Screw tap performance prediction method

Abstract

The invention discloses a screw tap performance prediction method and relates to the field of screw tap performance prediction. The problem that the screw tap performance cannot be accurately predicted through the process parameters, the geometric parameters and the performance parameters of the screw tap in the prior art is solved. Comprising the following steps of S1, performing data processing on technological parameters and geometric parameters and performance parameters of a to-be-predicted screw tap model; s2, the parameters after data processing are used for setting a BP neural network model; s3, obtaining the optimal weight and bias of the BP neural network model through a multi-particle swarm algorithm; s4, inputting the optimal weight and bias as initial weight and bias into a BP neural network model; s5, training a BP neural network model through the initial weight and the bias, and outputting a prediction result; the combination of the multi-particle swarm optimization and the neural network can prevent the screw tap performance prediction from falling into a local optimal solution, thereby significantly improving the prediction precision and the generalization ability of the model.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical processing, and in particular to a performance prediction technology in a tap processing process. Background Art

[0002] If unreasonable cutting parameters and tool geometry are selected during the cutting process, tap internal thread processing will be negatively affected in many aspects. These unreasonable choices may not only lead to low cutting efficiency and reduced processing quality, but also increase tool wear and even damage the equipment.

[0003] The spindle speed directly affects the cutting efficiency and cutting temperature. Too high or too low speed may lead to poor cutting effect or increased tool wear. The tap rake angle determines the sharpness of the cutting edge and the size of the cutting force. Reasonable rake angle selection can reduce cutting resistance and improve cutting efficiency. The tap back angle is related to the protection of the cutting edge and the stability of the cutting process. The appropriate back angle design can reduce friction and extend the tool life. The number of teeth in the chip part directly affects the stability of the cutting process and the processing quality of the thread. Reasonable number of teeth selection can ensure the stability of the cutting process and obtain high-quality threads. If the geometric parameters are unreasonable, it may also increase the friction between the cutting edge and the workpiece, making the workpiece surface rough and reducing the processing quality. These unreasonable cutting parameters and tool geometric parameters will also accelerate tool wear and shorten the tool life. This will not only affect the continuity and stability of the processing process, but also increase production costs, which will not only increase the maintenance cost of the equipment, but also may affect the normal operation of the production line, causing economic losses to the enterprise.

[0004] The performance of the tap tool determines the production cost during the processing. For example, excessive chip temperature will accelerate tool wear, affect cutting quality and tool life, and reduce cutting efficiency. Tool wear rate directly affects processing accuracy and surface quality. High wear rate requires frequent tool replacement, which increases costs. Excessive cutting force will increase tool wear, may cause cutting edge breakage, and affect the positioning and fixation of the workpiece. Since most taps are selected based on experience during processing, there is no accurate understanding of the processing performance of the selected taps. This may cause damage to the tap or the workpiece being processed, which in turn leads to increased production costs.

[0005] If the finite element analysis is performed on the tool, the simulation time will be too long, thereby increasing the tool design time. The present invention reduces the tool design time by establishing a database and constructing a neural network to replace the finite element simulation process of the tool.

[0006] As a typical artificial neural network, BP neural network has good self-learning and nonlinear mapping capabilities and has been widely used in performance prediction in mechanical processing. However, since the training process of BP neural network is prone to fall into local optimal solutions and its training speed is slow, the prediction accuracy and efficiency are limited. Summary of the invention

[0007] The present invention aims to solve the technical problem that there is a lack of the ability to quickly predict tap performance parameters in the prior art, and to solve the defect that the tap prediction technology in the prior art is prone to fall into a local optimal solution and has a slow training speed, resulting in limited prediction accuracy and efficiency.

