A GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials

Abstract

The present invention discloses a GABP network prediction method for the quality of ultrasonic vibration-assisted drilling of carbon fiber composite materials. First, the Hassin-Parker failure criterion of the material within the layer and the damage failure mechanism of the cohesive viscous layer are comprehensively considered to establish a finite element model for drilling of carbon fiber composite materials. Secondly, a BP neural network prediction model for axial force and delamination factor (i.e., a quantitative evaluation parameter for the drilling quality of carbon fiber composite materials) is constructed by using the orthogonal experimental design method and the training samples and prediction samples determined by the finite element simulation model of carbon fiber composite materials drilling. Finally, the prediction error of the output sample is used as the fitness function, and the BP neural network is optimized by a genetic algorithm to construct a GABP neural network prediction model for axial force and delamination factor. The present invention not only establishes an accurate finite element model for drilling of carbon fiber composite materials, provides computer assistance for carbon fiber composite material drilling research, and reduces experimental costs; but also establishes a GABP prediction model for the drilling quality of carbon fiber composite materials, improves the calculation efficiency of axial force and delamination factor, and provides a practical theoretical basis for the reasonable selection of carbon fiber composite material drilling process parameters.

Description

Technical Field

[0001] This invention relates to ultrasonic vibration-assisted drilling technology for carbon fiber composites, and is applied to predicting the axial force and delamination factor (i.e., quantitative evaluation parameters for carbon fiber composite drilling quality) generated during ultrasonic vibration-assisted drilling of carbon fiber composites. In particular, it relates to a GABP network-based prediction method for ultrasonic vibration-assisted drilling of carbon fiber composites. Background Art

[0002] Carbon fiber composites are widely used in the aerospace field due to their advantages such as high specific strength, large specific modulus, strong corrosion resistance and good fatigue resistance. Aircraft structural parts use carbon fiber composites and titanium alloys or aluminum alloys to form a laminated structure, which can greatly reduce the weight while achieving excellent comprehensive performance. During the assembly process, carbon fiber composite parts are usually connected to other parts with rivet holes and bolt holes, so the quality of the holes is particularly important. However, carbon fiber composites have disadvantages such as high brittleness, low interlaminar strength, poor thermal conductivity, weak impact resistance and anisotropy. During the drilling process, defects such as entry peeling, exit delamination, tearing, burrs, poor processing quality and tool wear are easily generated. These defects seriously restrict the assembly quality of the aircraft, and thus affect the application effect and prospects of carbon fiber composites in the aerospace field.

[0003] KOENIG et al. found that excessive axial force is the main cause of burrs, tearing and delamination defects in the drilling process of carbon fiber composite materials. Zhang Houjiang et al. believe that the lower the spindle speed, the greater the feed speed and the larger the drill diameter, the more serious the tearing damage at the exit of the carbon fiber composite hole. Zhang Minghui et al. found that the larger the diameter and top angle of the twist drill, the larger the delamination factor and the greater the delamination damage. PHADNIS et al. simulated the effect of ultrasonic vibration-assisted processing on the drilling of carbon fiber composite materials. Wang Weibin compared axial vibration drilling with ordinary drilling simulation and believed that vibration processing is better than traditional processing, which can reduce drilling force and improve the quality of carbon fiber composite holes. The above work mainly uses finite element simulation technology to study the influence of ordinary drilling or ultrasonic vibration-assisted drilling process parameters on drilling force. The calculation amount is large and the operating efficiency is low. Especially when considering multiple drilling process parameters at the same time, it is difficult to reveal the relationship between drilling parameters and drilling force and delamination factor using only finite element method. Summary of the Invention

[0004] Based on the above problems, the present invention provides a GABP network prediction method for the quality of ultrasonic vibration-assisted drilling of carbon fiber composite materials, which can effectively predict the axial force and delamination factor of ultrasonic vibration-assisted drilling of carbon fiber composite materials, improve processing efficiency, and provide a basis for the selection of processing parameters.

