Drilling force calculation method based on unit tool linear synthesis method and transfer learning

Abstract

The present invention discloses a method for calculating drilling force based on the unit tool linear synthesis method and transfer learning. The method first uses the unit tool linear synthesis method to divide the drilling tool into multiple unit tools. By calculating the cutting force on each unit tool and integrating them, the cutting force on the entire drilling edge is obtained. On this basis, the existing drilling force calculation neural network model is transferred to the new drilling task using transfer learning technology. By fine-tuning the model weights, the drilling force can be quickly and accurately calculated. This method combines the accuracy of the unit tool linear synthesis method with the efficiency of transfer learning. It can fully consider the influence of multiple factors such as the geometry of the twist drill and the workpiece material, and improve the accuracy and reliability of the drilling force calculation. This method has broad application prospects in mechanical processing and can be used to optimize cutting parameters, improve processing efficiency and workpiece quality, and provide effective reference and guidance for processing personnel.

Description

Technical Field

[0001] The present invention relates to the field of mechanical processing technology, and specifically to a method for calculating drilling force in drilling processing. The method combines a unit tool linear synthesis method with transfer learning technology to improve the calculation accuracy and efficiency of the drilling force. Background Art

[0002] In drilling operations, drilling force has a significant impact on both workpiece quality and drill life. Research on methods for calculating (predicting) drilling force not only contributes to a deeper understanding of the mechanical properties of the drilling process but also provides a theoretical basis for optimizing machining parameters, improving machining efficiency, and reducing costs. Currently, the main methods for calculating drilling force include theoretical analysis, empirical formulas, finite element methods, and neural networks. First, the theoretical analysis method decomposes the drill bit's cutting edge into multiple tiny cutting units (microelements) to calculate the drilling force. The cutting force of each unit is calculated using a simple microelement cutting force calculation model. The cutting forces of all units are then summed to obtain the drilling force of the entire twist drill. This method ignores nonlinear effects in the drilling process (such as the influence of drilling thermal effects, tool wear, and built-up edge on the drilling force). Accurate drilling force calculations are only possible when a very precise microelement cutting force calculation model is used. Second, the empirical formula method uses experimental data through regression analysis to establish a mathematical relationship between drilling force, drilling parameters, and drill bit structural parameters. However, this method suffers from poor accuracy and extrapolation. ③ The finite element method uses numerical methods to simulate the interaction between the material and the drill bit during the drilling process, and then calculates the drilling force of the twist drill. This method is computationally intensive and time-consuming. The accuracy of the drilling force is greatly affected by factors such as the selection of material constitutive model parameters and meshing accuracy. It is difficult to handle complex working conditions and requires a large amount of experimental data for verification and calibration. ④ The neural network method uses a large amount of experimental data to automatically learn the relationship between input variables (such as drilling amount and drill bit structural parameters) and drilling force. This method is highly dependent on data and requires a large amount of high-quality experimental data to train the model. When data is insufficient or feature selection is inappropriate, overfitting problems are prone to occur, resulting in poor model generalization ability.

[0003] In summary, most of the existing methods for calculating drilling force require a large amount of experimental data, and the calculation accuracy and efficiency of drilling force need to be further improved. Summary of the Invention

[0004] The purpose of the present invention is to provide a drilling force calculation method based on the unit tool linear synthesis method and transfer learning, so as to overcome the problems of the existing methods in calculating the drilling force, such as the large amount of experimental data required and low precision. The method decomposes the drill bit cutting edge into multiple unit tools, calculates the cutting force of each unit tool using a neural network model, and combines the linear synthesis method and transfer learning technology to realize the accurate calculation of the drilling force.

[0005] To achieve the above content, the present invention proposes the following technical solutions.

