude of variation; To mutate the connection weight W, randomly select a mutation point v, and then add a random number that follows a normal distribution to the vth element to generate a new W vector, which is expressed as: W′=[w 11 ,w 12 ,…,w v +δ,…,w 3M′ ] Where W′ is the new W vector; δ is a vector that follows the normal distribution N(0, σ w ) of random numbers, σ w is a smaller standard deviation, used to control the amplitude of variation; it satisfies 1≤v≤3M′; The variation of bias B randomly selects a variation point z, and then adds a random number that follows a normal distribution to the zth element to generate a new B vector, which is expressed as: B′=[b1,b2,…,b z +δ,…,b3] Where B′ is the new B vector; δ is a vector that follows the normal distribution N(0, σ b ) of random numbers, σ b A smaller standard deviation is used to control the amplitude of variation; it satisfies 1≤z≤3; Conditional termination submodule, used for termination conditions, if the maximum evolutionary generation G is reached max , or the fitness change of the population is less than the set threshold, stop the evolution, output the parameters and fitness of the optimal individual, and the corresponding radial basis neural network model; otherwise, set g = g + 1, return to calculate the fitness, and continue the evolution; The model performance evaluation module is used to evaluate the prediction performance of the radial basis neural network model using a test set, calculate the error between the predicted value and the actual value; determine whether the error meets the error range, if it meets the error range, output the radial basis neural network model; if it does not meet the error range, retrain the radial basis neural network model.
5. The abrasive jet cold cutting quality prediction system according to claim 4, characterized in that: The normalization processing division module specifically includes: Where N is the number of experimental data; X is the cutting process parameter matrix; Y is the cutting process quality index matrix; Data normalization is expressed as: In the formula, x ij and ij are the cutting process parameters and cutting quality indicators of the i-th sample respectively; x ij ′ and y ij ′ are the normalized cutting process parameters and cutting quality indicators of the i-th sample; x j and j are the cutting process parameters and cutting quality indicators of the jth column respectively; min(x j ) and max(x j ) are the minimum and maximum values of the cutting process parameters in the jth column respectively; min(y j ) and max(y j ) represent the minimum and maximum values of the cutting quality index of the jth column respectively.
6. The abrasive jet cold cutting quality prediction system according to claim 4, characterized in that: The radial basis network construction module specifically includes: The input layer of the radial basis neural network model is the cutting process parameters; the output layer is the cutting process quality index; the number of neurons in the hidden layer is M; The activation function of the hidden layer is expressed as: Where x is the data of the input layer, that is, the normalized cutting process parameters; c I is the center vector of the I-th hidden layer neuron; σ I is the width parameter of the I-th hidden layer neuron; φ is the radial basis function, that is, the Gaussian function; ||·|| is the Euclidean norm; The output function of the output layer is expressed as: In the formula, y J (x) is the output value of the Jth output layer neuron, i.e., the normalized cutting quality index; w IJ is the connection weight from the Ith hidden layer neuron to the Jth output layer neuron; b J is the bias of the Jth output layer neuron; The loss function is expressed as: In the formula, y kJ is the actual value of the Jth cutting quality index of the kth sample; is the predicted value of the Jth cutting quality index of the kth sample; is the number of samples in the training set, that is, the number of partial experimental data.
Citation Information
Patent Citations
Distributed big data system risk predicating method based on AHP-RBF
CN104978612A
Edge chamfering by mechanically processing laser cut glass
CN106102983A
Steel rail repair process based on high-pressure abrasive water jet
CN115122242A
Soil mass engineering parameter prediction method capable of considering damage influence of soil skeleton
CN115936211A
Metal powder recovery structure of abrasive jet cold cutting robot
CN117245568A
