Micro-milling parameter identification method considering random tool wear

Abstract

The application provides a micro-milling machining parameter identification method considering the influence of random tool wear, comprising the following steps: establishing a cutting force model under the influence of tool wear by considering tool runout and chip separation mechanism; updating the tool rotation radius according to the influence of runout and wear on the tool edge radius value, and obtaining the tool tip trajectory equation; collecting tool wear data in the actual machining process, training the neural network model, and optimizing the hyperparameters in the neural network model by using a bidirectional long short-term memory network; identifying the machining parameter values in the neural network model by using a particle filtering algorithm; simulating and calculating the cutting force and wear values under different working conditions by using the neural network model, comparing the experimental results, and evaluating the accuracy of the machining parameter identification. In the modeling process, the randomness of tool wear is fully considered, the prediction accuracy is significantly improved, and the application value is stronger.

Description

Technical Field

[0001] This invention relates to the field of manufacturing technology for small, complex three-dimensional parts, and in particular to a method for identifying micro-milling machining parameters that takes into account the influence of random tool wear. Background Technology

[0002] With the continuous development of aerospace, medical equipment, and electronic technologies, the manufacturing industry's demand for complex micro-parts is increasing, driving micromachining technology to become an important research and application direction in the machining field. By monitoring tool wear and analyzing tool stress in advance, the safety and sustainability of the micro-milling process can be significantly improved.

[0003] The impact of tool wear is more significant in micro-milling than in conventional milling. Because micro-milling involves small cutting depths and high cutting speeds, tool wear rapidly affects machining quality and efficiency. The paper "Analytical modelling and experimental validation of micro-ball-end milling forces with progressive tool flank wear" proposes a model based on tool wear mechanisms to estimate the impact of flank wear on cutting forces and optimize the machining process. The paper "Micro-Milling Tool Wear Monitoring via Nonlinear Cutting Force Model" establishes a cutting force model incorporating tool wear through a stepwise optimization method. Furthermore, due to the time-varying characteristics of micro-milling, accurate identification of real-time cutting parameters is crucial for modeling micro-milling mechanisms and predicting machining states. The paper "An improved time-varying stability analysis of micro-milling considering tool wear" derives cutting force coefficients by fitting experimental data and establishes a time-varying modified force model. The paper "Time-varying reliability and global sensitivity analysis of regenerative chatter stability in turning considering tool wear" uses a gamma process model to describe the impact of tool wear on the cutting force coefficients. The literature "Investigation of tool flank wear effect on system stability prediction in the milling of Ti-6Al-4V thin-walled workpiece" considers friction and process damping effects and proposes a model for solving time-varying milling force coefficients. These studies provide support for tool wear modeling and optimization in micro-milling, improving machining accuracy and efficiency.

[0004] The aforementioned studies have laid the foundation for identifying milling machining parameters. However, these studies assume that the machining parameters are deterministic; in reality, these parameters are stochastic in practical engineering due to the influence of loads and environmental conditions. Furthermore, accurately calculating random wear values ​​and incorporating the wear process into the solution of machining parameters is a relatively rare approach. Therefore, existing research has certain limitations. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for identifying micro-milling machining parameters that considers the influence of random tool wear. First, the invention analyzes the micro-milling mechanism, studies the variation law of tool tip wear and cutting thickness, and establishes a corresponding mathematical model. Next, an experimental platform is built to acquire micro-milling cutting test data. Based on the experimental data, a neural network framework is constructed to achieve real-time prediction of tool wear values. Using real-time cutting data and the cutting model, combined with an improved particle filtering algorithm, the dynamic cutting force coefficient is solved. This method can simulate random tool wear and integrate wear information into the machining parameter identification process, thereby improving the accuracy of the micro-milling mathematical model. This invention is suitable for accurate modeling of micro-milling, can monitor tool wear in real time, reduce equipment damage, and lower manufacturing costs.

[0006] The technical means employed in this invention are as follows:

[0007] A method for identifying micro-milling parameters considering the influence of random tool wear includes:

[0008] A cutting force model under the influence of tool wear is established, taking into account tool runout and chip separation mechanisms.

[0009] Based on the influence of runout and wear on the tool edge radius, the tool rotation radius is updated, and the tool tip trajectory equation is obtained;

[0010] Data on tool wear during actual machining processes are collected to train a neural network model, and a bidirectional long short-term memory network is used to optimize the hyperparameters in the neural network model.

[0011] The particle filtering algorithm is used to identify the processing parameter values ​​in the neural network model;

[0012] By using a neural network model to simulate and calculate cutting forces and wear values ​​under different working conditions, and comparing the experimental results, the accuracy of machining parameter identification is evaluated.

[0013] Furthermore, the establishment of a cutting force model under the influence of tool wear, considering tool runout and chip separation mechanisms, specifically includes:

[0014] Considering the contact principle between the surface of the workpiece and the cutting tool, cutting force models for the shear-dominated zone and the plowing-dominated zone are established respectively.