[0005] The object of the present invention is achieved by: A GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials, the steps of which are as follows:

[0006] Step 1: Establish a finite element model of carbon fiber composite material drilling. In ABAQUS, establish a solid model of the carbon fiber composite material workpiece, and import the twist drill model into ABAQUS, divide the workpiece and drill bit mesh, define the boundary conditions and contact properties, and the model is as follows: Figure 1 shown.

[0007] Step 2: Establish the intra-layer failure model of carbon fiber composite materials. Use the user-defined VUMAT subroutine interface provided by finite element ABAQUS to write the intra-layer material damage subroutine "In-layer Damage". The specific steps are as follows:

[0008] S01: The constitutive stress-strain relationship of carbon fiber composite materials is constructed through the stiffness matrix of generalized Hooke's law, and its expression is:

[0009]

[0010]

[0011] Where, σ ij is stress, ε ij is strain, E i is the elastic modulus, G ij is the shear modulus, ν ij is Poisson's ratio;

[0012] Δ=1-ν 12 ν 21 -ν 23 ν 32 -ν 13 ν 31 -2ν 21 ν 13 ν 32 .

[0013] S02: Establish the Hassin-Parker failure criterion for carbon fiber composite materials. That is, the Hassin criterion is used to judge fiber failure, and the Park criterion is used to judge matrix failure. The expression is as follows:

[0014] (1) Tensile failure of carbon fiber composite materials (σ 11 ≥0)

[0015]

[0016] (2) Fiber compression failure of carbon fiber composite materials (σ 11 <0)

[0017]

[0018] (3) Tensile failure of carbon fiber composite matrix (σ 22 +σ 33 ≥0)

[0019]

[0020] (4) Compression failure of carbon fiber composite matrix (σ 22 +σ 33 <0)

[0021]

[0022] Where, σ ij is each stress component, S ij is the shear strength, X 1t 、X 1c are the tensile and compressive strengths in direction 1, X 2t 、X 2c are the tensile and compressive strengths in direction 2, respectively.

[0023] S03: When a composite material is damaged, its bearing capacity changes and its stiffness degrades. As the damage increases, the stiffness gradually decreases to zero, eventually leading to material failure. Assuming that the material is linear elastic before failure, after damage occurs, the stiffness matrix C degenerates into Its expression is:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] Step 3: Establish an interlaminar failure model of carbon fiber composite materials.

[0031] The workpiece structure of carbon fiber composite material is a multi-layer composite structure formed by a single layer under certain conditions. There is a very thin resin layer between the layers, which is weaker than the fiber layer. Therefore, carbon fiber composite materials often produce delamination defects during processing. In order to make the simulation results closer to the actual processing, this paper constructs a cohesive unit viscous layer between the layers and adopts a double-line constitutive model, such as Figure 2 shown.

[0032] S04: When ε<ε 0 When , it is the linear elastic stage. If the stiffness is K, the constitutive relation is:

[0033]

[0034] Where n, s, and t are the normal and two tangential directions of the viscous layer; σ n is the normal stress, σ s , σ t is the tangential shear stress, ε n is the normal strain, ε s , ε t is the tangential shear strain.

[0035] S05: When ε 0 ≤ε<ε max When , it is the stiffness degradation stage, the material is damaged, the stiffness is (1-D)K, and the constitutive relationship is

[0036]

[0037] Where D is the damage coefficient, and

[0038]

[0039] in

[0040] S06: When ε≥ε max When the material fails, the stiffness degenerates to 0, and the interface layer is damaged and delaminated.

[0041] S07: Use the quadratic nominal stress criterion as the initial damage criterion to determine whether damage occurs. The expression is:

[0042]

[0043] Where, is the interlaminar tensile strength, is the interlaminar shear strength.

[0044] S08: Use the energy BK hybrid criterion as the final damage criterion to determine whether failure occurs. The expression is:

[0045]

[0046]

[0047] Where, is the positive complete destruction fracture energy, is the tangential complete destruction fracture energy.

