A method for optimizing GH4169 milling process parameters based on RFR-IDBO algorithm

Abstract

The application provides a GH4169 milling process parameter optimization method based on an RFR-IDBO algorithm, which is used for GH4169 high-temperature alloy, and the method takes a milling speed, a feed per tooth and an axial cutting depth as inputs, takes a surface roughness, a microhardness and a machining energy consumption of a GH4169 high-temperature alloy workpiece as outputs, establishes RFR models between the inputs and the three outputs respectively, improves the optimization ability of a DBO algorithm through improvement on the DBO algorithm, then weights and fuses the three established RFR models to obtain a comprehensive evaluation model, takes the model as a fitness function of the improved DBO algorithm, and optimizes the GH4169 milling process parameters by using the algorithm IDBO; and the application can realize GH4169 milling process parameter optimization.

Description

TECHNICAL FIELD

[0001] The application relates to the field of production manufacturing technology, and particularly relates to a GH4169 milling process parameter optimization method based on an RFR-IDBO algorithm. BACKGROUND

[0002] High-temperature alloys have excellent performance, high hardness and strength, good thermal stability, corrosion resistance, and good fatigue resistance. Therefore, they are widely used in the fields of aerospace, gas turbines, and petroleum chemical industry, and play an indispensable role in important parts of aircraft engines and steam turbines. According to the classification of organizational components, high-temperature alloys can be generally divided into iron-based, cobalt-based and nickel-based high-temperature alloys. Among them, the most widely used is the nickel-based high-temperature alloy. According to statistics, more than 50% of aircraft engine materials are composed of nickel-based high-temperature alloys, and among the nickel-based high-temperature alloys, GH4169 is the most widely used.

[0003] Due to the characteristics of low thermal conductivity, high temperature strength, high hardness and the like of the high-temperature alloy GH4169, the processing performance thereof is poor, and it belongs to a typical difficult-to-machine material. Therefore, during the machining process of GH4169, parts with unqualified surface quality are easily generated. The main machining mode of GH4169 is milling, and GH4169 is often used in key parts in important fields such as aerospace, and works under extreme working conditions such as high temperature and high pressure for a long time, so the machining surface quality requirement thereof is high. According to a large amount of research, the milling process parameters are the main factors affecting the surface roughness and surface microhardness of the GH4169 workpiece. The influence of the milling process parameters on the surface roughness and surface microhardness is very complex, so it is relatively complex to select appropriate process parameters in the milling process. Selecting appropriate machining process parameters helps to reduce the generation of unqualified parts. With the continuous development of industrialization, China's carbon emissions have ranked first in the world. In September 2020, China officially proposed "carbon peak" and "carbon neutrality". According to statistics, domestic industrial energy consumption accounts for about two-thirds of total social energy consumption. The energy consumption of machine tools in the machining process accounts for about 75% of the manufacturing industry, but the effective energy consumption of machine tools in the process of machining workpieces accounts for only about 30% of the total energy consumption. Therefore, in order to realize the national "double carbon strategy", it is of great significance to optimize the energy consumption in the machining process. Unreasonable milling parameters will lead to high energy consumption and low machining efficiency of machine tools during machining. Therefore, it is of great significance to carry out the optimization research of the milling process parameters of GH4169 for improving the milling quality of GH4169 and promoting the application of GH4169 in related fields.

[0004] Most current research uses traditional modeling methods for parameter optimization. However, different materials and equipment can lead to inaccurate models built using empirical formulas. Some studies use algorithms to optimize process parameters, but most use native algorithms, which are prone to getting trapped in local optima. Summary of the Invention

[0005] This invention proposes a method for optimizing milling process parameters of GH4169 based on the RFR-IDBO algorithm. Using GH4169 milling process parameters (milling speed, feed per tooth, axial depth of cut, radial depth of cut) along with GH4169 surface roughness, machining energy consumption, and microhardness as datasets, an improved RFR-IDBO algorithm is proposed, which combines Random Forest Regressor (RFR) with the Improved Dung Beetle Optimization (IDBO) algorithm, to achieve optimization of GH4169 milling process parameters.

[0006] The present invention adopts the following technical solution.

[0007] An optimization method for milling process parameters of GH4169 based on the RFR-IDBO algorithm is proposed for GH4169 high-temperature alloy. This method takes milling speed, feed per tooth, and axial depth of cut as inputs, and surface roughness, microhardness, and machining energy consumption of the GH4169 high-temperature alloy workpiece as outputs. RFR models are established between the inputs and the three outputs. The optimization capability of the DBO algorithm is improved by modifying it. Then, the three established RFR models are weighted and fused to obtain a comprehensive evaluation model. This model is used as the fitness function of the improved DBO algorithm, and the IDBO algorithm is used to optimize the milling process parameters of GH4169.

[0008] The optimization method includes the following steps;

[0009] Step S1: Using the surface roughness, microhardness and machining energy consumption of GH4169 material as optimization targets, and milling speed, feed per tooth and axial depth of cut as optimization variables, design an experimental scheme.

[0010] Step S2: Conduct the experiment according to the experimental plan, obtain the experimental data, and organize the dataset into a surface roughness dataset D. R Microhardness dataset D H and processing energy consumption dataset D E Data set D is divided according to a set ratio. R Divided into training set D t and verification set D v n subsets D are obtained using the bootstrap sampling method.t1 D t2 ..., D tn , respectively, train n CART (Classification And Regression Tree) decision tree models as base learners of the RFR model, and the prediction results of the n decision tree models are comprehensively averaged as the final prediction result of the RFR performance, finally obtaining the surface roughness model RFR1, and the microhardness model RFR2 and the machining energy consumption model RFR3 are obtained in the same way;

[0011] Step S3, weighting the three models RFR1, RFR2 and RFR3 obtained by training to obtain a multi-objective comprehensive evaluation model F m ;

[0012] Step S4, taking the multi-objective comprehensive evaluation model F m as the fitness function of the improved DBO algorithm, calculating the fitness value, and using the improved DBO algorithm to optimize the process parameters to obtain the optimal milling process parameters, and realize the optimization of the milling process parameters of GH4169.

[0013] Step S1 includes the following steps:

[0014] Step S1.1, taking the milling speed, feed per tooth and axial cutting depth as experimental factors, setting the levels of each factor, and designing orthogonal experiments; each factor is specifically:

[0015] Milling speed v min ≤v c ≤v max

[0016] Feed per tooth f min ≤f z ≤f max

[0017] Axial cutting depth a pmin ≤a p ≤a pmax

[0018] Among them

[0019] v min , v max are the minimum and maximum values of the milling speed v c , respectively;

[0020] f min , f max are the minimum and maximum values of the feed per tooth f z , respectively;

[0021] a pmin , a pmaxrespectively the minimum and maximum values of the axial depth of cut a p ;

[0022] Step S1.2, milling experiment is carried out, after each milling experiment, the surface roughness of the milled part, microhardness and processing energy consumption are obtained.