[0006] The drilling force calculation method based on the unit tool linear synthesis method and transfer learning includes the following steps:

[0007] S1: Unit tool division

[0008] The two main cutting edges and the chisel edge of the drill bit are divided into a series of unit tools connected end to end, each of which cuts with different rake angles, edge inclination angles and cutting amounts;

[0009] S2: Neural network model for right-angle cutting force calculation

[0010] A series of right-angle cutting force theoretical calculation results are obtained using the unequal shear zone model as the source domain data set to train the initial neural network model for right-angle cutting force calculation.

[0011] A target domain dataset was obtained through a small number of right-angle cutting experiments, and a neural network model for right-angle cutting force calculation was trained using transfer learning.

[0012] S3: Neural network model for calculating oblique cutting forces

[0013] The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and convert it into the oblique cutting force.

[0014] The bevel cutting force data obtained from a small number of bevel cutting experiments are used as the target domain dataset, and the final neural network model for bevel cutting force calculation is trained using the transfer learning method.

[0015] S4: Linear synthesis method

[0016] The cutting forces of each unit tool are accumulated using the linear synthesis method to obtain the theoretical calculation results of the drilling force of the entire twist drill;

[0017] S5: Transfer Learning

[0018] A series of theoretical calculation results of drilling forces using twist drills of different diameters and drilling loads were obtained and used as the source domain dataset to train an initial neural network model for drilling force calculation.

[0019] Conduct a small number of drilling experiments to obtain the measured data of drilling force experiments and use it as the target domain dataset;

[0020] Using the transfer learning method, part of the structure, weights or parameters of the initial drilling force calculation neural network model are transferred to the final drilling force calculation neural network. The weights of the first few layers of the neural network are frozen, and a small amount of actual drilling force experimental data is used for secondary training to obtain the final drilling force calculation neural network.

[0021] Furthermore, each cutting edge of the drill bit is divided into several unit tools connected end to end, specifically including:

[0022] Divide the drill bit radius into n equal parts, and get n radii ρ1, ρ2, ρ3, ... ρ arranged from large to small i With the drill axis as the axis and the radius of each equally divided cylinder as the radius, draw n coaxial cylinders. The tool part located on the same cutting edge of the drill bit between two adjacent cylindrical surfaces is defined as a unit tool. That is, the unit tool on the same cutting edge between the i-th cylindrical surface and the i+1-th cylindrical surface is defined as the i-th unit tool, and the corresponding cutting edge is P i P i+1 ;

[0023] Use formula (1) to calculate the position radius ρ of the unit tool i :

[0024] ρ i =x i i+y i j+z i k#(1)

[0025] Among them, x i ,y i , z i (i=1,2,……,n) is the dividing point (P i P i+1 The coordinates of the midpoint); i, j, k represent the unit vectors in the x, y, and z directions respectively;

[0026] Calculate the orientation b of the unit tool using formula (2) i :

[0027]

[0028] The synthetic cutting velocity vector V of the unit tool is calculated using formula (3): all :

[0029] V all =V c +V f =ρ i ×2πNj+Nf / 1000j#(3)

[0030] Among them, V cis the main motion linear velocity of the unit tool, V f is the feed motion of the unit tool, f is the feed per revolution of the drill (mm / r), and N is the spindle speed (r / min);

[0031] Formula (4) is used to calculate the unit normal vector of the working base surface r, working cutting plane s and working main section o of the unit tool:

[0032]

[0033] Calculate the working edge inclination angle λ of the unit tool using formula (5) se , working front angle γ oe and working principal deflection angle k re :

[0034]

[0035] Among them, f' is the unit vector of the projection direction of the unit vector f in the feed direction on the working base surface,

[0036] The cutting amount of the unit tool (cutting width a) is calculated using formula (6) c and cutting thickness a w ):

[0037]

[0038] Where z is the number of teeth, for a drill, z = 2; u is the angle between the synthetic cutting speed of the unit tool and the main motion linear velocity, sinu = j·r

[0039] Furthermore, the normal rake angle α of the unit tool at the chisel edge of the twist drill is calculated using formula (7): n :

[0040]

[0041] in, is the cutting edge angle, and ψ is the chisel edge bevel angle.