[0048] Step 4: Carbon fiber composite material drilling BP neural network prediction. BP neural network consists of three parts: input layer, hidden layer and output layer. Its structure is as follows Figure 3 As shown in Figure 2, the input layer has m = 4 neurons: speed x1, feed rate x2, frequency x3, amplitude x4, the output layer has n = 2 neurons, axial force a1 and stratification factor a2, and the hidden layer has u neurons.

[0049] Step 5: Train the network.

[0050] Assuming that the total number of samples is Z, the input vector X(z), hidden layer output vector H(z), output layer network output vector A(z) and actual output vector T(z) of the zth sample can be expressed as

[0051]

[0052] S09: Initialize the network. Randomly generate weights from [-1,1]. and threshold Where N = 1, j = 1, 2, ..., u, k = 1, 2, ..., n, which are the first-generation variable values. And input the normalized input sample X and the actual output sample T.

[0053] S10: Calculate the output of each layer

[0054]

[0055] Among them, f=tansig and g=purelin are the transfer functions between the hidden layer and the input and output layers respectively.

[0056] S11: Calculation error

[0057]

[0058]

[0059] S12: Modify weights and thresholds

[0060]

[0061]

[0062]

[0063]

[0064] S13: Determine the termination condition

[0065] e≤e0 (27)

[0066] N≥N max(28)

[0067] If equation (27) or equation (28) is satisfied, the algorithm ends; otherwise, set N=N+1 and return to S10.

[0068] Step 6: Optimize the network model. The initial weights and thresholds are crucial for whether the training process can reach a local minimum or converge. Therefore, a genetic algorithm is used to optimize the neural network. By finding the weights and thresholds that minimize the error, the random initial weights and thresholds of the BP neural network are replaced. The network is then trained for prediction.

[0069] S14: The optimization model of initial weights and thresholds can be expressed as:

[0070]

[0071] S15: Initialize the population. Using real number coding, randomly generate an initial population Θ consisting of P individuals y (1≤y≤P), individuals that do not meet the constraints in the optimization model should be eliminated.

[0072] S16: Define the fitness function. The reciprocal absolute value of the difference between the network's predicted output and the actual output is used as the fitness value. The larger the fitness value, the higher the individual quality. It can be expressed as:

[0073]

[0074] S17: Perform selection operation. Using the roulette method, the probability of each individual being selected is p y for:

[0075]

[0076] S18: Perform crossover operation. Since individuals are coded with real numbers, the crossover operation method uses the real number crossover method. The t-th chromosome Θ t and the rth chromosome Θ r The crossover operation at position s is as follows:

[0077]

[0078] Where b is the crossover probability.

[0079] S19: Perform mutation operation. i To perform mutation, the operation steps are as follows:

[0080]

[0081] Where r is the mutation probability, g is the current number of iterations, and G is the maximum number of evolutions.

[0082] S20: Determine the termination condition.

[0083] g≥G (34)

[0084] If equation (34) is satisfied, the algorithm ends and the optimal initial weight and threshold are returned to step 5; otherwise, set g=g+1 and return to S16.

[0085] Step 7: Based on steps 1 to 3, a finite element model for carbon fiber composite drilling was established in ABAQUS software, and the network input and training samples were obtained through orthogonal experimental design. From steps 4 to 6, a genetic algorithm was used to establish a GABP prediction model for the axial force and delamination factor of carbon fiber composite drilling, providing a practical theoretical basis for the rational selection of carbon fiber composite drilling process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is the drilling finite element model;

[0087] Figure 2 It is a two-line constitutive model of the cohesive viscous layer;

[0088] Figure 3 It is a neural network structure diagram. DETAILED DESCRIPTION

[0089] The present invention will be further described below with reference to the accompanying drawings and examples. Figures 1 to 3 A GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials is proposed. The specific steps are as follows:

[0090] Step 1: Establish a finite element model of carbon fiber composite material drilling. Figure 1 As shown in the figure, a model of a 16 mm × 16 mm × 2 mm carbon fiber composite workpiece is established in ABAQUS, and an 8 mm diameter twist drill model is imported into ABAQUS, the workpiece and drill bit meshes are divided, and the boundary conditions and contact properties are defined.