[0023] Step S2 includes the following steps:

[0024] Step S2.1, according to the designed experimental scheme, experimental data are obtained and the surface roughness data set is arranged as D R ={(x1,y1),(x2,y2),...,(x i ,y i )...,(x m ,y m )} , the data set is divided into training set D t and validation set D v according to the set proportion, a new sub-training set is formed by sampling m times from the training set D t with replacement, and n sub-training sets D t1 ,D t2 ,...,D ti ,...,D tn are obtained by repeating n times.

[0025] Step S2.2, the data is normalized, the input variable of the original data set is x={x1,x2,...,x i ,...,x m}, the normalization method is: the output variable is y={y1,y2,...,y i ,...,y m}, the normalization method is: the data after transformation is D R ={(x s1 ,y s1 ),(x s2 ,y s2 ),...,(x si ,y si )...,(x sm ,y sm )}

[0026] wherein,

[0027] x i , y i are the initial values of the input variable and the output variable respectively;

[0028] x s1 , y s1are the normalized values of the input variable and the output variable, respectively;

[0029] Min(x), Min(y) are the minimum values of the original input variable and the original output variable, respectively;

[0030] Max(x), Max(y) are the maximum values of the original input variable and the original output variable, respectively;

[0031] Step S2.3, using a sub-training set D ti Train the CART decision tree, recursively divide each region into two sub-regions and determine the output value on each sub-region within the input space where the training set is located, and construct the decision tree;

[0032] Step S2.4, select the optimal partition feature j and partition point s, for the regression problem, the CART decision tree uses the mean square error to measure the pros and cons of the feature partition point, for any partition feature j, the corresponding any partition point s divides the training set into D1 and D2 two parts, find the feature and feature value partition point corresponding to the minimum mean square error of D1 and D2 respectively, and the sum of the mean square error of D1 and D2 is the minimum. Calculate the mean square error of each partition point of each feature:

[0033]

[0034] Wherein,

[0035] c1 is the sample output value of the training set D1, D1 is a sub-region of D ti , D1(j, s) = {x | x (j) ≤ s};

[0036] c2 is the sample output value of the training set D2, D2 is another sub-region of D ti , D2(j, s) = {x | x (j) > s};

[0037] x i represents the input variable (or feature vector) of a sample in the training data set, which contains the values of multiple features or attributes;

[0038] y i represents the value of the target variable associated with x i .

[0039] It is easy to know that c1, c2 that make the square error in each region minimum is the mean value of y in the corresponding region, so the above formula can be modified as,

[0040]

[0041] Wherein,

[0042] d1 is the number of samples in D1;

[0043] d2 is the number of samples in D2;

[0044] Step S2.5, form a pair of (j, s) for each feature and its corresponding each partition point, and substitute each (j, s) pair into the formula to calculate, and select the (j, s) pair that makes the value of the formula minimum;

[0045] Step S2.6, divide the region with the selected (j, s) and determine the corresponding output value

[0046]

[0047] wherein,

[0048] D1(j, s) = {x | x (j) ≤ s};

[0049] D2(j, s) = {x | x (j > s};

[0050] Step S2.7, repeat the process of calling steps S2.5 to S2.6 on the two sub-regions until the set partition stop condition is met;

[0051] Step S2.8, divide the input space into M regions D1, D2,..., D i ,..., D M , and generate a CART decision tree:

[0052] wherein I is an indicator function,

[0053]

[0054] Step S2.9, repeat the process of steps S2.2 to S2.8, and train decision trees respectively using D t1 ,D t2 ,..., D ti ,..., D tn sub-training sets, and finally obtain n CART decision tree models M1, M2,..., M i ,..., M n ;

[0055] Step S2.10, use each decision tree for prediction, and n groups of prediction results c1, c2,..., c i ,..., c n; step S2.11, using the average method to synthesize the prediction results of n decision trees to obtain the final prediction result where c i is the i-th prediction result, and finally the surface roughness model RFR1 is obtained;

[0056] Step S2.12, repeating steps S2.1 to S2.11 to obtain the microhardness model RFR2 and the machining energy consumption RFR3.

[0057] The specific implementation of step S3 includes the following method: using the artificial weighting method to give the same weight to each model of the surface roughness model RFR1, the microhardness model RFR2 and the machining energy consumption RFR3 obtained in step S2, and performing weighted combination to construct a comprehensive evaluation model F m .

[0058] The specific implementation of step S4 includes the following steps:

[0059] Step S4.1, taking the constructed comprehensive evaluation model F m as the fitness function of the improved DBO algorithm to calculate the fitness value, and using the improved DBO algorithm to optimize the surface roughness, microhardness and machining energy consumption of GH4169, so as to obtain the milling process parameters corresponding to the optimal milling surface roughness, microhardness and machining energy consumption, thereby realizing the optimization of the milling process parameters;

[0060] Step S4.2, using the improved DBO algorithm to optimize the process parameters, setting the total number of scarab individuals as N, the maximum iteration number as T max , initializing the scarab individual as X={x1,x2,x3,x4}, setting the dimension x1 as the size of the milling speed v c , the dimension x2 as the size of the feed per tooth f z , the dimension x3 as the size of the axial depth of cut a p , and the dimension x4 as the size of the radial depth of cut a e ; setting the population variables for the four dimensions, and initializing the population using the Piecewise chaotic mapping:

[0061]

[0062] wherein,

[0063] p is the chaotic coefficient, and the value is 0.4;

[0064] represents the chaotic initialization value of the j-1th dimension of the i-th population individual, wherein x i,j represents the value of the j-1th dimension of the i-th population individual;

[0065] Lb1 , Ub 1 denotes the milling speed v c the minimum and maximum values of the constraint;

[0066] Lb 2 , Ub 2 denotes the feed per tooth f z the minimum and maximum values of the constraint;

[0067] Lb 3 , Ub 3 denotes the axial depth of cut a p the minimum and maximum values of the constraint;

[0068] Lb 4 , Ub 4 denotes the radial depth of cut a p the minimum and maximum values of the constraint;

[0069] Step S4.3, put the initialized individual of the dung beetle population into the comprehensive evaluation model established in S3 to obtain the fitness Y = {y1, y2,..., y N}

[0070] wherein,

[0071] Y denotes the fitness value of the dung beetle population;

[0072] y i denotes the fitness value of the individual of the dung beetle;