[0042] Furthermore, a neural network model for calculating right-angle cutting force is established, which specifically includes:

[0043] (1) Using the unequal shear zone model, a series of theoretical calculation results of right-angle cutting forces under different rake angles and cutting parameters are obtained as the source domain data set, and an initial neural network model for right-angle cutting force calculation is trained;

[0044] (2) Conduct a small number of right-angle cutting experiments to obtain a target domain dataset containing experimental measured data on right-angle cutting forces;

[0045] (3) Using the transfer learning method, part of the structure, weights or parameters of the initial right-angle cutting force calculation neural network model are transferred to the final right-angle cutting force calculation neural network, and the weights of the first few layers of the final neural network are frozen. A small amount of right-angle cutting force experimental measured data is used for secondary training to obtain the final right-angle cutting force calculation neural network.

[0046] Furthermore, the final neural network model for calculating the bevel cutting force is established, which specifically includes:

[0047] (1) The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and transform it into the oblique cutting force.

[0048] (2) A series of theoretical calculation results of oblique cutting forces under different cutting angles and cutting parameters based on the equivalent plane method and the right-angle cutting force calculation neural network are used as the source domain data set;

[0049] (3) Conduct a small number of oblique cutting experiments to obtain experimental measured data of oblique cutting forces as the target domain dataset;

[0050] (4) Using the transfer learning method, part of the structure, weights or parameters of the initial bevel cutting force calculation neural network model are transferred to the final bevel cutting force calculation neural network, the weights of the first few layers of the final neural network are frozen, and a small amount of experimental measured data of the bevel cutting force is used for secondary training to obtain the final bevel cutting force calculation neural network.

[0051] Furthermore, the equivalent plane method is used to transform the bevel cutting condition into an equivalent right-angle cutting condition, specifically including:

[0052] The equivalent plane P is defined according to the chip flow direction u' of the oblique cutting and the normal direction n' of the rake face equ ;

[0053] In the equivalent plane, use formula (8) to calculate the equivalent rake angle α e :

[0054] tanα n =tanα e ·cosη#(8)

[0055] Among them, α n is the normal to front angle, η is the chip angle;

[0056] Establish a rectangular coordinate system xyz, where x is the unit vector of the main cutting motion direction and y is the unit vector of the feed motion direction. Rotate the rectangular coordinate system xyz around the x-axis by λ0 (edge ​​inclination angle) and then rotate it around the transformed y-axis by α n(normal forward angle), and finally rotate η (chip angle) around the transformed x-axis to obtain the coordinate system x'y'z', where x', y', and z' are unit vectors in the x', y', and z' directions;

[0057] In the equivalent plane, use formula (9) to calculate the equivalent feed f equ :

[0058] f equ =f e / (D fe ·z')#(9)

[0059] Among them, f e is the feed rate in the bevel cutting condition, D fe is the unit vector in the equivalent feed direction,

[0060] In the equivalent plane, the equivalent cutting speed V is calculated using formula (10): equ :

[0061] V equ =V e ·(D fe ·z′)#(10)

[0062] Among them, V e is the cutting speed in the bevel cutting condition, D te is the unit vector in the direction of the equivalent cutting main motion,

[0063] In the equivalent plane, the equivalent cutting width w is calculated using formula (11): equ :

[0064] w equ =w e / (D te ·x')#(11)where w e is the cutting width in the bevel cutting condition;

[0065] The equivalent right-angle cutting force is converted into oblique-angle cutting force using formula (12):

[0066]

[0067] Among them, F t is the main cutting force under the equivalent right-angle cutting condition, F f is the feed resistance under equivalent right-angle cutting conditions; F r is the main cutting force under oblique cutting conditions, F s is the feed resistance under oblique cutting conditions, F o It is the cutting depth resistance under oblique cutting conditions.