[0091] Step 2: Establish the intra-layer failure model of carbon fiber composite materials. Use the user-defined VUMAT subroutine interface provided by finite element ABAQUS to write the intra-layer material damage subroutine "In-layerDamage". The specific steps are as follows:

[0092] S01: The constitutive stress-strain relationship of carbon fiber composite materials is constructed through the stiffness matrix of generalized Hooke's law, and its expression is:

[0093]

[0094]

[0095] Where, σ ij is stress, ε ij is strain, E i is the elastic modulus, G ij is the shear modulus, ν ij is Poisson's ratio;

[0096] Δ=1-ν 12 ν 21 -ν 23 ν 32 -ν 13 ν 31 -2ν 21 ν 13 ν 32 .

[0097] S02: Establish the Hassin-Parker failure criterion for carbon fiber composite materials. That is, the Hassin criterion is used to judge fiber failure, and the Park criterion is used to judge matrix failure. The expression is as follows:

[0098] (1) Tensile failure of carbon fiber composite materials (σ 11 ≥0)

[0099]

[0100] (2) Fiber compression failure of carbon fiber composite materials (σ 11 <0)

[0101]

[0102] (3) Tensile failure of carbon fiber composite matrix (σ 22 +σ 33 ≥0)

[0103]

[0104] (4) Compression failure of carbon fiber composite matrix (σ 22 +σ 33 <0)

[0105]

[0106] Where, σ ij is each stress component, S ij is the shear strength, X 1t 、X 1c are the tensile and compressive strengths in direction 1, X 2t 、X 2c are the tensile and compressive strengths in direction 2, respectively.

[0107] This paper uses carbon fiber composite material model M21 / T700, and its material parameters are shown in Table (1,2).

[0108] Table 1 M21 / T700 performance parameters

[0109]

[0110] Table 2 Damage failure parameters of M21 / T700 CFRP (MPa)

[0111]

[0112] S03: When a composite material is damaged, its bearing capacity changes and its stiffness degrades. As the damage increases, the stiffness gradually decreases to zero, eventually leading to material failure. Assuming that the material is linear elastic before failure, after damage occurs, the stiffness matrix C degenerates into Its expression is:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] Step 3: Establish an interlaminar failure model of carbon fiber composite materials.

[0120] The workpiece structure of carbon fiber composite material is a multi-layer composite structure formed by a single layer under certain conditions. There is a very thin resin layer between the layers, which is weaker than the fiber layer. Therefore, carbon fiber composite materials often produce delamination defects during processing. In order to make the simulation results closer to the actual processing, this paper constructs a cohesive unit viscous layer between the layers and adopts a double-line constitutive model, such as Figure 2 shown.

[0121] S04: When ε<ε 0 When , it is the linear elastic stage. If the stiffness is K, the constitutive relation is:

[0122]

[0123] Where n, s, and t are the normal and two tangential directions of the viscous layer; σ n is the normal stress, σ s , σ t is the tangential shear stress, ε n is the normal strain, ε s , εt is the tangential shear strain.

[0124] S05: When ε 0 ≤ε<ε max When , it is the stiffness degradation stage, the material is damaged, the stiffness is (1-D)K, and the constitutive relationship is

[0125]

[0126] Where D is the damage coefficient, and

[0127]

[0128] in

[0129] S06: When ε≥ε max When the material fails, the stiffness degenerates to 0, and the interface layer is damaged and delaminated.

[0130] S07: Use the quadratic nominal stress criterion as the initial damage criterion to determine whether damage occurs. The expression is:

[0131]

[0132] Where, is the interlaminar tensile strength, is the interlaminar shear strength.

[0133] S08: Use the energy BK hybrid criterion as the final damage criterion to determine whether failure occurs. The expression is:

[0134]

[0135]

[0136] Where, is the positive complete destruction fracture energy, is the tangential complete destruction fracture energy.