[0073] Step S4.4, sort the fitness of the individual of the dung beetle, wherein the individual with better fitness is the ball rolling dung beetle, which performs the ball rolling or dancing behavior, the second is the breeding dung beetle, the third is the small dung beetle, and the fourth is the stealing dung beetle, and the population proportion of the four kinds of dung beetles needs to be set according to specific problems. The ball rolling dung beetle is responsible for quickly moving the food to prevent it from being robbed by other dung beetles, the breeding dung beetle is responsible for laying eggs to breed the next generation, the small dung beetle is responsible for foraging, and the stealing dung beetle will rob the food of other dung beetles;

[0074] Step S4.5, update the position of the ball rolling dung beetle, and plan to process both the case of no obstacle and the case of obstacle;

[0075] Step S4.6, the small dung beetle that grows into an adult will drill out from the ground to forage, and the best foraging area needs to be established to guide the dung beetle to forage. The best foraging area is dynamically changed, and the foraging area boundary selection strategy is as follows: Lb b = max(X b ×(1-R), Lb)

[0076] Ub b = min(X b ×(1+R), Ub)

[0077] wherein,

[0078] Lb b is the lower bound of the optimal foraging region;

[0079] Ub b is the upper bound of the optimal foraging region;

[0080] X b is the global optimal foraging position;

[0081] R = 1 - t / T max , T max denotes the maximum number of iterations;

[0082] Lb denotes the lower bound of the optimization problem;

[0083] Ub denotes the upper bound of the optimization problem.

[0084] The foraging behavior position update formula is as follows:

[0085] x i (t+1) = x i (t) + C1 x i (t) - Lb b ) + C2 x i (t) - Ub b )

[0086] wherein,

[0087] C1 denotes a random number subject to normal distribution, i.e. C1 ~ N(0, 1);

[0088] C2 denotes a 1 x D random vector belonging to (0, 1);

[0089] In the small scarab beetles, adaptive t-distributed variation is introduced, and a [0, 1] random number rand is introduced, when rand > 0.5, the mutation operation is executed

[0090] x i (t+1) = X b + t(iter) X b

[0091]

[0092] wherein

[0093] x i (t+1) is the new position of the i-th scarab beetle individual after the adaptive t-distributed variation is executed;

[0094] t(iter) is the t-distribution adopted by the algorithm;

[0095] ν is the degree of freedom parameter of t-distribution;

[0096] X b is the global best foraging position;

[0097] Step S4.7, the female O. marginata will select a suitable oviposition site to provide a safe environment for offspring. Once the oviposition area is determined, the female O. marginata will select the oviposition area for oviposition. The oviposition area boundary selection strategy of the breeding O. marginata is as follows:

[0098] Lb * = max(X * ×(1-R), Lb)

[0099] Ub * = min(X * ×(1+R), Ub)

[0100] wherein,

[0101] Lb * represents the lower boundary of the oviposition area;

[0102] Ub * represents the upper boundary of the oviposition area;

[0103] X * represents the current local best position;

[0104] Each female O. marginata produces only one egg in each iteration. The boundary range of the oviposition area is dynamically changed with the number of iterations, which is mainly determined by the value of R. The position of the egg ball is also dynamically changed;

[0105] The position update formula of the breeding behavior is:

[0106] B i (t+1) = X * + b1×(B i (t) - Lb * ) + b2×(B i (t) - Ub * )

[0107] wherein,

[0108] B i (t) is the position information of the i-th egg ball at the t-th iteration (which is actually an individual x i );

[0109] b1 and b2 are two independent random vectors with a size of 1xD, and D represents the dimension of the optimization problem. The position of the egg ball is strictly limited within the oviposition area;

[0110] Step S4.8, some O. marginata in the O. marginata population will steal food from other O. marginata, Xb For the best food source position, that is, to compete for the best position of food, the position update formula of the thief is as follows:

[0111] x i (t+1)=X b +S×g×(|x i (t)-X * |+|x i (t)-X b |)

[0112] Wherein,

[0113] g is a random vector subject to normal distribution, with a size of 1×D;

[0114] S is a constant value.

[0115] In step S4.5, the algorithm handles the case without obstacles as follows:

[0116] Suppose that when there is no obstacle, the beetle uses the sun to navigate and moves in a given direction in the entire search space, the intensity of the light source will affect the trajectory of the beetle, and natural factors will cause the beetle to deviate from the original direction; then during the rolling process, the position update formula of the rolling beetle is as follows:

[0117] x i (t+1)=x i (t)+α×k×x i (t-1)+b×Δx

[0118] Δx=|x i (t)-X ω |

[0119] Wherein,

[0120] t represents the current iteration number of the algorithm;

[0121] x i (t) represents the position information of the i-th beetle at the t-th iteration;

[0122] k is a constant, with a value range of (0, 0.2], representing a deflection coefficient, generally set to 0.1;

[0123] b is a constant, with a value range of (0, 1), generally set to 0.3;

[0124] α is a natural coefficient, with a value of 1 or -1, -1 indicating deviation from the original direction, and 1 indicating no deviation;

[0125] X ω is the global worst position;

[0126] Δx simulates the change of light intensity;

[0127] The global exploration stage position updating strategy of the fish eagle optimization algorithm is introduced into the ball roller:

[0128] x i (t+1)=x i (t)+q×(G-H×x i (t))

[0129] Wherein,

[0130] q is a random number between 0 and 1;

[0131] G is a better one of the current positions of the roller;

[0132] H is a random number in the set {1,2}.

[0133] In step S4.5, the algorithm has the following processing method for the case of obstacles:

[0134] When the roller encounters an obstacle, it will adopt a dance behavior to reposition and select a new direction for position updating. A tangent function is used to simulate the dance behavior, and the dance behavior position updating is as follows:

[0135] x i (t+1)=x i (t)+tanθ|x i (t)-x i (t-1)|

[0136] Wherein,

[0137] θ is the deflection angle, and the value range is [0, π]. If θ = 0, π / 2, π, the position x i is not updated.

[0138] |x i (t)-x i (t-1)| represents the difference between the position of the i-th roller at the t-th iteration and the position at the t-1-th iteration.

[0139] The GH4169 milling process adopts a hard alloy cutter.

[0140] The GH4169 milling process is carried out under dry milling conditions.

[0141] The present application aims at the deficiency of domestic and foreign research on the GH4169 milling process parameter optimization method, and provides a GH4169 milling process parameter optimization method, which takes the GH4169 milling process parameters (milling speed, feed per tooth, axial depth of cut, radial depth of cut) and the GH4169 surface roughness, processing energy consumption and microhardness as a data set, proposes an improved RFR-IDBO algorithm combining random forest regression (Random forest Regressor, RFR) and improved dung beetle optimization algorithm (Improved DungBeetle Optimization, IDBO), and realizes the optimization of GH4169 milling process parameters.