[0068] Furthermore, the cutting forces of each unit tool are superimposed using the linear synthesis method, including:

[0069] Through coordinate transformation, the tangential force F of the i-th unit tool in the drill bit global coordinate system xyz is obtained using formula (13): eli , radial force F emi and axial force F eni :

[0070]

[0071] Among them, l i 、m i 、n i Represents the unit vectors of the tangential, radial and axial directions of the position of the i-th unit tool; r i 、s i 、o i Respectively represent the unit normal vector of the unit tool working base surface, the unit normal vector of the working cutting plane, and the unit normal vector of the working orthogonal plane; F eri 、F esi 、F eoi They represent the cutting forces in the main motion direction, feed motion direction, and cutting depth direction of the unit tool respectively.

[0072] The torque T of the unit tool is obtained using formula (14): ei :

[0073]

[0074] Among them, x i 、z i They represent the x and z coordinates of the i-th unit tool in the global coordinate system xyz;

[0075] Using formula (15), we can get the torque T and axial force F of the twist drill as a whole: N :

[0076]

[0077] Beneficial effects of the present invention:

[0078] 1. Compared with the drilling force obtained by empirical formula method or neural network method, the method proposed in this invention can obtain more accurate drilling force without a large amount of experimental data;

[0079] 2. Compared with obtaining drilling force through finite element method, the method proposed in this invention greatly reduces the calculation time;

[0080] 3. Compared with the theoretical analytical method for obtaining the drilling force, the method proposed in the present invention improves the calculation accuracy of the drilling force.

[0081] Figures in the specification

[0082] Figure 1 The drilling force calculation process is based on the unit tool linear synthesis method and transfer learning;

[0083] Figure 2 It is the global coordinate system of the twist drill and the local coordinate system of the unit tool;

[0084] Figure 3 Schematic diagram of the equivalent plane method. DETAILED DESCRIPTION

[0085] The scheme of the present invention is described in more detail and completely below.

[0086] Specific embodiment 1,

[0087] The drilling force calculation method based on the unit tool linear synthesis method and transfer learning includes the following steps:

[0088] S1: Unit tool decomposition

[0089] Decompose the cutting edge of the drill into a series of unit tools connected end to end, each cutting with different rake angles, edge inclination angles and cutting amounts;

[0090] S2: Final neural network model for right-angle cutting force calculation

[0091] A series of right-angle cutting force theoretical calculation results are obtained using the unequal shear zone model as the source domain data set to train the initial neural network model for right-angle cutting force calculation.

[0092] A target domain dataset is obtained through a small number of right-angle cutting force experiments, and the final neural network model for right-angle cutting force calculation is trained using the transfer learning method.

[0093] S3: Neural network model for calculating oblique cutting forces

[0094] The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and convert it into the oblique cutting force.

[0095] A target domain dataset is obtained through a small number of oblique cutting experiments, and the final neural network model for oblique cutting force calculation is trained using the transfer learning method.

[0096] S4: Linear synthesis method

[0097] The cutting forces of each unit tool are superimposed using the linear synthesis method to obtain the theoretical calculation results of the drilling force of the entire twist drill.

[0098] S5: Transfer Learning

[0099] A series of theoretical calculation results of drilling forces using twist drills of different diameters and drilling parameters are obtained as source domain data sets to train an initial neural network model for drilling force calculation.

[0100] Conduct a small number of drilling experiments and obtain drilling force experimental results as the target domain dataset;

[0101] Using the transfer learning method, part of the structure, weights or parameters of the initial drilling force calculation neural network model are transferred to the final drilling force calculation neural network. The weights of the first few layers of the final neural network are frozen, and secondary training is performed using a small amount of drilling force experimental results to obtain the final drilling force calculation neural network.