[0015] In the shear-dominant region, material separation occurs to form chips, and the radial and axial cutting force components are expressed as:

[0016]

[0017] Wherein dF cs and dF ts dF represents the shear force component in the shear-dominant region. cpand dF tp dF represents the plowing force component in the shear-dominated region. cv and dF tv K represents the frictional force component in the shear-dominant region. ns and K fs K represents the shear cutting coefficient in the shear-dominant region. np and K fp K represents the plowing coefficient in the shear-dominated zone. nv and K fv dz represents the wear coefficient in the shear-dominant region; α represents the height of the discrete element of the cutting edge; e ζ represents the instantaneous effective rake angle of the shear; h represents the instantaneous effective clearance angle of the shear. ψk Indicates the instantaneous cutting thickness; h min Indicates the minimum cutting thickness; h er It is the elastic recovery height of the workpiece; ζ v It is the effective included angle;

[0018] No chips are formed in the plowing-dominant zone; material accumulates in front of the blades. The radial and axial cutting force components are represented as follows:

[0019]

[0020] Wherein dF cp 'and dF tp 'Indicates the plowing force component in the dominant plowing area; dF cv 'and dF tv 'Indicates the frictional component in the dominance zone of plowing; α p P represents the instantaneous effective leading angle of plowing. e Indicates the material's elastic recovery rate; h ac The height of the material stack is represented as:

[0021]

[0022] Where E and σ0 are the performance parameters of the workpiece; r vp This indicates the radius of the cutting edge after tool wear;

[0023] Combining the radial and axial cutting force components of the shear-dominant zone and the plowing-dominant zone, the cutting force model under the influence of tool wear is expressed as follows:

[0024]

[0025] Where, ψ k It is the instantaneous cutting position angle of the tool, and M is the total number of axial cutting units of the cutting edge.

[0026] Further, obtaining the blade tip trajectory equation includes:

[0027] Based on the impact of runout and wear on the tool edge radius value, the tool rotation radius is updated as follows:

[0028] R v-update =(R v cos(r a +2π(k-1) / K)+r v ) 2 +(R v sin r a ) 2 -VP

[0029] Among them, R v Indicates the nominal diameter of the cutting tool; R v-update This represents the tool radius after considering tool runout and tool wear; r v Indicates the jump distance; r a Indicates the runout angle; k represents the current cutting tooth, K represents the total number of cutting teeth, and VP represents the aggregate distance related to the tool wear width;

[0030] By establishing a coordinate system, the equation of the blade tip trajectory is obtained:

[0031]

[0032] Where, x k and y k Indicates the coordinates of the tool tip position; x Ok and y Ok Indicates the coordinates of the tool center position; f tk Indicates cutting speed; r t It is the radius of the original cutting edge; r vp ω represents the wear radius of the cutting edge of the tool; ω represents the angular velocity of the tool's rotation.

[0033] Furthermore, the training of the neural network model and the optimization of its hyperparameters using a bidirectional long short-term memory network specifically include:

[0034] Feature extraction and representation techniques are used to perform wavelet transform denoising on the tool wear data collected during the actual machining process, reducing the data dimensionality without losing tool wear features; the unbiased contribution of each feature output is calculated using the max-min normalization method.

[0035]

[0036] in, This represents the data before normalization. This represents the normalized data. This represents the maximum data value before normalization. This represents the minimum data value before normalization.

[0037] Based on the trend of data T RE Monotonicity M ON and correlation coefficient C OR Used as a feature selection criterion for screening:

[0038]

[0039] Where N is the sample size. It is feature y k The i-th value, y k 'is y k The average value, t k Indicates a work cycle. dy represents the average value across all work cycles. k express and The difference between them, where w represents tool wear;

[0040] A bidirectional long short-term memory network is used to construct training and validation sets based on tool wear data and create an objective function to optimize the hyperparameters in the neural network model. The network framework of the neural network model includes a forget gate, an input gate, and an output gate.

[0041] The forget gate portion is obtained from the input data and is used to control information C. t-1 The degree of forgetting is determined in the following ways:

[0042] f t =σ(w f [s t-1 ,x t ]+b f )

[0043] Among them, f t Indicates the output of the forget gate, [s t-1 ,x t ] represents the hidden state s of the previous moment. t-1 Input x at the current time t splicing, w f Let b be the weight matrix of the forget gate. f For the bias term of the forget gate;

[0044] The input gate section is used to control the degree of information updating, and is represented as follows:

[0045] i t =σ(w i [s t-1 ,x t ]+b i )

[0046] Among them, it b represents the output of the input gate. i The bias term for the input gate memory unit value; w i This represents the weight matrix of the input gate;

[0047] Current new LSTM state c t Based on the state c of the previous moment t-1 Represented as:

[0048] c t =f t ×c t-1 +i t tanh(w c [s t-1 ,x t ]+b c )

[0049] Among them, w c b represents the weight matrix associated with the candidate memory cell values. c The bias term for the input gate and candidate memory cell values ​​is passed through the output gate o. t Control information output, the unit output is represented as:

[0050] s t =o t ×tanh(c t )=σ(w o [s t-1 ,x t ]+b o )×tanh(c t )

[0051] Among them, w o and b o These are the weight matrix and offset vector of the aforementioned gate, where σ and tanh represent the sigmoid and hyperbolic tangent activation functions, respectively.