[0142] The present application relates to the field of machining, and aims at the deficiency of domestic and foreign literature in the GH4169 high-temperature alloy milling process parameter optimization method, and proposes a GH4169 milling process parameter optimization method based on RFR-IDBO algorithm. The method takes the milling speed, feed per tooth and axial depth of cut as the input, and the surface roughness, microhardness and processing energy consumption of the GH4169 high-temperature alloy workpiece as the output, respectively establishes the RFR model between the input and the three outputs, and improves the DBO algorithm to improve the optimization ability of the DBO algorithm. Then the three established RFR models are weighted and fused to obtain a comprehensive evaluation model, and the model is taken as the fitness function of the IDBO algorithm.

[0143] The present application is a scheme for optimizing the GH4169 milling process parameters by using the IDBO algorithm. When comparing the optimization results with the unimproved RFR-DBO algorithm, the results show that the RFR-IDBO algorithm of the present application has better effect. BRIEF DESCRIPTION OF DRAWINGS

[0144] The present application will be further described in detail below in combination with the drawings and specific embodiments:

[0145] The accompanying drawings are schematic flowcharts of the present application. Figure 1 DETAILED DESCRIPTION

[0146] ​As shown in the figure, a GH4169 milling process parameter optimization method based on RFR-IDBO algorithm is used for GH4169 high-temperature alloy, the method takes milling speed, feed per tooth, axial depth of cut as input, and the surface roughness, microhardness, and machining energy consumption of the GH4169 high-temperature alloy workpiece as output, and establishes RFR models between the input and the three outputs respectively, and improves the optimization ability of the DBO algorithm of the dung beetle algorithm through the improvement of the DBO algorithm of the dung beetle algorithm, and then the three established RFR models are weighted and fused to obtain a comprehensive evaluation model, which is used as the fitness function of the improved dung beetle algorithm IDBO, and the GH4169 milling process parameters are optimized by using the algorithm IDBO.

[0147] The optimization method comprises the following steps:

[0148] Step S1, taking the surface roughness, microhardness and machining energy consumption of GH4169 material milling as optimization objectives, and taking milling speed, feed per tooth and axial depth of cut as optimization variables, an experimental scheme is designed;

[0149] Step S2, according to the experimental scheme, experiments are carried out, experimental data are obtained, and the data set is arranged into surface roughness data set D R , microhardness data set D H and machining energy consumption data set D E , the data set D R is divided into training set D t and verification set D v according to the set proportion, n sub-training sets D t1 , D t2 ,..., D tn are obtained by using the bootstrap sampling method, n CART (Classification And Regression Tree) decision tree models are trained respectively as the base learners of the RFR model, the prediction results of the n decision tree models are comprehensively averaged as the final prediction results of the RFR performance, and finally the surface roughness model RFR1 is obtained, and the microhardness model RFR2 and the machining energy consumption model RFR3 are obtained in the same way;

[0150] Step S3, the three models RFR1, RFR2 and RFR3 trained are weighted to obtain a multi-objective comprehensive evaluation model F m ;

[0151] Step S4, taking the multi-objective comprehensive evaluation model F m as the fitness function of the improved DBO algorithm, calculating the fitness value, and optimizing the process parameters by using the improved DBO algorithm to obtain the optimal milling process parameters, and realizing the optimization of the GH4169 milling process parameters.

[0152] Step S1 comprises the following steps;

[0153] Step S1.1, taking the milling speed, the feed per tooth, the axial cutting depth as experimental factors, setting the levels of each factor, designing an orthogonal experiment; each factor is specifically:

[0154] Milling speed v min ≤v c ≤v max

[0155] Feed per tooth f min ≤f z ≤f max

[0156] Axial cutting depth a pmin ≤a p ≤a pmax

[0157] Wherein

[0158] v min , v max are the minimum and maximum values of the milling speed v c respectively;

[0159] f min , f max are the minimum and maximum values of the feed per tooth f z respectively;

[0160] a pmin , a pmax are the minimum and maximum values of the axial cutting depth a p respectively;

[0161] Step S1.2, milling experiment is carried out, after each milling experiment, the surface roughness, microhardness and processing energy consumption of the milled part are obtained.

[0162] Step S2 comprises the following steps;

[0163] Step S2.1, according to the designed experimental scheme, experimental data are obtained and the surface roughness data set is arranged as D R ={(x1,y1),(x2,y2),...,(x i ,y i )...,(x m ,y m )}; according to the set proportion, the data set is divided into training set D t and verification set D v ; using the bootstrap sampling method, the training set D t is sampled m times with replacement, a new sub-training set is formed, and the process is repeated n times to obtain n sub-training sets D t1D t2 ,...,D ti ,...,D tn ;

[0164] Step S2.2, normalize the data, the input variable of the original data set is x = {x1, x2,..., x i ,...,x m}, the normalization method is: The output variable is y = {y1, y2,..., y i ,...,y m}, the normalization method is: The data after transformation is D R = {(x s1 ,y s1 ), (x s2 ,y s2 ),..., (x si ,y si ),..., (x sm ,y sm )}

[0165] Wherein,

[0166] x i , y i are the initial values of the input variable and the output variable, respectively;

[0167] x s1 , y s1 are the normalized values of the input variable and the output variable, respectively;

[0168] Min(x), Min(y) are the minimum values of the original input variable and the original output variable, respectively;

[0169] Max(x), Max(y) are the maximum values of the original input variable and the original output variable, respectively;

[0170] Step S2.3, use a sub-training set D ti to train the CART decision tree, recursively divide each region into two sub-regions and determine the output value on each sub-region in the input space where the training set is located, and build a decision tree;

[0171] Step S2.4, select the optimal partition feature j and partition point s, for the regression problem, CART decision tree uses mean square error to measure the advantages and disadvantages of feature partition point, for any partition feature j, the corresponding arbitrary partition point s divides the training set into D1 and D2 two parts, find the feature and feature value partition point corresponding to the minimum mean square error of D1 and D2 respectively, and the minimum sum of mean square error of D1 and D2. Calculate the mean square error of each partition point of each feature:

[0172]

[0173] wherein,

[0174] c1 is the sample output value of D1 training set, D1 is a sub-region of D ti , D1(j, s) = {x | x (j) ≤ s};

[0175] c2 is the sample output value of D2 training set, D2 is another sub-region of D ti , D2(j, s) = {x | x (j) > s};

[0176] x i represents the input variable (or feature vector) of a sample in the training data set, containing the values of multiple features or attributes;

[0177] y i represents the value of the target variable associated with x i .