[0102] Furthermore, each cutting edge of the drill bit is divided into several unit tools connected end to end, specifically including:

[0103] Divide the drill bit radius into n equal parts, and get n radii ρ1, ρ2, ρ3, ... ρ arranged from large to small i With the drill axis as the axis and the radius of each equally divided cylinder as the radius, draw n coaxial cylinders. The tool part located on the same cutting edge of the drill bit between two adjacent cylindrical surfaces is defined as a unit tool. That is, the unit tool on the same cutting edge between the i-th cylindrical surface and the i+1-th cylindrical surface is defined as the i-th unit tool, and the corresponding cutting edge is P i P i+1 ;

[0104] Use formula (1) to calculate the position radius ρ of the unit tool i :

[0105] ρ i =x i i+y i j+z i k#(1)

[0106] Among them, x i ,y i , z i (i=1,2,……,n) is the dividing point (P i P i+1 The coordinates of the midpoint); i, j, k represent the unit vectors in the x, y, and z directions respectively;

[0107] Calculate the orientation b of the unit tool using formula (2) i :

[0108]

[0109] The synthetic cutting velocity vector V of the unit tool is calculated using formula (3): all :

[0110] V all =V c +V f =ρ i ×2πNj+Nf / 1000j#(3)

[0111] Among them, V c is the main motion linear velocity of the unit tool, V f is the feed motion of the unit tool, f is the feed per revolution of the drill (mm / r), and N is the spindle speed (r / min);

[0112] Formula (4) is used to calculate the unit normal vector of the working base surface r, working cutting plane s and working main section o of the unit tool:

[0113]

[0114] Calculate the working edge inclination angle λ of the unit tool using formula (5) se , working front angle γ oe and working principal deflection angle k re :

[0115]

[0116] Among them, f' is the unit vector of the projection direction of the unit vector f in the feed direction on the working base surface,

[0117] The cutting amount of the unit tool (cutting width a) is calculated using formula (6) c and cutting thickness a w ):

[0118]

[0119] Where z is the number of teeth, for a drill, z = 2; u is the angle between the synthetic cutting speed of the unit tool and the main motion linear velocity, sinu = j·r

[0120] Furthermore, the normal rake angle α of the unit tool at the chisel edge of the twist drill is calculated using formula (7): n :

[0121]

[0122] in, is the cutting edge angle, and ψ is the chisel edge bevel angle.

[0123] Furthermore, a neural network model for calculating right-angle cutting force is established, which specifically includes:

[0124] (1) Using the unequal shear zone model (reference Zhou Danyu. Improvement of the unequal shear zone model and its application in modeling the sudden change of chip angle in face turning [D / OL]. Huazhong University of Science and Technology, 2022.), a series of theoretical calculation results of right-angle cutting forces under different rake angles and cutting parameters are obtained as the source domain data set to train an initial neural network model for right-angle cutting force calculation;

[0125] (2) Conduct a small number of right-angle cutting experiments to obtain a target domain dataset containing the right-angle cutting force experimental results;

[0126] (3) Using the transfer learning method, part of the structure, weights or parameters of the initial right-angle cutting force calculation neural network model are transferred to the final right-angle cutting force calculation neural network, the weights of the first few layers of the final neural network are frozen, and a small amount of right-angle cutting force experimental results are used for secondary training to obtain the final right-angle cutting force calculation neural network.

[0127] Furthermore, a neural network model for calculating the bevel cutting force is established, which specifically includes:

[0128] (1) The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and transform it into the oblique cutting force.

[0129] (2) A series of theoretical calculation results of oblique cutting forces under different cutting angles and cutting parameters based on the equivalent plane method and the right-angle cutting force calculation neural network are used as the source domain data set;

[0130] (3) Conduct a small number of oblique cutting experiments to obtain the oblique cutting force experimental results as the target domain dataset;

[0131] (4) Using the transfer learning method, part of the structure, weights or parameters of the initial bevel cutting force calculation neural network model are transferred to the final bevel cutting force calculation neural network, and the weights of the first few layers of the final neural network are frozen. A small amount of bevel cutting force experimental results are used for secondary training to obtain the final bevel cutting force calculation neural network.