[0008] The technical solution adopted by the present invention to solve the above technical problems is a tap performance prediction method, which comprises the following steps:

[0009] S1: performing data processing on process parameters and geometric parameters and performance parameters of a tap model to be predicted, wherein the process parameters include spindle speed, the geometric parameters include tap rake angle, tap clearance angle and number of teeth, and the performance parameters include cutting temperature, tool wear rate and cutting force;

[0010] S2: Use the parameters after data processing to set the BP neural network model;

[0011] S3: The optimal weights and biases of the BP neural network model are obtained through the multi-particle swarm algorithm, specifically:

[0012] S31: Initialize the particle swarm, the parameters of which include the particle swarm size N, the initial inertia weight ω max , maximum number of iterations T, particle velocity v, learning factor and particle dimension D;

[0013] S32: Calculate the comprehensive fitness of each particle in the particle group;

[0014] S33: updating the current optimal solution and the global optimal solution of each particle according to the comprehensive fitness of each particle;

[0015] S34: dynamically adjusting the inertia weight according to the current optimal solution of each particle and the global optimal solution;

[0016] S35: updating the parameters of the particle swarm according to the dynamically adjusted inertia weight;

[0017] S36: Determine whether the comprehensive fitness of the particle meets the termination condition. If the comprehensive fitness of the particle does not meet the termination condition, repeat S32-S35 for iteration. If the comprehensive fitness of the particle meets the termination condition, output the global optimal solution to obtain the optimal weight and bias.

[0018] S4: inputting the optimal weights and biases into the BP neural network model as initial weights and biases;

[0019] S5: training the BP neural network model through the initial weights and biases and outputting the prediction results;

[0020] S6: Obtaining a loss function through the prediction result output by the BP neural network model and calculating the error of the BP neural network model;

[0021] S7: updating the weights and biases of the BP neural network model according to the error, and taking the weights and biases as the optimal weights and biases;

[0022] S8: Obtaining prediction data by updating the BP neural network model weights and the biased BP neural network model;

[0023] S9: Compare the predicted data with the actual simulation data, and detect the accuracy of the prediction model by calculating the mean root mean square error;

[0024] S10: Determine whether the mean square error of the prediction model accuracy reaches the termination condition. If the determination result is yes, repeat S5-S10. If the determination result is no, output the model performance parameters of the tap.

[0025] Furthermore, the data processing described in S1 includes obtaining an orthogonal table of spindle speed and geometric parameters through an orthogonal test, collecting simulation data of performance parameters through the orthogonal table, and normalizing the simulation data.

[0026] Furthermore, the initialization of the particle swarm in S31 includes encoding each particle in a manner of constructing a vector, and the vector equation is:

[0027] P(a)=(α 11 ,α 12 ,…,α ij ,…,α mn ,α 11 ,α 12 ,…,α pq ,…α nw ,β 1 ,β 2 ,…,β i …β n )

[0028] where α ij represents the weight between the i-th input layer node and the j-th hidden layer node, α pq represents the weight between the pth hidden layer node and the qth output layer node, β i is the bias of the i-th hidden layer node, m is the number of input layer nodes, n is the number of hidden layer nodes, and w is the number of output layer nodes.

[0029] Furthermore, the comprehensive fitness of each particle in S32 is expressed by the inverse of the mean square error of each performance parameter predicted by the BP neural network model. The mean square error formula of each performance parameter predicted by the BP neural network model is:

[0030]

[0031] Among them, y i is the actual simulation data value, is the predicted data value, N is the sample size, and the sample size N is the same as the particle population size N.

[0032] Furthermore, the comprehensive fitness of each particle described in S32 is the inverse of the sum of the mean square errors of the performance parameters predicted by the BP neural network model. The comprehensive fitness function of the particle is:

[0033]

[0034] Where MSE i , i=1,2,3 are the mean square errors of each performance parameter, C i Indicates the weight represented by each performance indicator.

[0035] Furthermore, the formula for dynamically adjusting the inertia weight described in S3 is:

[0036]

[0037] Among them, t represents the current iteration number, T max represents the final number of iterations, ω max =0.9,ω min =0.4.