[0178] It is known that c1, c2 that make the square error in each region minimum are the mean values of y in the corresponding regions, so the above formula can be modified as,

[0179]

[0180] wherein,

[0181] d1 is the number of samples in D1;

[0182] d2 is the number of samples in D2;

[0183] Step S2.5, each feature and its corresponding each division point to form a (j, s) pair, each (j, s) pair is substituted into the formula for calculation, and the (j, s) pair that makes the formula value minimum is selected;

[0184] Step S2.6, the selected (j, s) divides the region and determines the corresponding output value

[0185]

[0186] wherein,

[0187] D1(j, s) = {x | x (j) ≤ s};

[0188] D2(j, s) = {x | x (j) > s};

[0189] Step S2.7, repeat the process of calling steps S2.5 to S2.6 on two sub-regions until the set partition stop condition is met;

[0190] Step S2.8, divide the input space into M regions D1, D2,...,D i ,...,D M , generate a CART decision tree:

[0191] where I is an indicator function,

[0192]

[0193] Step S2.9, repeat the process of steps S2.2 to S2.8, respectively, using D t1 ,D t2 ,...,D ti ,...,D tn sub-training set to train the decision tree, finally get n CART decision tree models M1, M2,...,M i ,...,M n ;

[0194] Step S2.10, use each decision tree for prediction, can get n groups of prediction results c1, c2,...c i ,...,c n ; Step S2.11, using the average method to integrate the prediction results of n decision trees, get the final prediction result where c i is the i-th prediction result, finally get the surface roughness model RFR1;

[0195] Step S2.12, repeat steps S2.1 to S2.11, get the microhardness model RFR2, processing energy consumption RFR3.

[0196] The specific implementation of step S3 includes the following method: the surface roughness model RFR1, microhardness model RFR2 and processing energy consumption RFR3 obtained by step S2 are given the same weight of each model by artificial weighting method, weighted combination to construct a comprehensive evaluation model F m .

[0197] The specific implementation of step S4 includes the following steps:

[0198] Step S4.1, the comprehensive evaluation model F mThe fitness value is calculated using the fitness function of the improved DBO algorithm, and the surface roughness, microhardness and machining energy consumption of GH4169 are optimized using the improved DBO algorithm. The milling process parameters corresponding to the optimal milling surface roughness, microhardness and machining energy consumption are obtained, thereby realizing the optimization of milling process parameters.

[0199] Step S4.2: Optimize process parameters using the improved DBO algorithm, setting the total number of dung beetle individuals to N and the maximum number of iterations to T. max Initialize the dung beetle individuals as X = {x1, x2, x3, x4}, and let dimension x1 be the milling speed v. c Size, dimension x2 is the feed per tooth f z Size, dimension x3 is the axial depth a p The size, dimension x4 is the radial depth a e The size of the population; setting population variables for four dimensions, and initializing the population using a Piecewise chaotic mapping:

[0200]

[0201] in,

[0202] p is the chaos coefficient with a value of 0.4;

[0203] Let represent the chaotic initialization value of the (j-1)th dimension of the i-th individual in the population, where x i,j This represents the value of the (j-1)th dimension of the i-th individual in the population;

[0204] Lb 1 Ub 1 Indicates milling speed v c Minimum and maximum values ​​of constraints;

[0205] Lb 2 Ub 2 f represents the feed per tooth. z Minimum and maximum values ​​of constraints;

[0206] Lb 3 Ub 3 Indicates the axial depth of cut a p Minimum and maximum values ​​of constraints;

[0207] Lb 4 Ub 4 Indicates radial depth of cut a p Minimum and maximum values ​​of constraints;

[0208] Step S4.3, put the initialized individual of the dung beetle population into the comprehensive evaluation model established in S3 to obtain the fitness Y={y1, y2,..., y N}

[0209] wherein,

[0210] Y represents the fitness value of the dung beetle population;

[0211] y i represents the fitness value of the individual of the dung beetle;

[0212] Step S4.4, sort the fitness of the individual of the dung beetle, wherein the individual with better fitness is the ball rolling dung beetle, which performs the ball rolling or dancing behavior, the second is the breeding dung beetle, the third is the small dung beetle, and the fourth is the stealing dung beetle, and the population proportion of the four kinds of dung beetles needs to be set according to specific problems. The ball rolling dung beetle is responsible for quickly moving the food to prevent it from being robbed by other dung beetles, the breeding dung beetle is responsible for laying eggs to breed the next generation, the small dung beetle is responsible for foraging, and the stealing dung beetle will rob the food of other dung beetles;

[0213] Step S4.5, update the position of the ball rolling dung beetle, and plan to process both the case of no obstacle and the case of obstacle;

[0214] Step S4.6, the small dung beetle that grows into an adult will drill out from the underground to forage, and the best foraging area needs to be established to guide the dung beetle to forage, and the best foraging area is dynamically changed, and the foraging area boundary selection strategy is as follows: Lb b = max(X b ×(1-R), Lb)

[0215] Ub b = min(X b ×(1+R), Ub)

[0216] wherein,

[0217] Lb b is the lower limit of the optimal foraging area;

[0218] Ub b is the upper limit of the optimal foraging area;

[0219] X b is the global optimal foraging position;

[0220] R=1-t / T max , T max represents the maximum number of iterations;

[0221] Lb represents the lower limit of the optimization problem;

[0222] Ub represents the upper limit of the optimization problem.

[0223] The foraging behavior position update formula is as follows:

[0224] x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b )

[0225] wherein,

[0226] C1 represents a random number subject to normal distribution, that is, C1 ~ N(0, 1);

[0227] C2 represents a 1x D random vector belonging to (0, 1);

[0228] In the small scarab beetles, adaptive t-distributed variation is introduced, and a [0, 1] random number rand is introduced, when rand > 0.5, the mutation operation is executed

[0229] x i (t+1)=X b +t(iter)·X b

[0230]

[0231] wherein

[0232] x i (t+1) is the new position of the i-th scarab beetle individual after executing the adaptive t-distributed variation;

[0233] t(iter) is the t-distribution adopted by the algorithm;

[0234] ν is the degree of freedom parameter of the t-distribution;

[0235] X b is the global best foraging position;

[0236] Step S4.7, the female scarab beetle will select a suitable oviposition site to provide a safe environment for offspring, and once the oviposition area is determined, the female scarab beetle will select the area for oviposition; the oviposition area boundary selection strategy formula of the breeding scarab beetle is as follows:

[0237] Lb * =max(X * ×(1-R),Lb)

[0238] Ub * =min(X * ×(1+R),Ub)

[0239] wherein,

[0240] Lb * denotes the lower boundary of the egg-laying region;

[0241] Ub * denotes the upper boundary of the egg-laying region;

[0242] X * denotes the current local optimal position;