[0132] Furthermore, the equivalent plane method is used to transform the bevel cutting condition into an equivalent right-angle cutting condition, specifically including:

[0133] The equivalent plane P is defined according to the chip flow direction u' of the oblique cutting and the normal direction n' of the rake face equ ;

[0134] In the equivalent plane, use formula (8) to calculate the equivalent rake angle α e :

[0135] tanα n =tanα e ·cosη#(8)

[0136] Among them, α n is the normal to front angle, η is the chip angle;

[0137] Establish a rectangular coordinate system xyz, where x is the unit vector of the main cutting motion direction and y is the unit vector of the feed motion direction. Rotate the rectangular coordinate system xyz around the x-axis by λ0 (edge ​​inclination angle) and then rotate it around the transformed y-axis by α n (normal forward angle), and finally rotate η (chip angle) around the transformed x-axis to obtain the coordinate system x'y'z', where x', y', and z' are unit vectors in the x', y', and z' directions;

[0138] In the equivalent plane, use formula (9) to calculate the equivalent feed f equ :

[0139] f equ =f e / (D fe ·z')#(9)

[0140] Among them, f e is the feed rate in the bevel cutting condition, D fe is the unit vector in the equivalent feed direction,

[0141] In the equivalent plane, the equivalent cutting speed V is calculated using formula (10): equ :

[0142] V equ =V e ·(D fe ·z′)#(10)

[0143] Among them, V e is the cutting speed in the bevel cutting condition, D te is the unit vector in the direction of the equivalent cutting main motion,

[0144] In the equivalent plane, the equivalent cutting width w is calculated using formula (11): equ :

[0145] w equ =w e / (D te ·x')#(11)where w e is the cutting width in the bevel cutting condition;

[0146] The equivalent right-angle cutting force is converted into oblique-angle cutting force using formula (12):

[0147]

[0148] Among them, F t is the main cutting force under the equivalent right-angle cutting condition, F f is the feed resistance under equivalent right-angle cutting conditions; F r is the main cutting force under oblique cutting conditions, F s is the feed resistance under oblique cutting conditions, F o It is the cutting depth resistance under oblique cutting conditions.

[0149] Furthermore, the cutting forces of each unit tool are accumulated using the linear synthesis method, including:

[0150] Through coordinate transformation, the tangential force F of the i-th unit tool in the drill bit global coordinate system xyz is obtained using formula (13): eli , radial force F emi and axial force F eni :

[0151]

[0152] Among them, l i 、m i 、n i Represents the unit vectors of the tangential, radial and axial directions of the position of the i-th unit tool; r i 、s i 、o i Respectively represent the unit normal vector of the unit tool working base surface, the unit normal vector of the working cutting plane, and the unit normal vector of the working orthogonal plane; F eri 、F eso 、F eoi They represent the cutting forces in the main motion direction, feed motion direction, and cutting depth direction of the unit tool respectively.

[0153] The torque T of the unit tool is obtained using formula (14): ei :

[0154]

[0155] Among them, x i 、z i They represent the x and z coordinates of the i-th unit tool in the global coordinate system xyz;

[0156] Using formula (15), we can get the torque T and axial force F of the twist drill as a whole: N :

[0157]

[0158] In summary, the present invention provides a drilling force calculation method based on the unit tool linear synthesis method and transfer learning. This method combines the advantages of multiple technologies, has the characteristics of high calculation accuracy, small amount of required experimental data and high calculation efficiency, is suitable for various drilling processing scenarios, and has broad application prospects.