[0038] Furthermore, the updating of the particle swarm parameters in S3 includes the position and speed of the particles,

[0039] The velocity formula of a particle is:

[0040] The particle position formula is:

[0041] Among them, v i t represents the velocity vector of particle i at the tth iteration, x i t represents the position vector of particle i at the tth iteration, p i t represents the optimal position of particle i, g t represents the global optimal position among all particles, ω represents the inertia weight, c 1, c 2 represents the learning factor, c 1 =2.05, c 2 =2.05, r 1 t ,r 2 t : A random number between [0,1].

[0042] Furthermore, the termination condition in S36 is that the fitness of the particle reaches a preset accuracy or a maximum number of iterations.

[0043] Furthermore, the loss function described in S6 is:

[0044]

[0045] in, Indicates the actual simulation data of cutting force, represents the actual simulation data of cutting temperature, μ represents the actual simulation data of cutting temperature.

[0046] Furthermore, the termination condition described in S10 is that the mean square error of the prediction model accuracy is less than a given value.

[0047] Compared with the prior art, the present invention has the following effects:

[0048] The present invention constructs a tap model through a BP neural network model, and the model reflects the nonlinear mapping relationship between the process parameters, geometric parameters and performance parameters of the tap, thereby realizing the prediction of the performance parameters based on the process parameters and geometric parameters.

[0049] The present invention obtains the optimal weights and biases of the BP neural network model through a multi-particle swarm optimization algorithm, effectively avoiding the problem that the BP neural network model is prone to fall into a local optimal solution and the training speed is slow during the training process. The prediction method of the present invention is more accurate in predicting multiple performance indicators such as the cutting force, tool wear rate and chip temperature of the tap. At the same time, it accelerates the network training process and shortens the convergence time of the model. Compared with the traditional BP network training time, it is shortened by about 10% to 20%.

[0050] In addition, the prediction method described in the present invention is applicable to the design process of taps. In this process, the method described in the present invention can be used to predict the tap performance parameters based on the tap model in the design scheme, and the BP neural network model processed by the multi-particle swarm optimization algorithm can simultaneously predict the performance parameters of multiple tap models, and optimize the parameters in the design scheme based on the prediction results, thereby improving the efficiency of product design, providing a more comprehensive reference for the optimization of tap parameters, and improving the design quality.

[0051] In application scenarios where taps are used, the prediction method described in the present invention can be used to predict the performance parameters of the taps. According to the prediction results, the maintenance cycle of the taps can be reasonably arranged to avoid the degradation of processing quality caused by excessive wear of the taps, improve production efficiency, and solve production maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0053] Figure 1 is a flow chart of the present invention;

[0054] Figure 2 It is a schematic diagram of the BP neural network model hierarchy;

[0055] Figure 3 Schematic diagram of the number of teeth N on the chip cone part of the tap geometric parameters;

[0056] Figure 4 The tap rake angle γ is the tap geometric parameter p and tap relief angle α p Schematic diagram of . DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely explain the technical solutions in the embodiments of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict, and the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0058] In the description of the present invention, unless otherwise clearly specified and limited, the terms "connected", "connected", and "fixed" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0059] A method for predicting tap performance, see Figure 1 , including the following steps:

[0060] S1: Data processing is performed on process parameters and geometric parameters and performance parameters of a model of a tap with prediction, wherein the process parameters are machining process parameters including spindle speed, geometric parameters include tap rake angle, tap clearance angle and number of teeth, and performance parameters include cutting temperature, tool wear rate and cutting force;

[0061] S2: Use the parameters after data processing to set the BP neural network model;

[0062] S3: Obtain the optimal weights and biases of the BP neural network model through the multi-particle swarm algorithm, specifically;

[0063] S31: Initialize the particle swarm, the parameters of which include the particle swarm size N, the initial inertia weight ω max , maximum number of iterations T, particle velocity v, learning factor and particle dimension D;

[0064] S32: Calculate the comprehensive fitness of each particle in the particle group;

[0065] S33: updating the current optimal solution and the global optimal solution of each particle according to the comprehensive fitness of each particle;

[0066] S34: dynamically adjusting the inertia weight according to the current optimal solution of each particle and the global optimal solution;

[0067] S35: updating the parameters of the particle swarm according to the dynamically adjusted inertia weight;

[0068] S36: Determine whether the comprehensive fitness of the particle meets the termination condition. If the comprehensive fitness of the particle does not meet the termination condition, repeat S32-S35 for iteration. If the comprehensive fitness of the particle meets the termination condition, output the global optimal solution to obtain the optimal weight and bias.