[0243] Each female scarab produces only one egg in each iteration. The boundary of the egg-laying region is dynamically changed with the number of iterations, which is mainly determined by the R value, because the position of the egg sphere is also dynamically changed;

[0244] The position update formula of the reproductive behavior is:

[0245] B * (t+1) = X i + b1 x (B * (t) - Lb i ) + b2 x (B * (t) - Ub i )

[0246] wherein,

[0247] B i (t) is the position information of the i-th egg sphere at the t-th iteration (actually, it is the individual x b );

[0248] b1 and b2 are two independent random vectors with a size of 1 x D, and D represents the dimension of the optimization problem. The position of the egg sphere is strictly limited within the egg-laying region;

[0249] Step S4.8, some scarabs in the scarab population will steal food from other scarabs, X i is the optimal food source position, that is, the optimal position for competing for food, and the position update formula of the stealing scarab is as follows:

[0250] x b (t+1) = X i + S x g x (x * (t) - X i | + |x b (t) - X i |)

[0251] wherein,

[0252] g is a random vector with a size of 1 x D, which is subject to a normal distribution;

[0253] S is a constant value.

[0254] In step S4.5, the algorithm handles the case without obstacles as follows:

[0255] When there is no obstacle, the sun is used by the dung beetle to navigate the movement in the whole search space along a given direction, the intensity of the light source will affect the trajectory of the movement of the dung beetle, and natural factors will cause the dung beetle to deviate from the original direction; then in the rolling process, the position update formula of the rolling dung beetle is as follows:

[0256] x i (t+1)=x i (t)+α×k×x i (t-1)+b×Δx

[0257] Δx=|x i (t)-X ω |

[0258] Wherein,

[0259] t represents the current iteration number of the algorithm;

[0260] x i (t) represents the position information of the ith dung beetle at the tth iteration;

[0261] k is a constant, the value range is (0, 0.2], which represents the deflection coefficient, and is generally set to 0.1;

[0262] b is a constant, the value range is (0, 1), and is generally set to 0.3;

[0263] α is a natural coefficient, the value is 1 or -1, -1 represents deviation from the original direction, and 1 represents no deviation;

[0264] X ω is the global worst position;

[0265] Δx simulates the change of light intensity;

[0266] The position update strategy of the global exploration stage of the fish eagle optimization algorithm is introduced into the rolling dung beetle:

[0267] x i (t+1)=x i (t)+q×(G-H×x i (t))

[0268] Wherein;

[0269] q is a random number between 0 and 1;

[0270] G is a better dung beetle at the current position;

[0271] H is a random number in the set {1, 2}.

[0272] In step S4.5, the algorithm handles the case with obstacles as follows:

[0273] When the dung beetle encounters an obstacle, it will adopt a dance behavior to reposition and choose a new direction for position updating. A tangent function is used to simulate the dance behavior, and the position updating of the dance behavior is as follows:

[0274] x i (t+1)=x i (t)+tanθ|x i (t)-x i (t-1)|

[0275] Wherein,

[0276] θ is the deflection angle, and the value range is [0, π]. If θ = 0, π / 2, π, the position x is not updated. i ;

[0277] |x i (t)-x i (t-1)| represents the difference between the position of the i-th dung beetle at the t-th iteration and the position at the t-1-th iteration.

[0278] The GH4169 milling process adopts a hard alloy cutter.

[0279] The GH4169 milling process is carried out under dry milling conditions.

[0280] Embodiment:

[0281] In this example, a high-temperature alloy GH4169 plate of lxdh = 100x80x5mm is side milled on a vertical machining center. The radial depth of cut is set to a fixed value of 5mm, and the milling speed, feed per tooth and axial depth of cut are factors. A three-factor four-level orthogonal experiment is designed, and the orthogonal experiment table is shown in Table 1.

[0282] Table 1 GH4169 orthogonal experiment design factor and level table

[0283]

[0284] Table 2 GH4169 orthogonal experiment design table

[0285]

[0286]

[0287] According to the experimental design of Table 2, side milling of the high-temperature alloy GH4169 plate with lxdxh = 100x80x5mm is carried out, and the surface roughness, microhardness and machining energy consumption of each milling experiment need to be obtained after the test is completed. The surface roughness is measured by a surface roughness measuring instrument, and the specific method is: after each milling experiment is completed, the surface roughness of the workpiece is measured three times and the average value is taken as the surface roughness value of the GH4169 milling processing; the microhardness is measured by HYHVS-1000A digital display automatic turret microhardness tester, and the specific method is: after each milling experiment is completed, the surface microhardness of the workpiece is measured three times and the average value is taken as the microhardness value of the GH4169 milling processing. The machining energy consumption is collected by LCDG-DG3-71760TH intelligent collector combined with Zigbee gateway host station. The specific data is shown in Table 3.

[0288] Table 3 GH4169 milling experiment data table

[0289]

[0290]

[0291] Unimproved DBO algorithm optimization result

[0292] Table 4 is the process parameter value (milling speed, feed per tooth, axial depth of cut, radial depth of cut) obtained by using the unimproved RFR-DBO algorithm, and the surface roughness, microhardness and machining energy consumption obtained by GH4169 milling experiment based on the process parameter are shown in the table.

[0293] Table 4 unimproved algorithm optimization process parameter milling experiment result table

[0294]

[0295] Improved DBO algorithm optimization result

[0296] Table 5 is the process parameter value (milling speed, feed per tooth, axial depth of cut, radial depth of cut) obtained by using the improved RFR-DBO algorithm, and the surface roughness, microhardness and machining energy consumption obtained by GH4169 milling experiment based on the process parameter are shown in the table.

[0297] Table 5 improved algorithm optimization process parameter milling experiment result table

[0298]

[0299]

[0300] In the above embodiment, the optimization result is compared with the unimproved RFR-DBO algorithm, and the result shows that the improved RFR-IDBO algorithm has better effect.