[0052] A backward layer was added to the forward layer of the neural network model, and the forward and backward hidden layer vectors were concatenated. The mathematical expression is as follows:

[0053]

[0054] in, and O t represents the final output of the forward layer, backward layer, and output layer at time t, respectively. f is the network unit, g is ReLU, and w* is the weight.

[0055] Furthermore, the step of using a particle filtering algorithm to identify the processing parameter values ​​in the neural network model specifically includes:

[0056] For multiple cutting force coefficients in the neural network model, a particle filtering algorithm is used to identify the parameter values ​​at any given time.

[0057] Assume the dynamics of the system using the state equation and observation equation:

[0058]

[0059] Where, x k It is time t k The system state at y k It is related to time t k The state x at time k The corresponding observed value, z k-1 and o k Let f represent the process noise and observation noise, respectively. Let f represent the current and previous system states, and h represent the relationship between the system state and the current time measurement. Substitute the obtained posterior probability density into the next prediction calculation to form a recursive state update.

[0060] P(x k |y 1:k )=P(y k |x k )P(x k |y 1:k-1 ) / P(y k |y 1:k-1 )

[0061] Wherein, P(x k |y 1:k-1 ) is the prior probability, P(y) k |x 1:k ) is the posterior probability, P(y) k |x k ) is the likelihood function, P(y) k |y k-1 ) is the normalization constant;

[0062] By introducing the Monte Carlo method to replace the integration process with the average value, the expected value of the function f(x) can be approximately estimated from the known sampling distribution Q(x). t |y 1:t The sampling in the sample and the calculation of the expectation of f(x) are as follows:

[0063]

[0064] Among them, w k These are normalized weights, calculated recursively using ordinal importance sampling. The recursive form of the particle weight value is expressed as:

[0065] w k (i) =wk-1 (i) P(y k |x k (i) )P(x k (i) |x k-1 (i) ) / Q(x t (i) |x 0:k-1 (i) ,y 1:k )

[0066] The particle swarm that caused particle degradation was resampled, and the degree of particle degradation was determined by the number of effective particles N. eff To measure:

[0067]

[0068] Where N is the preset effective sample count threshold;

[0069] Considering the effects of chatter and tool runout in the machining system, the time index t is treated as a variable in the wear process modeling. The obtained wear prediction model is integrated into the mechanical modeling process. Based on the established cutting force model, the future cutting force prediction state equation and observation equation, including tool wear, are expressed as follows:

[0070]

[0071] Where, maxF xk 、maxF yk rmsF xk and rmsF yk V represents the observed maximum cutting force and the root mean square values ​​of the cutting forces in the x-axis and y-axis directions, respectively. ns v np v fs v fp v nv and v fv It is Gaussian distributed process noise, f k It is observation noise;

[0072] The cutting coefficient value is derived based on the measured data and the above equations, so that the values ​​of each parameter of the model are fully obtained.

[0073] Compared with the prior art, the present invention has the following advantages:

[0074] This invention provides a method for identifying micro-milling parameters considering the influence of random tool wear. It establishes a cutting force model under the influence of tool wear, taking into account tool runout and chip separation mechanisms. Based on the influence of runout and wear on the tool edge radius, the tool rotation radius is updated to obtain the tool tip trajectory equation. Tool wear data from actual machining processes is collected to train the neural network model, and a bidirectional long short-term memory network is used to optimize the hyperparameters in the neural network model. A particle filtering algorithm is used to identify the machining parameter values ​​in the neural network model. The neural network model is then used to simulate and calculate cutting forces and wear values ​​under different working conditions, and experimental results are compared to evaluate the accuracy of the machining parameter identification.

[0075] This invention provides a method for identifying micro-milling parameters considering the influence of random tool wear. First, it analyzes the impact of tool wear on instantaneous tool tip trajectory and chip separation state, models the wear process, and integrates it into the cutting force model. Next, a bidirectional long short-term memory (BiLSTM) network is used to predict random tool wear, and the hyperparameters in the network are optimized. Given the time-varying characteristics of micro-milling, experimental data is further combined with particle filtering to obtain the cutting force coefficients, ultimately establishing the mechanical model. Finally, a series of micro-milling experiments are conducted to verify the consistency between the simulation results and experimental data, thus ensuring the accuracy of the model. Compared with traditional micro-milling modeling methods, this invention fully considers the randomness of tool wear during the modeling process, significantly improving prediction accuracy and possessing stronger practical application value.

[0076] Based on the above reasons, this invention can be widely applied in fields such as the manufacturing of small, complex three-dimensional parts. Attached Figure Description

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0078] Figure 1 This is a flowchart of the micro-milling machining parameter identification method considering the influence of random tool wear in this invention.