[0069] S4: inputting the optimal weights and biases into the BP neural network model as initial weights and biases;

[0070] S5: training the BP neural network model through the initial weights and biases and outputting the prediction results;

[0071] S6: Obtaining a loss function through the prediction result output by the BP neural network model and calculating the error of the BP neural network model;

[0072] S7: updating the weights and biases of the BP neural network model according to the error, and taking the weights and biases as the optimal weights and biases;

[0073] S8: Obtaining prediction data by updating the BP neural network model weights and the biased BP neural network model;

[0074] S9: Compare the predicted data with the actual simulation data, and detect the accuracy of the prediction model by calculating the mean root mean square error;

[0075] S10: Determine whether the mean square error of the prediction model accuracy reaches the termination condition. If the determination result is yes, repeat S5-S10. If the determination result is no, output the model performance parameters of the tap.

[0076] The data processing described in S1 includes obtaining an orthogonal table for the spindle speed and geometric parameters through an orthogonal experiment, collecting simulation data of performance parameters through the orthogonal table, and normalizing the simulation data. Specifically, the purpose of setting up an orthogonal experiment is to find the optimal combination of factors through a limited number of experiments. Through orthogonal tables with different coding levels, the experimental results under different combinations can be compared to determine the optimal combination. Through different coding levels, the number of experiments and time costs can be reduced while ensuring the reliability of the experimental results.

[0077] The specific method of setting the BP neural network model in S2 is:

[0078] Establish BP neural network model. Figure 2 The network structure includes input layer, hidden layer and output layer.

[0079] The input layer contains the geometric parameters and processing conditions of the tap, and the output layer is the corresponding cutting performance index. Determine the number of nodes in each layer. The number of nodes in the output layer is The number of nodes in the input layer is ρ. The relationship between the number of nodes in the hidden layer and the number of nodes in the input layer is:

[0080] k≤2ρ+1

[0081] Among them, k is the number of hidden layer nodes, and ρ is the number of input layer nodes.

[0082] Set the error accuracy. The BP neural network model is allowed to have errors, but the error must be within an acceptable range. The smaller the error, the higher the prediction accuracy of the BP neural network model. The error accuracy of the present invention is set to 10 -3 .

[0083] The number of input layer nodes is equal to the number of geometric parameters and processing conditions. The formula for the number of hidden layer nodes is:

[0084]

[0085] Where ρ is the number of nodes in the output layer, is the number of nodes in the output layer, and c is a random number between 1 and 10.

[0086] The number of nodes in the input layer includes the tap rake angle γ p 、Tap back angle α p , the number of teeth on the chip cone is N, and the spindle speed is n. The output layer nodes include chip force F, cutting temperature Q, and tool wear rate W.

[0087] Select the BP neural network model activation function. Commonly used activation functions are Sigmoid, Tanh, Relu and

[0088] Four kinds of ELU, four kinds of activation functions are combined and paired and trained, and the comprehensive mean square error of different combinations is

[0089] When the comprehensive MAE is the minimum among all combinations, the activation function combination is the optimal activation function combination for the neural network.

[0090] Among them, the Sigmoid activation function is:

[0091]

[0092] The tanh activation function is:

[0093]

[0094] The Relu activation function is:

[0095] Relu=max(0,x)

[0096] The ELU activation function is:

[0097]

[0098] The particle swarm initialization in step S31 includes encoding each particle by constructing a vector, and the vector equation is:

[0099] P(a)=(α 11 ,α 12 ,…,α ij ,…,α mn ,α 11 ,α 12 ,…,α pq ,…α nw ,β 1 ,β 2 ,…,β i …β n )

[0100] where α ij represents the weight between the i-th input layer node and the j-th hidden layer node, α pq represents the weight between the pth hidden layer node and the qth output layer node, β i is the bias of the i-th hidden layer node, m is the number of input layer nodes, n is the number of hidden layer nodes, and w is the number of output layer nodes.