[0137] The viscosity layer parameters of carbon fiber composite material M21 / T700 are shown in Table 3.

[0138] Table 3 Viscous layer parameters

[0139]

[0140] Through the carbon fiber composite material drilling finite element model established from steps one to three, the axial force and delamination factor of the simulation output can be obtained. The orthogonal experimental design method is applied, and according to the value ranges of the drilling process parameters, which are 3000 rpm ≤ x1 ≤ 9000 rpm, 355 mm / min ≤ x2 ≤ 685 mm / min, 20 kHz ≤ x3 ≤ 35 kHz, and 1 micron ≤ x4 ≤ 4 microns, 64 sets of training data and 16 sets of test data required for the neural network prediction model are obtained.

[0141] Step 4: Carbon fiber composite material drilling BP neural network prediction. BP neural network consists of three parts: input layer, hidden layer and output layer. Its structure is as follows Figure 3 As shown in Figure 2, the input layer has m = 4 neurons: speed x1, feed rate x2, frequency x3, amplitude x4, the output layer has n = 2 neurons, axial force a1 and stratification factor a2, and the hidden layer has u neurons.

[0142] Step 5: Train the network.

[0143] Assuming that the total number of samples is Z, the input vector X(z), hidden layer output vector H(z), output layer network output vector A(z) and actual output vector T(z) of the zth sample can be expressed as

[0144]

[0145] S09: Initialize the network. Randomly generate weights from [-1,1]. and threshold Where N = 1, j = 1, 2, ..., u, k = 1, 2, ..., n, which are the first-generation variable values. And input the normalized input sample X and the actual output sample T.

[0146] S10: Calculate the output of each layer

[0147]

[0148] Among them, f=tansig and g=purelin are the transfer functions between the hidden layer and the input and output layers respectively.

[0149] S11: Calculation error

[0150]

[0151]

[0152] S12: Modify weights and thresholds

[0153]

[0154]

[0155]

[0156]

[0157] S13: Determine the termination condition

[0158] e≤e0 (27)

[0159] N≥N max (28)

[0160] If equation (27) or equation (28) is satisfied, the algorithm ends; otherwise, set N=N+1 and return to S10.

[0161] Given the maximum number of iterations N of BP neural network parameters max =5000, target error e0=1×10 -5 , adjustment coefficient η = 1.05, learning rate α = 0.05, training to obtain the network prediction model of axial force and delamination factor, which is A = sim(net, X)

[0162] Step 6: Optimize the network model. The initial weights and thresholds are crucial for whether the training process can reach a local minimum or converge. Therefore, a genetic algorithm is used to optimize the neural network. By finding the weights and thresholds that minimize the error, the random initial weights and thresholds of the BP neural network are replaced. The network is then trained for prediction.

[0163] S14: The optimization model of initial weights and thresholds can be expressed as:

[0164]

[0165] S15: Initialize the population. Using real number coding, randomly generate an initial population Θ consisting of P individuals y (1≤y≤P), individuals that do not meet the constraints in the optimization model should be eliminated.

[0166] S16: Define the fitness function. The reciprocal absolute value of the difference between the network's predicted output and the actual output is used as the fitness value. The larger the fitness value, the higher the individual quality. It can be expressed as:

[0167]

[0168] S17: Perform selection operation. Using the roulette method, the probability of each individual being selected is p y for:

[0169]

[0170] S18: Perform crossover operation. Since individuals are coded with real numbers, the crossover operation method uses the real number crossover method. The t-th chromosome Θ t and the rth chromosome Θ r The crossover operation at position s is as follows:

[0171]

[0172] Where b is the crossover probability.

[0173] S19: Perform mutation operation. To perform mutation, the operation steps are as follows:

[0174]

[0175] Where r is the mutation probability, g is the current number of iterations, and G is the maximum number of evolutions.

[0176] S20: Determine the termination condition.