[0079] Figure 2 This is a diagram showing the change in the edge radius of the micro-milling tool due to wear in an embodiment of the present invention.

[0080] Figure 3 This is a schematic diagram illustrating the principle of tool runout measurement in an embodiment of the present invention.

[0081] Figure 4This is a schematic diagram of neural network prediction of tool wear in an embodiment of the present invention.

[0082] Figure 5 This is a schematic diagram illustrating the basic concept of particle filtering in an embodiment of the present invention.

[0083] Figure 6 This is a comparison diagram of simulated wear of micro-milling tools in an embodiment of the present invention.

[0084] Figure 7 This is a comparison chart of micro-milling force simulation and testing in an embodiment of the present invention. Detailed Implementation

[0085] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0086] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0088] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0089] like Figure 1 As shown, the present invention provides a method for identifying micro-milling machining parameters considering the influence of random tool wear, including:

[0090] A cutting force model under the influence of tool wear is established, taking into account tool runout and chip separation mechanisms.

[0091] In a specific implementation, as a preferred embodiment of the present invention, the step of establishing a cutting force model under the influence of tool wear, considering tool runout and chip separation mechanisms, specifically includes:

[0092] Considering the contact principle between the surface of the workpiece and the cutting tool, cutting force models for the shear-dominated zone and the plowing-dominated zone are established respectively.

[0093] In the shear-dominant region, material separation occurs to form chips, and the radial and axial cutting force components are expressed as:

[0094]

[0095] Wherein dF cs and dF ts dF represents the shear force component in the shear-dominant region. cp and dF tp dF represents the plowing force component in the shear-dominated region. cv and dF tv K represents the frictional force component in the shear-dominant region. ns and K fs K represents the shear cutting coefficient in the shear-dominant region. np and K fp K represents the plowing coefficient in the shear-dominated zone. nv and K fv dz represents the wear coefficient in the shear-dominant region; α represents the height of the discrete element of the cutting edge; eζ represents the instantaneous effective rake angle of the shear; h represents the instantaneous effective clearance angle of the shear. ψk Indicates the instantaneous cutting thickness; h min Indicates the minimum cutting thickness; h er It is the elastic recovery height of the workpiece; ζ v It is the effective included angle;

[0096] No chips are formed in the plowing-dominant zone; material accumulates in front of the blades. The radial and axial cutting force components are represented as follows:

[0097]

[0098] Wherein dF cp 'and dF tp 'Indicates the plowing force component in the dominant plowing area; dF cv 'and dF tv 'Indicates the frictional component in the dominance zone of plowing; α p P represents the instantaneous effective leading angle of plowing. e Indicates the material's elastic recovery rate; h ac The height of the material stack is represented as:

[0099]

[0100] Where E and σ0 are the performance parameters of the workpiece; r vp This indicates the radius of the cutting edge after tool wear;

[0101] Combining the radial and axial cutting force components of the shear-dominant zone and the plowing-dominant zone, the cutting force model under the influence of tool wear is expressed as follows:

[0102]

[0103] Where, ψ k It is the instantaneous cutting position angle of the tool, and M is the total number of axial cutting units of the cutting edge.

[0104] Based on the influence of runout and wear on the tool edge radius, the tool rotation radius is updated, and the tool tip trajectory equation is obtained;

[0105] In a specific implementation, as a preferred embodiment of the present invention, obtaining the blade tip trajectory equation includes:

[0106] The tool tip radius changes with tool wear, and this change further affects the instantaneous cutting thickness. The change in the actual tool edge radius due to tool wear can be categorized into three cases. On the other hand, tool runout also affects the cutting force. The runout distance and angle reflect the misalignment of the tool's geometric axis with its central rotation axis, leading to changes in the actual tool position and rotation radius. Based on the influence of runout and wear on the tool edge radius, the tool rotation radius is updated as follows:

[0107] R v-update =(R v cos(r a +2π(k-1) / K)+r v ) 2 +(R v sinr a ) 2 -VP

[0108] Among them, R v Indicates the nominal diameter of the cutting tool; R v-update This represents the tool radius after considering tool runout and tool wear; r v Indicates the jump distance; r a Indicates the runout angle; k represents the current cutting tooth, K represents the total number of cutting teeth, and VP represents the aggregate distance related to the tool wear width;

[0109] The instantaneous cutting thickness during the cutting process changes over time and is influenced by both the tool tip radius and the tool cutting trajectory. By establishing a coordinate system, the tool tip trajectory equation can be obtained:

[0110]

[0111] Where, x k and y k Indicates the coordinates of the tool tip position; x Ok and y Ok Indicates the coordinates of the tool center position; f tk Indicates cutting speed; r t It is the radius of the original cutting edge; r vp ω represents the wear radius of the cutting edge of the tool; ω represents the angular velocity of the tool's rotation.