[0101] The comprehensive fitness of each particle in S32 is expressed by the inverse of the mean square error of each performance parameter predicted by the BP neural network model. The formula for the mean square error of each performance parameter predicted by the BP neural network model is:

[0102]

[0103] Among them, y i is the actual simulation data value, is the predicted data value, N is the sample size, and the sample size N is the same as the particle population size N.

[0104] The comprehensive fitness of each particle described in S32 is the inverse of the sum of the mean square errors of the performance parameters predicted by the BP neural network model. The comprehensive fitness function of the particle is:

[0105]

[0106] Where MSE i , i=1,2,3 are the mean square errors of each performance parameter, C i Indicates the weight represented by each performance indicator.

[0107] Introducing the weight represented by each performance indicator in the particle fitness function can make the particle fitness function more accurately represent the accuracy of the BP neural network model prediction after each multi-particle swarm algorithm training. Since the proportion of each performance parameter is different in actual processing, the material of the workpiece and other influencing factors of the processing parameters must also be considered when setting the comprehensive weight.

[0108] The formula for dynamically adjusting the inertia weight described in S3 is:

[0109]

[0110] Among them, t represents the current iteration number, T max represents the final number of iterations, ω max =0.9,ω min =0.4.

[0111] S3 updates the particle swarm parameters including the position and velocity of the particles.

[0112] The velocity formula of a particle is:

[0113] The particle position formula is:

[0114] Among them, v i t represents the velocity vector of particle i at the tth iteration, x i trepresents the position vector of particle i at the tth iteration, p i t represents the optimal position of particle i, g t represents the global optimal position among all particles, ω represents the inertia weight, which reflects the inertia of the particle and controls the influence of the current speed on the next speed, c 1 , c 2 represents the learning factor, c 1 =2.05, c 2 =2.05, r 1 t ,r 2 t : represents a random number between [0,1], which increases the randomness of the algorithm and promotes global search.

[0115] The termination condition in S36 is that the fitness of the particle reaches a preset accuracy or a maximum number of iterations.

[0116] The loss function in S6 is:

[0117]

[0118] in, Indicates the actual simulation data of cutting force, Indicates the actual simulation data of cutting temperature, represents the actual simulation data of cutting tool wear rate, F′ represents the cutting force prediction data, T′ represents the cutting temperature prediction data, and μ′ represents the tool wear rate prediction data.

[0119] The termination condition described in S10 is that the mean square error of the prediction model accuracy is less than a given value.

[0120] Obviously, the embodiments of the present invention disclosed above are only used to help illustrate the present invention. The embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. According to the content of this specification, many modifications and changes can be made. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. It is not necessary and impossible to exhaustively list all implementation methods here.

Claims

1. A method for predicting tap performance, characterized in that: The specific steps are as follows: S1: Data processing is performed on process parameters and geometric parameters and performance parameters of a model of a tap with prediction, wherein the process parameters include spindle speed, the geometric parameters include tap rake angle, tap clearance angle and number of teeth, and the performance parameters include cutting temperature, tool wear rate and cutting force; S2: Use the parameters after data processing to set the BP neural network model; S3: Obtain the optimal weights and biases of the BP neural network model through the multi-particle swarm algorithm, specifically; S31: Initialize the particle swarm, the parameters of which include the particle swarm size N, the initial inertia weight ω max , maximum number of iterations T, particle velocity v, learning factor and particle dimension D; S32: Calculate the comprehensive fitness of each particle in the particle group; S33: updating the current optimal solution and the global optimal solution of each particle according to the comprehensive fitness of each particle; S34: dynamically adjusting the inertia weight according to the current optimal solution of each particle and the global optimal solution; S35: updating the parameters of the particle swarm according to the dynamically adjusted inertia weight; S36: Determine whether the comprehensive fitness of the particle meets the termination condition. If the comprehensive fitness of the particle does not meet the termination condition, repeat S32-S35 for iteration. If the comprehensive fitness of the particle meets the termination condition, output the global optimal solution to obtain the optimal weight and bias. S4: inputting the optimal weights and biases into the BP neural network model as initial weights and biases; S5: training the BP neural network model through the initial weights and biases and outputting the prediction results; S6: Obtaining a loss function through the prediction result output by the BP neural network model and calculating the error of the BP neural network model; S7: updating the weights and biases of the BP neural network model according to the error, and taking the weights and biases as the optimal weights and biases; S8: Obtaining prediction data by updating the BP neural network model weights and the biased BP neural network model; S9: Compare the predicted data with the actual simulation data, and detect the accuracy of the prediction model by calculating the mean root mean square error; S10: Determine whether the mean square error of the prediction model accuracy reaches the termination condition. If the determination result is yes, repeat S5-S10. If the determination result is no, output the model performance parameters of the tap.