[0159] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. Throughout this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A drilling force calculation method based on unit tool linear synthesis method and transfer learning, characterized by: The following steps are involved: S1: Unit tool division The two main cutting edges and the chisel edge of the drill bit are divided into a series of unit tools connected end to end, each of which cuts with different rake angles, edge inclination angles and cutting amounts; S2: Neural network model for right-angle cutting force calculation A series of right-angle cutting force theoretical calculation results are obtained using the unequal shear zone model as the source domain data set to train the initial neural network model for right-angle cutting force calculation. A target domain dataset was obtained through a small number of right-angle cutting experiments, and a neural network model for right-angle cutting force calculation was trained using transfer learning. S3: Neural network model for calculating oblique cutting forces The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and convert it into the oblique cutting force. The bevel cutting force data obtained from a small number of bevel cutting experiments are used as the target domain dataset, and the final neural network model for bevel cutting force calculation is trained using the transfer learning method. S4: Linear synthesis method The cutting forces of each unit tool are accumulated using the linear synthesis method to obtain the theoretical calculation results of the drilling force of the entire twist drill; S5: Transfer Learning A series of theoretical calculation results of drilling forces using twist drills of different diameters and drilling loads were obtained and used as the source domain dataset to train an initial neural network model for drilling force calculation. Conduct a small number of drilling experiments to obtain the measured data of drilling force experiments and use it as the target domain dataset; Using the transfer learning method, part of the structure, weights or parameters of the initial drilling force calculation neural network model are transferred to the final drilling force calculation neural network. The weights of the first few layers of the neural network are frozen, and a small amount of actual drilling force experimental data is used for secondary training to obtain the final drilling force calculation neural network.

2. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 1 is characterized in that: The drill cutting edge is divided into several unit tools, including: The drill bit radius is divided into n equal parts, and n radii are arranged from large to small: ρ1, ρ2, ρ3, ... ρ i ; With the drill axis as the axis and the radius of each equally divided cylinder as the radius, draw n coaxial cylinders. The tool part located on the same cutting edge of the drill bit sandwiched between two adjacent cylindrical surfaces is defined as a unit tool. That is, the unit tool on the same cutting edge sandwiched between the i-th cylindrical surface and the i+1-th cylindrical surface is defined as the i-th unit tool, and the corresponding cutting edge is P i P i+1 ; Use formula (1) to calculate the position radius ρ of the unit tool i : ρ i =x i i+y i j+z i k #(1) Among them, x i ,y i , z i (i=1,2,……,n) is the dividing point (P i P i+1 The coordinates of the midpoint); i, j, k represent the unit vectors in the x, y, and z directions respectively; Calculate the orientation b of the unit tool using formula (2) i : The synthetic cutting velocity vector V of the unit tool is calculated using formula (3): all : V all =V c +V f =ρ i ×2πNj+Nf / 1000j #(3) Among them, V c is the main motion linear velocity of the unit tool, V f is the feed motion of the unit tool, f is the feed per revolution of the drill (mm / r), and N is the spindle speed (r / min); Formula (4) is used to calculate the unit normal vector of the working base surface r, working cutting plane s and working main section o of the unit tool: Calculate the working edge inclination angle λ of the unit tool using formula (5) se , working front angle γ oe and working principal deflection angle k re : Among them, f' is the unit vector of the projection direction of the unit vector f in the feed direction on the working base surface, The cutting amount of the unit tool (cutting width a) is calculated using formula (6) c and cutting thickness a w ): Among them, z is the number of teeth. For a drill, z = 2; u is the angle between the synthetic cutting speed of the unit tool and the main motion linear velocity, sinu = j·r.

3. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 1 is characterized in that: Formula (7) is used to calculate the normal rake angle α of the unit tool at the chisel edge of the twist drill. n : in, is the cutting edge angle, and ψ is the chisel edge bevel angle.

4. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 1 is characterized in that: Establish a neural network model for right-angle cutting force calculation, including: (1) Using the unequal shear zone model, a series of theoretical calculation results of right-angle cutting forces under different rake angles and cutting parameters are obtained as the source domain data set, and an initial neural network model for right-angle cutting force calculation is trained; (2) Conduct a small number of right-angle cutting experiments to obtain a target domain dataset containing experimental measured data on right-angle cutting forces; (3) Using the transfer learning method, part of the structure, weights or parameters of the initial right-angle cutting force calculation neural network model are transferred to the final right-angle cutting force calculation neural network, and the weights of the first few layers of the final neural network are frozen. A small amount of right-angle cutting force experimental measured data is used for secondary training to obtain the final right-angle cutting force calculation neural network.

5. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 1 is characterized in that: Establish a neural network model for calculating bevel cutting force, including: (1) The equivalent plane method is used to transform the oblique cutting condition into an equivalent right-angle cutting condition. The equivalent rake angle and equivalent right-angle cutting amount are calculated in the equivalent plane. The right-angle cutting force neural network model is used to obtain the equivalent right-angle cutting force and transform it into the oblique cutting force. (2) A series of theoretical calculation results of oblique cutting forces under different cutting angles and cutting parameters based on the equivalent plane method and the right-angle cutting force calculation neural network are used as the source domain data set; (3) Conduct a small number of oblique cutting experiments to obtain experimental measured data of oblique cutting forces as the target domain dataset; (4) Using the transfer learning method, part of the structure, weights or parameters of the initial bevel cutting force calculation neural network model are transferred to the final bevel cutting force calculation neural network, the weights of the first few layers of the final neural network are frozen, and a small amount of experimental measured data of the bevel cutting force is used for secondary training to obtain the final bevel cutting force calculation neural network.

6. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 5 is characterized in that: The equivalent plane method is used to convert the oblique cutting condition into an equivalent right-angle cutting condition, specifically including: The equivalent plane P is defined according to the chip flow direction u' of the oblique cutting and the normal direction n' of the rake face equ ; In the equivalent plane, use formula (8) to calculate the equivalent rake angle α e : tanα n =tanα e ·cosη #(8) Among them, α n is the normal to front angle, η is the chip angle; Establish a rectangular coordinate system xyz, where x is the unit vector of the main cutting motion direction and y is the unit vector of the feed motion direction. Rotate the rectangular coordinate system xyz around the x-axis by λ0 (edge ​​inclination angle) and then rotate it around the transformed y-axis by α n (normal forward angle), and finally rotate η (chip angle) around the transformed x-axis to obtain the coordinate system x'y'z', where x', y', and z' are unit vectors in the x', y', and z' directions; In the equivalent plane, use formula (9) to calculate the equivalent feed f equ : f equ =f e / (D fe ·z') #(9) Among them, f e is the feed rate in the bevel cutting condition, D fe is the unit vector in the equivalent feed direction, In the equivalent plane, the equivalent cutting speed V is calculated using formula (10): equ : In equ =V e ·(D fe ·z′) #(10) Among them, V e is the cutting speed in the bevel cutting condition, D te is the unit vector in the direction of the equivalent cutting main motion, In the equivalent plane, the equivalent cutting width w is calculated using formula (11): equ : w equ =w e / (D te ·x') #(11) Among them, w e is the cutting width in the bevel cutting condition; The equivalent right-angle cutting force is converted into oblique-angle cutting force using formula (12): Among them, F t is the main cutting force under the equivalent right-angle cutting condition, F f is the feed resistance under equivalent right-angle cutting conditions; F r is the main cutting force under oblique cutting conditions, F s is the feed resistance under oblique cutting conditions, F o It is the cutting depth resistance under oblique cutting conditions.

7. The drilling force calculation method based on unit tool linear synthesis method and transfer learning according to claim 1 is characterized in that: The cutting forces of each unit tool are superimposed using the linear synthesis method, including: Through coordinate transformation, the tangential force F of the i-th unit tool in the drill bit global coordinate system xyz is obtained using formula (13): eli , radial force F emi and axial force F eni : Among them, l i 、m i 、n i Represents the unit vectors of the tangential, radial and axial directions of the position of the i-th unit tool; r i 、s i 、o i Respectively represent the unit normal vector of the unit tool working base surface, the unit normal vector of the working cutting plane, and the unit normal vector of the working orthogonal plane; F eri 、F esi 、F eoi Respectively represent the cutting forces in the main motion direction, feed motion direction, and cutting depth direction of the unit tool; The torque T of the unit tool is obtained using formula (14): ei : Among them, x i 、z i They represent the x and z coordinates of the i-th unit tool in the global coordinate system xyz; Using formula (15), we can get the torque T and axial force F of the twist drill as a whole: N : 。

Images