[0112] Data on tool wear during actual machining processes are collected to train a neural network model, and a bidirectional long short-term memory network is used to optimize the hyperparameters in the neural network model.

[0113] In a specific implementation, as a preferred embodiment of the present invention, the step of training the neural network model and optimizing the hyperparameters in the neural network model using a bidirectional long short-term memory network specifically includes:

[0114] Feature extraction and representation techniques are used to perform wavelet transform denoising on the tool wear data collected during the actual machining process, reducing the data dimensionality without losing tool wear features. Since the extracted features have varying amplitudes, directly inputting the features into the neural network negatively impacts the algorithm's learning phase and convergence speed. The unbiased contribution of each feature output is calculated using the max-min normalization method.

[0115]

[0116] in, This represents the data before normalization. This represents the normalized data. This represents the maximum data value before normalization. This represents the minimum data value before normalization.

[0117] Based on the trend of data T RE Monotonicity M ON and correlation coefficient C OR Used as a feature selection criterion for screening:

[0118]

[0119] Where N is the sample size. It is feature y k The i-th value, y k 'is y k The average value, t k Indicates a work cycle. dy represents the average value across all work cycles. k express and The difference between them, where w represents tool wear;

[0120] A bidirectional long short-term memory network is employed to construct training and validation sets based on tool wear data and create an objective function. Hyperparameters in the neural network model are then optimized to ensure more accurate predictions. The network framework of the neural network model includes a forget gate, an input gate, and an output gate.

[0121] The forget gate portion is obtained from the input data and is used to control information C. t-1 The degree of forgetting is determined in the following ways:

[0122] f t =σ(w f [s t-1 ,x t ]+b f )

[0123] Among them, f t Indicates the output of the forget gate, [s t-1,x t ] represents the hidden state s of the previous moment. t-1 Input x at the current time t splicing, w f Let b be the weight matrix of the forget gate. f For the bias term of the forget gate;

[0124] The input gate section is used to control the degree of information updating, and is represented as follows:

[0125] i t =σ(w i [s t-1 ,x t ]+b i )

[0126] Among them, i t b represents the output of the input gate. i The bias term for the input gate memory unit value; w i This represents the weight matrix of the input gate;

[0127] Current new LSTM state c t Based on the state c of the previous moment t-1 Represented as:

[0128] c t =f t ×c t-1 +i t tanh(w c [s t-1 ,x t ]+b c )

[0129] Among them, w c b represents the weight matrix associated with the candidate memory cell values. c The bias term for the input gate and candidate memory cell values ​​is passed through the output gate o. t Control information output, the unit output is represented as:

[0130] s t =o t ×tanh(c t )=σ(w o [s t-1 ,x t ]+b o )×tanh(c t )

[0131] Among them, w o and b o These are the weight matrix and offset vector of the aforementioned gate, where σ and tanh represent the sigmoid and hyperbolic tangent activation functions, respectively.

[0132] A backward layer was added to the forward layer of the neural network model, and the forward and backward hidden layer vectors were concatenated. The mathematical expression is as follows:

[0133]

[0134] in, and O t represents the final output of the forward layer, backward layer, and output layer at time t, respectively. f is the network unit, g is ReLU, and w* is the weight.

[0135] The new measurement data is fed into the neural network training model to obtain the actual wear value at any time, and then substituted into the proposed model to simulate the cutting process.

[0136] The particle filtering algorithm is used to identify the processing parameter values ​​in the neural network model;

[0137] The basic process of particle filtering is to extract a set of random samples from the state space and use the sample mean instead of integration to obtain a minimum variance estimate of the system state, thus approximating the probability density function. In a preferred embodiment of this invention, the use of particle filtering to identify processing parameter values ​​in a neural network model specifically includes:

[0138] For multiple cutting force coefficients in the neural network model, a particle filtering algorithm is used to identify the parameter values ​​at any given time.

[0139] Assume the dynamics of the system using state equations and observation equations, such as Figure 5 As shown:

[0140]

[0141] Where, x k It is time t k The system state at y k It is related to time t k The state x at time k The corresponding observed value, z k-1 and o k Let f represent the process noise and observation noise, respectively. Let f represent the current and previous system states, and h represent the relationship between the system state and the current time measurement. Substitute the obtained posterior probability density into the next prediction calculation to form a recursive state update.