[0177] g≥G (34)

[0178] If equation (34) is satisfied, the algorithm ends and the optimal initial weight and threshold are returned to step 5; otherwise, set g=g+1 and return to S16.

[0179] Set the population size to P = 30, the number of evolutions to G = 500, the crossover probability to b = 0.5, and the mutation probability to r = 0.1, and perform training. The GABP neural network prediction model can be obtained as A = sim(opt_net, X).

[0180] The above-described embodiments are only preferred embodiments of the present invention and are not exhaustive of all feasible implementations of the present invention. For those skilled in the art, any obvious modifications made thereto without departing from the principles and spirit of the present invention should be considered to be included within the scope of protection of the claims of the present invention.

Claims

1. A GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials, characterized by: The prediction method includes the following steps: Step 1, establishing a finite element model of carbon fiber composite material drilling: In ABAQUS, a solid model of the carbon fiber composite material workpiece is established, and the twist drill model is imported into ABAQUS, the workpiece and the drill bit mesh are divided, and the boundary conditions and contact properties are defined; Step 2, establishing a carbon fiber composite material intra-layer failure model: Using the user-defined VUMAT subroutine interface provided by the finite element ABAQUS, the intra-layer material damage subroutine "In-layerDamage" is written; Step 3, establishing a carbon fiber composite material inter-layer failure model: The carbon fiber composite material workpiece structure is a multi-layer composite structure formed by a single-layer plate under certain conditions. There is a very thin resin layer between the layers, and its strength is weaker than the fiber layer. Therefore, carbon fiber composite materials often produce delamination defects during processing. In order to make the simulation results closer to actual processing, a cohesive unit viscosity layer is constructed between the layers, and a double-line constitutive model is adopted; Step 4. Carbon fiber composite material drilling BP neural network prediction: The BP neural network consists of three parts: input layer, hidden layer and output layer. The input layer has m = 4 neurons: speed x1, feed speed x2, frequency x3, amplitude x4, the output layer has n = 2 neurons, axial force a1 and stratification factor a2 respectively, and the hidden layer is u neurons; Step 5. Training network: Assuming that the total number of samples is Z, the input vector X(z), hidden layer output vector H(z), output layer network output vector A(z) and actual output vector T(z) of the zth sample can be expressed as Step 6. Optimize the network model: The initial weights and thresholds have a great influence on whether the training process can reach the local minimum or whether the training can converge. For this reason, a genetic algorithm is used to optimize the neural network. The random initial weights and thresholds of the BP neural network are replaced by the weights and thresholds with the smallest error, and then the network is trained for prediction. Step 7. According to steps 1 to 3, a finite element model of carbon fiber composite material drilling is established in ABAQUS software, and the input samples and training samples of the network are obtained through the orthogonal experimental design method. In steps 4 to 6, a genetic algorithm is used to establish a GABP prediction model for the axial force and delamination factor of carbon fiber composite material drilling, which provides a practical theoretical basis for the reasonable selection of carbon fiber composite material drilling process parameters.