2. A tap performance prediction method according to claim 1, characterized in that: The data processing described in S1 includes obtaining an orthogonal table of spindle speed and geometric parameters through orthogonal experiments, collecting simulation data of performance parameters through the orthogonal table, and normalizing the simulation data.

3. A tap performance prediction method according to claim 1, characterized in that: The particle swarm initialization in step S31 includes encoding each particle by constructing a vector, and the vector equation is: P(a)=(a 11 ,a 12 ,…,a ij ,…,a mn ,a 11 ,a 12 ,…,a pq ,…a nw ,β1,β2,…,β i …b n ) where α ij represents the weight between the i-th input layer node and the j-th hidden layer node, α pq represents the weight between the pth hidden layer node and the qth output layer node, β i is the bias of the i-th hidden layer node, m is the number of input layer nodes, n is the number of hidden layer nodes, and w is the number of output layer nodes.

4. A tap performance prediction method according to claim 1, characterized in that: The comprehensive fitness of each particle in S32 is expressed by the inverse of the mean square error of each performance parameter predicted by the BP neural network model. The formula for the mean square error of each performance parameter predicted by the BP neural network model is: Among them, y i is the actual simulation data value, is the predicted data value, N is the sample size, and the sample size N is the same as the particle population size N.

5. A tap performance prediction method according to claim 4, characterized in that: The comprehensive fitness of each particle described in S32 is the inverse of the sum of the mean square errors of the performance parameters predicted by the BP neural network model. The comprehensive fitness function of the particle is: Where MSE i , i=1,2,3 are the mean square errors of each performance parameter, C i Indicates the weight represented by each performance indicator.

6. A tap performance prediction method according to claim 1, characterized in that: The formula for dynamically adjusting the inertia weight described in S3 is: Among them, t represents the current iteration number, T max represents the final number of iterations, ω max =0.9,ω min =0.

4.

7. A tap performance prediction method according to claim 1, characterized in that: S3 updates the particle swarm parameters including the position and velocity of the particles. The velocity formula of a particle is: The particle position formula is: Among them, v i t represents the velocity vector of particle i at the tth iteration, x i t represents the position vector of particle i at the tth iteration, p i t represents the optimal position of particle i, g t represents the global optimal position among all particles, ω represents the inertia weight, c1, c2 represent the learning factor, c1 = 2.05, c2 = 2.05, r1 t ,r2 t : A random number between [0,1].

8. A tap performance prediction method according to claim 1, characterized in that: The termination condition in S36 is that the fitness of the particle reaches a preset accuracy or a maximum number of iterations.

9. A tap performance prediction method according to claim 1, characterized in that: The loss function in S6 is: in, Indicates the actual simulation data of cutting force, Indicates the actual simulation data of cutting temperature, represents the actual simulation data of cutting tool wear rate, F′ represents the cutting force prediction data, T′ represents the cutting temperature prediction data, and μ′ represents the tool wear rate prediction data.

10. A tap performance prediction method according to claim 1, characterized in that: The termination condition described in S10 is that the mean square error of the prediction model accuracy is less than a given value.

Images