[0142] P(x k |y 1:k )=P(y k |x k )P(x k |y 1:k-1 ) / P(yk |y 1:k-1 )

[0143] Wherein, P(x k |y 1:k-1 ) is the prior probability, P(y) k |x 1:k ) is the posterior probability, P(y) k |x k ) is the likelihood function, P(y) k |y k-1 ) is the normalization constant;

[0144] By introducing the Monte Carlo method to replace the integration process with the average value, the expected value of the function f(x) can be approximately estimated from the known sampling distribution Q(x). t |y 1:t The sampling in the sample and the calculation of the expectation of f(x) are as follows:

[0145]

[0146] Among them, w k These are normalized weights. Directly calculating the weight of each particle would reduce computational efficiency. We use ordered importance sampling to calculate the weights recursively. The recursive form of the particle weight value is expressed as:

[0147] w k (i) =w k-1 (i) P(y k |x k (i) )P(x k (i) |x k-1 (i) ) / Q(x t (i) |x 0:k-1 (i) ,y 1:k )

[0148] The particle swarm that causes particle degradation is resampled to prevent critical weights from concentrating on a few particles. The degree of particle degradation is determined by the number of effective particles N. eff To measure:

[0149]

[0150] Where N is the preset effective sample count threshold;

[0151] Considering the effects of chatter and tool runout in the machining system, the time index t is treated as a variable in the wear process modeling. The obtained wear prediction model is integrated into the mechanical modeling process. Based on the established cutting force model, the future cutting force prediction state equation and observation equation, including tool wear, are expressed as follows:

[0152]

[0153] Where, maxF xk 、maxF yk rmsF xk and rmsF yk V represents the observed maximum cutting force and the root mean square values ​​of the cutting forces in the x-axis and y-axis directions, respectively. ns v np v fs v fp v nv and v fv It is Gaussian distributed process noise, f k This is observation noise; due to the existence of multiple observation equations, the recommended distribution is a joint normal distribution. The cutting coefficient values ​​are derived based on measured data and the above equations, ensuring that all model parameter values ​​are obtained completely.

[0154] By using a neural network model to simulate and calculate cutting forces and wear values ​​under different working conditions, and comparing the experimental results, the accuracy of machining parameter identification is evaluated.

[0155] Example 1

[0156] In this embodiment, Figure 2 Taking a sample, this study investigates the influence of random tool wear on the cycloidal trajectory of the cutting edge, analyzes the law governing the change in the geometry of the tool tip edge with varying wear values, and shows that the change in the actual tool tip radius caused by tool wear can be further divided into... Figure 2 There are three scenarios.

[0157] Figure 2 (a) To account for the geometric details of the tool tip considering the effect of wear value VP, the distance on the diagram can be calculated as follows:

[0158]

[0159] Where γ is the tool clearance angle, α is the effective rake angle of the tool, and r t It is the radius of the original cutting edge. According to the appendix... Figure 2 In geometric relationships, the line segment distance value in the above formula can be expressed as:

[0160]

[0161] Appendix Figure 2The distance V*W*(*=1,2,3) in (a) can be used as follows: Figure 2 The mapping distance V in (b) o W o To describe it, it can be solved as follows:

[0162]

[0163] Tool runout affects the cutting force value; the measurement process is shown in the attached figure. Figure 3 As shown. The tool rotation radius affected by runout can be updated as follows:

[0164]

[0165] Among them, R v Indicates the nominal diameter of the cutting tool; R r Tool radius considering tool runout; r v Indicates the jump distance; r a The value indicates the runout angle; k represents the current cutting tooth, and K represents the total number of cutting teeth.

[0166] The tool rotation radius affected by runout and wear can be updated as follows:

[0167]

[0168] The time-dependent nonlinear equations are obtained using Newton-Raphson iteration, combined with the appendix. Figure 2 The geometric relationships in (c) require the actual instantaneous uncut thickness to be calculated by superposition:

[0169]

[0170] Wherein, λ is a proportionality parameter obtained experimentally, and P e It is the material's elastic recovery rate.

[0171] like Figure 4 As shown in Table 1, the algorithm for predicting tool wear using a neural network model is executed.

[0172] Table 1 Algorithm Execution Process

[0173]

[0174]

[0175] Based on the obtained model, cutting forces and wear values ​​under different working conditions were simulated and calculated. The experimental and simulation results were compared to evaluate the accuracy of the machining parameter identification and verify the model's predictive accuracy. Experimental conditions are shown in Table 2, and the comparison results are as follows: Figure 6 and Figure 7 As shown.

[0176] Table 2 Processing Conditions

[0177]

[0178] Table 3 shows the predicted values, actual values, and relative errors listed in the table for each method. (See appendix...) Figure 6 It can be seen that the wear trend predicted by the wear prediction is very consistent with the experimental values, and the prediction effect of the proposed method is better than that of the comparison methods. Therefore, it is reasonable to determine the tool wear value based on the proposed BiLSTM identification method.

[0179] Table 3. Tool wear prediction error under different methods (T1-3-2 represents tooth 2 of the third tool under condition 1)

[0180]

[0181] The results in Table 4 show that the calculated cutting parameters differ as the cutting time increases. To verify the accuracy of the mechanical model, coefficients were substituted into the model to calculate the cutting force, as shown in the appendix. Figure 7 As shown, the simulation-measurement comparison curves demonstrate that the model constructed based on the proposed parameter identification method is accurate.

[0182] Table 4. Calibration results of cutting force coefficient and runout parameters

[0183]

[0184] Example 2

[0185] The present invention also provides a method for constructing a micro-milling machining center, specifically including: a model building module, a measurement module, a tool wear prediction module, a parameter identification module, and a model verification module.