Claims

1. A GH4169 milling process parameter optimization method based on RFR-IDBO algorithm for GH4169 high-temperature alloy, characterized in that: The method takes the milling speed, the feed per tooth, and the axial depth of cut as inputs, and the surface roughness, the microhardness, and the machining energy consumption of the GH4169 high-temperature alloy workpiece as outputs, RFR models are respectively established between the inputs and the three outputs, the optimization ability of the dung beetle algorithm DBO is improved through the improvement of the DBO, and then the three established RFR models are weighted and fused to obtain a comprehensive evaluation model, the model is taken as the fitness function of the improved dung beetle algorithm IDBO, and the GH4169 milling process parameters are optimized by using the algorithm IDBO. The optimization method Comprise the following steps. Step S1, taking the milling surface roughness, the microhardness, and the machining energy consumption of the GH4169 material as optimization objectives, and taking the milling speed, the feed per tooth, and the axial depth of cut as optimization variables, an experimental scheme is designed. Step S2, perform experiments according to the experimental scheme, obtain experimental data and arrange the data set into the surface roughness data set D R , the microhardness data set D H and the machining energy consumption data set D E , divide the data set D R into a training set D t and a verification set D v according to a set proportion, obtain n sub-training sets D t1 ,D t2 ,...,D tn by using a bootstrap sampling method, train n CART decision tree models respectively as base learners of the RFR model, comprehensively average the prediction results of the n decision tree models as the final prediction result of the RFR performance, finally obtain the surface roughness model RFR1, and obtain the microhardness model RFR2 and the machining energy consumption model RFR3 in the same way; Step S3, the three trained models RFR1, RFR2, RFR3 are weighted to obtain a multi-objective comprehensive evaluation model F m ; Step S4, taking multi-target comprehensive evaluation model F m As the improved DBO algorithm fitness function, the fitness value is calculated, and the improved DBO algorithm is used to optimize the process parameters to obtain the optimal milling process parameters, and the optimization of GH4169 milling process parameters is realized.

2. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm according to claim 1, characterized in that: ； Step S1 comprises the following steps. Step S1.1, taking the milling speed, the feed per tooth, and the axial depth of cut as experimental factors, setting the levels of each factor, and designing an orthogonal experiment; each factor is specifically: milling speed v min ≤ v c ≤ v max feed per tooth f min ≤ f z ≤ f max axial cutting depth a pmin ≤ a p ≤ a pmax Wherein v min , v max are respectively the minimum and maximum values of the milling speed v c . f min , f max are the minimum and maximum values of the feed per tooth f z , respectively; a pmin , a pmax are minimum and maximum values of the axial cut depth a p , respectively; Step S1.2, milling processing experiment is performed, and after each milling experiment, the surface roughness, the microhardness, and the machining energy consumption of the milled part are obtained.

3. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm according to claim 1, characterized in that: Step S2 comprises the following steps. Step S2.1, perform experiments according to the designed experimental scheme to obtain experimental data and arrange the surface roughness data set as D R ={(x1,y1),(x2,y2),...,(x i ,y i )...,(x m ,y m )} , divide the data set into training set D t and validation set D v according to the set proportion, use the bootstrap sampling method to sample m times from the training set D t with replacement, form a new sub-training set, repeat n times to obtain n sub-training sets D t1 ,D t2 ,...,D ti ,...,D tn ; Step S2.2, normalizing the data, the input variable of the original data set is x = {x1, x2,..., x i ,...,x m}, the normalization method is: the output variable is y = {y1, y2,..., y i ,...,y m}, the normalization method is: the data after transformation is D R = {(x s1 ,y s1 ), (x s2 ,y s2 ),..., (x si ,y si ),..., (x sm ,y sm )} Wherein, x i , y i are initial values of the input variable and the output variable, respectively; x s1 , y s1 are the normalized values of the input and output variables, respectively; Min(x), Min(y) are the minimum values in the original input variables and the original output variables respectively; Max(x), Max(y) are the maximum values in the original input variable and the original output variable, respectively; step S2.3, using a sub-training set D ti training a CART decision tree, recursively dividing each region into two sub-regions and determining the output value on each sub-region within the input space where the training set is located, to construct a decision tree; Step S2.4, the optimal division feature j and the division point s are selected, for the regression problem, the CART decision tree adopts the mean square error to measure the advantages and disadvantages of the feature division point, for any division feature j, the corresponding any division point s divides the training set into two parts D1 and D2, the feature and the feature value division point corresponding to the minimum mean square error of D1 and D2 respectively are found; the mean square error of each division point of each feature is calculated: Wherein, c1 is the sample output value of D1 training set, D1 is one sub-region of D ti , D1(j,s) = {x|x (j) ≤s}; c2 is the sample output value of D2 training set, D2 is another sub-region of D ti , D2(j,s) = {x|x (j) >s}; x i represents the input variable or feature vector of a sample in the training data set, containing the values of multiple features or attributes; y i represents a value of a target variable associated with x i ; It is easy to know that c1, c2 that make the square error in each region minimum is the mean value of y in the corresponding region, so the above formula is modified as, Wherein, d1 is the number of samples in D1; d2 is the number of samples in D2; Step S2.5, each feature and each corresponding division point form a (j, s) pair, each (j, s) pair is substituted into the formula for calculation, and the (j, s) pair that makes the formula value minimum is selected; Step S2.6, the selected (j, s) is used to divide the region and determine the corresponding output value Wherein, D1 (j, s) = {x | x (j) ≤ s}; D2(j,s) = {x I x (j) > s}; Step S2.7, the steps S2.5 to S2.6 are repeatedly called for the two sub-regions until the set division stop condition is met. Step S2.8, divide the input space into M regions D1, D2,..., DM i ,...,D M , generate a CART decision tree: Wherein I is an indicator function, Step S2.9, repeat the process of steps S2.2 to S2.8, respectively, using D t1 , t2 ,...,D ti ,...,D tn Sub-training set training decision tree, finally get n CART decision tree model M1, M2,..., M i ,...,M n ; Step S2.10, predicting with each decision tree to obtain n groups of prediction results c1, c2,... c i ,...,c n ; Step S2.11, the prediction results of the n decision trees are averaged by using an average method to obtain a final prediction result where c i is the i-th prediction result, and finally a surface roughness model RFR1 is obtained. Step S2.12, steps S2.1 to S2.11 are repeated to obtain the microhardness model RFR2 and the machining energy consumption RFR3.

4. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm of claim 1, wherein: The specific implementation of step S3 includes the following method: the surface roughness model RFR1, the microhardness model RFR2 and the machining energy consumption RFR3 obtained in step S2 are given the same weight of each model by using an artificial weighting method, and a comprehensive evaluation model F is constructed by weighted combination m .

5. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm of claim 1, wherein: The specific implementation of step S4 comprises the following steps: Step S4.1, build the comprehensive evaluation model F m The fitness value is calculated as an improved DBO algorithm fitness function, and the improved DBO algorithm is used to optimize the surface roughness, microhardness and machining energy consumption of GH4169, so as to obtain the corresponding milling process parameters at the optimal milling surface roughness, microhardness and machining energy consumption, thereby realizing the optimization of the milling process parameters. Step S4.2, process parameter optimization is carried out using improved DBO algorithm, set the total number of scarab individuals as N, the maximum number of iterations T max , initialize the scarab individual as X={x1, x2, x3, x4}, let the dimension x1 be the size of milling speed v c , the dimension x2 be the size of feed per tooth f z , the dimension x3 be the size of axial depth of cut a p , and the dimension x4 be the size of radial depth of cut a e ; set population variables for the four dimensions, and initialize the population using Piecewise chaotic mapping: Wherein, P is a chaos coefficient, and the value is 0.4; xj-1(i) represents the (j-1)th dimension chaotic initialization value of the ith population individual, wherein x i,j xj-1(i) represents the value of the (j-1)th dimension of the ith population individual; Lb 1 , Ub 1 denotes the milling speed v c minimum and maximum values of the constraint; Lb 2 , Ub 2 denotes the feed per tooth f z the minimum and maximum values of the constraint; Lb 3 , Ub 3 indicates the axial depth of cut a p minimum and maximum values of the constraint; Lb 4 , Ub 4 denotes the radial depth of cut a p the minimum and maximum values of the constraint; Step S4.3, put the initialized individual of the population of the dung beetle into the comprehensive evaluation model established in S3 to obtain the fitness Y = {y1, y2,..., y N} Wherein, Y represents the fitness value of the dung beetle group. y i represents a fitness value of a dung beetle individual; Step S4.4, the fitness of the individual is sorted, and the individual with better fitness is the rolling ball dung beetle, which performs rolling ball or dancing behavior, followed by the breeding dung beetle, the small dung beetle, and the stealing dung beetle, and the population proportion of the four types of dung beetles needs to be set according to specific problems; the rolling ball dung beetle is responsible for quickly moving the food to prevent it from being stolen by other dung beetles, the breeding dung beetle is responsible for laying eggs to breed the next generation, the small dung beetle is responsible for foraging, and the stealing dung beetle will steal the food of other dung beetles; Step S4.5, the position update of the rolling ball dung beetle is planned to process both obstacle-free and obstacle-containing cases; Step S4.6, the small scarab beetles that grow into adults will drill out from the ground, foraging, and need to establish the best foraging area to guide the scarab beetles to forage, and the best foraging area is dynamically changing, and the foraging area boundary selection strategy is as follows: Lb b = max(X b ×(1-R), Lb) Ub b = min(X b ×(1+R), Ub) Wherein, Lb b is the lower bound of the optimal foraging area; Ub b upper bound for the optimal foraging area; X b is the global best foraging position; R = 1 - t / T max , T max denotes the maximum number of iterations; Lb represents the lower bound of the optimization problem; Ub represents the upper bound of the optimization problem; The foraging behavior position update formula is as follows: x i (t+1) = x i (t) + C1 x i (t) - Lb b ) + C2 x i (t) - Ub b ) Wherein, C1 represents a random number subject to normal distribution, that is, C1 ~ N(0, 1); C2 represents a 1×D random vector belonging to (0, 1); An adaptive t-distributed variation is introduced in the small dung beetle, and a [0, 1] random number rand is introduced, and when rand > 0.5, the variation operation is performed x i (t+1) = X b + t(iter) · X b Wherein x i (t + 1) is the new position of the i-th Stink-Horn individual after performing adaptive t-distributed variation; t(iter) is the t-distribution adopted by the algorithm; ν is the degree of freedom parameter of the t-distribution; X b is the global best foraging position; Step S4.7, the female dung beetle will select a suitable egg-laying site to provide a safe environment for the offspring, and once the egg-laying area is determined, the female dung beetle will select the area to lay eggs; the egg-laying area boundary selection strategy formula of the breeding dung beetle is as follows: Lb * = max(X * ×(1-R), Lb) Ub * = min(X * ×(1+R), Ub) Wherein, Lb * lower boundary of the egg laying area; Ub * represents the upper bound of the egg laying area; X * represents the current local optimum position; Each female dung beetle produces only one egg in each iteration; and the boundary range of the egg-laying area is dynamically changed with the number of iterations, which is mainly determined by the R value, and the position of the egg ball is also dynamically changed; The breeding behavior position update formula is: B i (t+1) = X * + b1 x (B i (t) - Lb * ) + b2 x (B i (t) - Ub * ) Wherein, B i (t) is the position information of the ith egg ball at the tth iteration, which is actually an individual x i ; b1 and b2 represent two independent random vectors of size 1×D, and D represents the dimension of the optimization problem; the position of the egg ball is strictly limited within the egg-laying area; Step S4.8, Some of the dung beetle population will steal food from other dung beetles, X b For the best food source location, i.e., to compete for the best location of food, the position update formula for the thief dung beetle is as follows: x i (t+1) = X b + S x g x (|x i (t) - X * | + |x i (t) - X b |) Wherein, g is a random vector of size 1×D subject to normal distribution; S is a constant value.

6. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm according to claim 5, characterized in that: In step S4.5, the algorithm processes the obstacle-free case as follows: When there is no obstacle, the dung beetle uses the sun to navigate and moves in a given direction in the entire search space, the intensity of the light source will affect the motion trajectory of the dung beetle, and natural factors will cause the dung beetle to deviate from the original direction; therefore, during the rolling process, the position update formula of the rolling ball dung beetle is as follows: x i (t+1) = x i (t) + a x k x x i (t-1) + b x D x Δx = |x i (t)-X ω | Wherein, t represents the current iteration number of the algorithm; x i (t) denotes the position information of the i-th dung beetle at the t-th iteration; k is a constant, and its value range is (0, 0.2], representing a deflection coefficient, and it is set to 0.1; b is a constant value, and its value range is (0, 1), and it is set to 0.3; α is a natural coefficient, and its value is 1 or -1, -1 indicating deviation from the original direction, and 1 indicating no deviation; X ω is the global worst position; Δx simulates the change of light intensity; The global exploration stage position update strategy of the fish eagle optimization algorithm is introduced in the rolling ball dung beetle as follows: x i (t+1) = x i (t) + q x (G - H x i (t)) Wherein; q is a random number between 0 and 1; G is a better dung beetle at the current position; H is a random number in the set {1, 2}.

7. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm according to claim 5, characterized in that: In step S4.5, the algorithm has the processing method for the case of having obstacles: when the dung beetle encounters obstacles, the dung beetle will adopt the dancing behavior to reposition, select a new direction to update the position, adopt a tangent function to simulate the dancing behavior, and the dancing behavior position update is as follows: x i (t+1) = x i (t) + tan θ | x i (t) - x i (t - 1) | Wherein, θ is the deflection angle, taking values in the range [0, π], if θ = 0, π / 2, π then the position x is not updated i ; | x i (t) - x i (t - 1) | represents the difference between the position of the ith dung beetle at the tth iteration and the position at the t - 1th iteration.

8. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm of claim 1, wherein: The GH4169 milling process adopts a hard alloy cutter.

9. The GH4169 milling process parameter optimization method based on RFR-IDBO algorithm of claim 1, wherein: The GH4169 milling process is carried out under dry milling conditions.

Images