2. The GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials according to claim 1 is characterized in that: The establishment of the carbon fiber composite material intra-layer failure model includes the following steps: Sa1: Carbon fiber composite material is an orthotropic material, and its constitutive stress-strain relationship is constructed by the stiffness matrix of the generalized Hooke's law, and the relationship is: ； ； Where, σ ij is stress, ε ij is strain, E i is the elastic modulus, G ij is the shear modulus, ν ij is Poisson's ratio, Δ=1-ν 12 ν 21 -ν 23 ν 32 -ν 13 ν 31 -2ν 21 ν 13 ν 32 Sa2: Carbon fiber composite materials have a certain toughness. The failure principles of fibers under tension and pressure are different. Excessive load will also cause matrix failure. Therefore, the Hassin-Parker failure criterion of carbon fiber composite materials is established, that is, the Hassin criterion is used to judge fiber failure, and the Park criterion is used to judge matrix failure. The expression is as follows: (1) Carbon fiber composite material fiber tensile failure, σ 11 ≥0; ； (2) Fiber compression failure of carbon fiber composite materials, σ 11 <0; ； (3) Tensile failure of carbon fiber composite matrix, σ 22 +σ 33 ≥0; ； (4) Compression failure of carbon fiber composite matrix, σ 22 +σ 33 <0; ； Where, σ ij is each stress component, S ij is the shear strength, X 1t 、X 1c are the tensile and compressive strengths in direction 1, X 2t 、X 2c are the tensile and compressive strengths in direction 2, respectively; Sa3: When damage occurs to the composite material, the material's bearing capacity changes and the stiffness degrades. As the degree of damage increases, the stiffness gradually decreases to zero, eventually leading to material failure. Assuming that the material is linear elastic before failure, after damage occurs, the stiffness matrix C degenerates into Its expression is: ； ； ； ； ； 3. The GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials according to claim 1 is characterized in that: The establishment of the carbon fiber composite material interlaminar failure model comprises the following steps: Sb1, when ε < ε 0 When , it is the linear elastic stage. If the stiffness is K, the constitutive relation is: ； Where n, s, and t are the normal and two tangential directions of the viscous layer; σ n is the normal stress, σ s , σ t is the tangential shear stress, ε n is the normal strain, ε s , ε t is the tangential shear strain; Sb2, when ε 0 ≤ε<ε max When , it is the stiffness degradation stage, the material is damaged, the stiffness is (1-D)K, and the constitutive relationship is ； Where D is the damage coefficient, and ; in Sb3, when ε≥ε max When , the material fails, the stiffness degenerates to 0, and the interface layer is damaged and delaminated; Sb4, using the quadratic nominal stress criterion as the initial damage criterion to determine whether damage occurs, its expression is: ； Where, is the interlaminar tensile strength, is the interlaminar shear strength; Sb5, the energy BK mixed criterion is used as the final damage criterion to determine whether failure occurs, and its expression is: ； ； Where, is the positive complete destruction fracture energy, is the tangential complete destruction fracture energy.

4. The GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials according to claim 1 is characterized in that: The training network includes the following steps: Sc1, initializing the network: randomly generating weights from [-1, 1] and threshold Where N = 1, j = 1, 2, ..., u, k = 1, 2, ..., n, which are the first-generation variable values, and the normalized input sample X and the actual output sample T are input; Sc2, calculate the output of each layer: ； Where f = tansig and g = purelin are the transfer functions between the hidden layer and the input and output layers respectively; Sc3, calculation error: ； Sc4. Modify weights and thresholds: ； ； ； Sc5. Determine the termination condition: e≤e0 N≥N max If the termination condition is satisfied, the algorithm ends; otherwise, set N=N+1 and return to Sc2.

5. The GABP network prediction method for ultrasonic vibration-assisted drilling quality of carbon fiber composite materials according to claim 1 is characterized in that: The optimization network model comprises the following steps: The optimization model of Sd1, initial weight and threshold is expressed as: Sd2, Initialization population: Use real number coding to randomly generate an initial population consisting of P individuals y , 1≤y≤P, individuals that do not meet the constraints in the optimization model should be eliminated; Sd3. Define the fitness function: take the reciprocal absolute value of the difference between the network predicted output and the actual output as the fitness value. The larger the fitness value, the higher the individual quality, which can be expressed as: ; Sd4, perform selection operation: using the roulette method, the probability of each individual being selected is p y for: Sd5, perform crossover operation: because individuals are coded with real numbers, the crossover operation method uses the real number crossover method, the tth chromosome t and the rth chromosome r The crossover operation at position s is as follows: ； Where b is the crossover probability; Sd6, perform mutation operation: for the i-th gene of the j-th individual To perform mutation, the operation steps are as follows: ； Where r is the mutation probability, g is the current number of iterations, and G is the maximum number of evolutions; Sd7, determine the termination condition: g≥G. If the termination condition is met, end the algorithm and return to the optimal initial weight and threshold to step 5; otherwise, set g=g+1 and return to Sd3.

Images