[0186] The model building module derives the cycloidal trajectory equation of the cutting edge under the combined effects of tool runout and tool wear; and updates the mathematical model in micro-milling mechanics modeling, which is mainly based on shearing and plowing, considering the influence of wear on the cutting edge radius.

[0187] The measurement module is an experimental platform consisting of a multi-axis machining center, a triaxial force sensor, a scanning electron microscope, a data acquisition system, a laser displacement sensor, and computer control software. It completes the accurate storage of data through cutting operations and disassembly / assembly tests between machine tools, cutting tools, and workpieces.

[0188] The tool wear prediction module introduces a neural network and trains and validates it using measurement data to build a tool wear prediction framework, thereby achieving accurate prediction of wear values.

[0189] The parameter identification module, based on the mechanical model and measurement data, uses a particle filtering algorithm to perform probability updates on dynamic data and establishes a real-time coefficient solving equation to obtain the model coefficient values ​​at any time.

[0190] The model verification module compares the cutting simulation results and test results under multiple working conditions, analyzes the accuracy of tool wear prediction, and measures the accuracy of model establishment.

[0191] like Figure 3 As shown, a cutting experiment was conducted on a vertical machining center with a maximum spindle speed of 24,000 rpm. The workpiece used in the cutting experiment was Al6061. The micro end mill was a double-edged tungsten carbide coated end mill with a radius of 0.5 mm and a helix angle of 30°. A CL-YD-3210 force sensor was mounted on the worktable, and two accelerometers were fixed to the x- and y-direction surfaces of the workpiece, respectively. During processing, the signals collected by the channel were transmitted to the data acquisition system through a multi-channel charge amplifier.

[0192] Wear values ​​were measured using a high-precision scanning electron microscope. Furthermore, a laser displacement sensor fixed to the machine tool was used to measure the real-time deformation of the tool during machining. Multi-condition milling tests were conducted on a micromachining center to determine the optimal tool wear measurement time, enabling efficient microscopic measurement of tool wear values.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying micro-milling machining parameters considering the influence of random tool wear, characterized in that, include: A cutting force model under the influence of tool wear is established, taking into account tool runout and chip separation mechanisms. Specifically, this includes: Considering the contact principle between the surface of the workpiece and the cutting tool, cutting force models for the shear-dominated zone and the plowing-dominated zone are established respectively. In the shear-dominant region, material separation occurs to form chips, and the radial and axial cutting force components are expressed as: in, dF cs and dF ts This represents the shear force components in the shear-dominant region. dF cp and dF tp This represents the plowing force component in the shear-dominated zone. dF cv and dF tv This represents the frictional force component in the shear-dominant region; K ns and K fs This represents the shear cutting coefficient in the shear-dominant region. K np and K fp Indicates the plowing coefficient under the shear-dominated zone. K nv and K fv This represents the wear coefficient in the shear-dominated region; dz This represents the height of the discrete element of the cutting edge; α e Indicates the effective rake angle at the instantaneous shearing moment; ζ Indicates the effective clearance angle at the instantaneous shearing; h ψk Indicates the instantaneous cutting thickness; h min Indicates the minimum cutting thickness; h er It is the elastic recovery height of the workpiece; ζ v It is the effective included angle; No chips are formed in the plowing-dominant zone; material accumulates in front of the blades. The radial and axial cutting force components are represented as follows: in, dF cp ' and dF tp ' Indicates the plowing force component in the dominant plowing area; dF cv ' and dF tv ' This indicates the frictional force component in the dominant tillage area; α p Indicates the instantaneous effective leading angle of plowing. Indicates the material's elastic recovery rate; h ac The height of the material stack is represented as: in, E and σ 0 represents the performance parameters of the workpiece; Indicates the radius of the cutting edge after tool wear; Combining the radial and axial cutting force components of the shear-dominant zone and the plowing-dominant zone, the cutting force model under the influence of tool wear is expressed as follows: in, ψ k It is the instantaneous cutting position angle of the tool. M This represents the total number of axial cutting units on the cutting edge; Based on the influence of runout and wear on the tool edge radius, the tool rotation radius is updated, and the tool tip trajectory equation is obtained; Data on tool wear during actual machining processes are collected to train a neural network model, and a bidirectional long short-term memory network is used to optimize the hyperparameters in the neural network model. The particle filtering algorithm is used to identify the processing parameter values ​​in the neural network model; By using a neural network model to simulate and calculate cutting forces and wear values ​​under different working conditions, and comparing the experimental results, the accuracy of machining parameter identification is evaluated.

2. The method for identifying micro-milling machining parameters considering the influence of random tool wear according to claim 1, characterized in that, The process of obtaining the blade tip trajectory equation includes: Based on the impact of runout and wear on the tool edge radius value, the tool rotation radius is updated as follows: in, R v Indicates the nominal diameter of the cutting tool; R v-update This represents the tool radius after taking into account tool runout and tool wear. r v Indicates the jump distance; r a Indicates the angle of jump; k Indicates the current cutting tooth. K Indicates the total number of cutting teeth. VP This represents the set distance related to the tool wear width; By establishing a coordinate system, the equation of the blade tip trajectory is obtained: in, x k and y k Indicates the coordinates of the blade tip position; x Ok and y Ok Indicates the coordinates of the tool center position; f tk Indicates cutting speed; r t It is the radius of the original cutting edge; r vp It is the radius of the cutting edge due to tool wear; ω This indicates the angular velocity of the tool rotation.

3. The method for identifying micro-milling machining parameters considering the influence of random tool wear according to claim 1, characterized in that, The training of the neural network model and the optimization of its hyperparameters using a bidirectional long short-term memory network specifically include: Feature extraction and representation techniques are used to perform wavelet transform denoising on the tool wear data collected during the actual machining process, reducing the data dimensionality without losing tool wear features; the unbiased contribution of each feature output is calculated using the max-min normalization method. in, xi k This represents the data before normalization. yi k This represents the normalized data. xmax k This represents the maximum data value before normalization. xmin k This represents the minimum data value before normalization. Based on data trends T RE Monotonicity M ON and correlation coefficient C OR Used as a feature selection criterion for screening: in, N It's the sample size. yi k It is a feature y k The i One value, y k ' yes y k The average value, t k Indicates a work cycle. This represents the average value across all work cycles. dy k express yi k and yi -1 k The difference between them w Indicates tool wear; A bidirectional long short-term memory network is used to construct training and validation sets based on tool wear data and create an objective function to optimize the hyperparameters in the neural network model. The network framework of the neural network model includes a forget gate, an input gate, and an output gate. The forget gate portion is obtained from the input data and is used for control information. C t-1 The degree of forgetting is determined in the following ways: in, f t Indicates the output of the forget gate, [ s t-1 , x t [Indicates the hidden state at the previous moment] s t-1 Input at the current time x t splicing, w f Here is the weight matrix for the forget gate. b f For the bias term of the forget gate; The input gate section is used to control the degree of information updating, and is represented as follows: in, i t Indicates the output of the input gate. b i This is a bias term for the input gate memory unit value; w i This represents the weight matrix of the input gate; Current new LSTM state c t Based on the state at the previous moment c t-1 Represented as: in, w c This represents the weight matrix associated with the candidate memory cell values. b c The bias term for the input gate and candidate memory cell values ​​is passed through the output gate. o t Control information output, the unit output is represented as: in, w o and b o These are the weight matrix and offset vector of the aforementioned gates. σ and tanh Represents the sigmoid and hyperbolic tangent activation functions; A backward layer was added to the forward layer of the neural network model, and the forward and backward hidden layer vectors were concatenated. The mathematical expression is as follows: in, h← t , h→ t and O t They represent t The final outputs of the forward layer, backward layer, and output layer at each moment. f It is a network unit. g It's ReLU. w It's the weight.

4. The method for identifying micro-milling machining parameters considering the influence of random tool wear according to claim 1, characterized in that, The process of using a particle filtering algorithm to identify processing parameter values ​​in a neural network model specifically includes: For multiple cutting force coefficients in the neural network model, a particle filtering algorithm is used to identify the parameter values ​​at any given time. Assume the dynamics of the system using the state equation and observation equation: in, x k It is time t k The system status at that point, y k It is with time t k state of time x k The corresponding observed values, z k-1 and o k These represent process noise and observation noise, respectively. f Represents the current and previous system states. h This represents the relationship between the system state and the current time measurement. The obtained posterior probability density is substituted into the next prediction calculation to form a recursive state update: in, P ( x k | y 1:k-1 ) is the prior probability. P ( y k | x 1:k ) is the posterior probability. P ( y k | x k ) is the likelihood function. P ( y k | y k-1 ) is the normalization constant; Introducing the Monte Carlo method to replace the integration process with the average value, the function... f ( x The expected value can be approximately estimated from the known sampling distribution. Q ( x t | y 1:t Sampling in ) and solving f ( x The expected value problem is calculated as follows: in, w k These are normalized weights, calculated recursively using ordinal importance sampling. The recursive form of the particle weight value is expressed as: The particle swarm that caused particle degradation was resampled, and the degree of particle degradation was determined by the number of effective particles. N eff To measure: in, N The preset effective sample count threshold; Considering the effects of chatter in the machining system and tool runout, the time index is... t Considered as variables in the wear process modeling; the obtained wear prediction model is integrated into the mechanical modeling process, and based on the established cutting force model, the future cutting force prediction state equation and observation equation, including tool wear, are expressed as: in, maxF xk , maxF yk , rmsF xk and rmsF yk Representing the observed maximum cutting force and x shaft and y The root mean square value of the cutting force in the axial direction v ns , v np , v fs , v fp , v nv and v fv It is Gaussian distributed process noise. f k It is observation noise; The cutting coefficient value is derived based on the measured data and the above equations, so that the values ​​of each parameter of the model are fully obtained.

Images