A tool wear state monitoring method and system considering physical processes

Abstract

The application discloses a tool wear state monitoring method and system considering physical processes. In view of the problem that existing deep learning models for predicting tool wear states do not consider cutting physical processes and have high requirements for the quality and quantity of training data, a deep learning model is constructed, which combines a tool cutting force model to optimize the weight and threshold value in the training process. The method considers the cutting physical process, can accurately and real-timely monitor the tool wear state based on the cutting force signal of the cutting process, and the prediction result has strong interpretability and low dependence on the number and quality of samples in the data set.

Description

Technical Field

[0001] This invention relates to the field of metal cutting technology, and in particular to a method and system for monitoring the wear condition of cutting tools used on weakly rigid parts. Background Technology

[0002] In high-precision milling of weakly rigid parts, the cutting tool is a key factor affecting machining quality. Tool condition monitoring can monitor the wear status of the tool in real time, promptly detect abnormal wear, and is crucial for reducing the risk of tool breakage. It can predict the remaining tool life, provide a reference for production planning, and avoid production downtime or similar losses caused by sudden tool failure. Furthermore, it can maintain machining accuracy and maximize tool life by providing corrective measures for tool wear.

[0003] Numerous and uncertain factors influence tool wear during milling, making it difficult to establish accurate physical models to describe tool performance degradation. Furthermore, the effectiveness of purely data-driven deep learning-based methods relies on the quantity and quality of training samples, lacking consideration of the physical processes of tool wear, resulting in poor interpretability of predictions. Summary of the Invention

[0004] The technical problem to be solved by this invention is: to address the difficulty in simulating existing purely physical models, and the problem that purely data-driven deep learning models have high requirements for the quantity and quality of data samples and poor interpretability, this invention proposes a tool wear state monitoring method and system that considers physical processes, so as to efficiently and accurately monitor tool wear state in real time with a small number of cutting data samples.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] On the one hand, this application provides a method for monitoring tool wear condition that takes into account physical processes, characterized by including the following steps:

[0007] S1: Using the same cutting parameters, collect the actual wear values ​​of the flank face of multiple tools during each wear process, and simultaneously collect the corresponding cutting force signals to construct a cutting dataset;

[0008] S2: Calculate the correlation between cutting force signal and tool wear value, and screen tool wear sensitive features in time domain and frequency domain;

[0009] S3: Construct a physical model of cutting forces that takes into account tool wear;

[0010] S4: Building a deep learning model;

[0011] S5: Construct the tool wear value loss function and the sensitive feature loss function;

[0012] S6: Construct a total loss function that includes the two loss functions from S5;

[0013] S7: Input the cutting force signal from S1 into the deep learning model constructed in step S4 to obtain the tool wear prediction value;

[0014] S8: Extract the sensitive features of the cutting force signal input in S7;

[0015] S9: Calculate the loss function between the predicted tool wear value in S7 and the actual wear value in S1;

[0016] S10: Combining the cutting parameters in S1 with the tool wear prediction value obtained in S7, the cutting force simulation signal is obtained through the cutting force physical model that considers tool wear, and the sensitive features in the cutting force simulation signal are extracted.

[0017] S11: Calculate the loss function between the sensitive features in S8 and S10;

[0018] S12: Calculate the total loss function;

[0019] S13: Repeat steps S7 to S12 using 70% of the cutting dataset from S1, until the physical model of cutting force considering tool wear converges, and obtain the optimal tool wear state monitoring model.

[0020] S14: Input the cutting force signal to be monitored into the optimal tool wear condition monitoring model to obtain the tool wear prediction value that takes into account the physical process.

[0021] As an optional implementation, in step S1, the tool back face wear value is obtained by taking pictures and marking the marks using an industrial camera.

[0022] In some embodiments, the method further includes: using the remaining 30% of the cutting dataset in S1 to verify the accuracy of the monitoring model.

[0023] As an optional implementation, the correlation calculation in step S2 can use the Pearson correlation calculation method to screen the sensitive feature variables of tool wear in the cutting force signal.

[0024] In the above technical solution, the cutting force physical model considering tool wear mentioned in step S3 refers to the cutting force physical model that incorporates a friction effect force model. The shearing force generated by the shearing action of the rake face, and the friction and pressure generated by the wear of the flank face together constitute the cutting force of the worn tool. Among them, the friction and clamping force generated by the wear of the flank face are collectively referred to as the friction effect force, and the expression is as follows:

[0025]

[0026] In the formula, dFt To account for the resultant force of the tangential infinitesimal element after tool wear, dF r To consider the radial resultant force of the infinitesimal element after tool wear; where the tangential frictional force dF tw and radial infinitesimal pressure dF rw The calculation method is as follows:

[0027]

[0028] In the formula, F tw (VB) is the tangential frictional force per unit cutting length on the flank face, F rw (VB) represents the radial pressure per unit width of the flank face, which is a function of the actual wear value of the flank face, VB.

[0029] Tangential frictional force F per unit cutting edge length on the back face tw (VB) and radial pressure F rw (VB) is as follows:

[0030]

[0031] Where x is the blade length distance, VB * Let τ0 be the fixed width of the elastic contact region, and σ0 be the tangential stress and normal stress, respectively.

[0032] Preferably, the physical model of cutting force considering tool wear described in S3 is selected according to different working conditions.

[0033] As an optional implementation, the physical model of cutting force considering tool wear described in S3 can be expressed as a tapered helical tool cutting force model considering tool wear, as follows:

[0034]

[0035] In the formula, F x F y F z These represent the cutting forces in the X, Y, and Z directions during side milling with a graduated helical tool; D is the tool diameter; N is the number of tool teeth / cutting edges; and the helix angle of the tool side cutting edge is β = {β1, β2, ..., β...}. i , ..., β N}, where 1≤i≤N, and the helix angle of the i-th cutting edge is β. i And β imin ≤β i ≤β imax , where β imin With β imax These are the helix start angle and helix end angle of the gradual helix angle of the i-th cutting edge; the axial depth of cut is a. p ;β iap For the tool axis ap The helix angle of the i-th cutting edge at point φ; i K represents the position angle of the i-th cutting edge element at time t; tci K rci K aci These are the tangential, radial, and axial shear force coefficients of the cutting micro-element, respectively; K tei K rei K aei These are the tangential, radial, and axial plowing force coefficients of the cutting micro-element, respectively; h D (φ i ) represents the cutting thickness of the i-th cutting edge element at time t, in mm; g(φ) i u(β) is the window function for determining whether to perform a cut; i ) is the function representing the gradual change law of the helix angle of the gradually changing helix tool.

[0036] As an optional implementation, the deep learning model in step S4 can be a common deep learning model such as CNN, LSTM, GRU, BiGRU, or a deep learning model that combines two or more models such as CNN-Transformer.

[0037] In the above technical solution, the tool wear value loss function in step S5 refers to the loss function between the predicted tool wear value and the actual tool wear value. The Adam optimizer is used to minimize the loss function, and L2 regularization is introduced. The expression is as follows:

[0038]

[0039] In the formula, m is the signal length; y is the predicted value of tool flank wear; i λ is the actual value of tool flank wear; λ is the regularization factor; w is the weight vector;

[0040] The sensitive feature loss function refers to the loss function between the sensitive features of the simulated cutting force signal and the experimental cutting force signal obtained by the cutting force model. The expression is as follows:

[0041]

[0042] In the formula, m is the signal length; k is the number of sensitive features; j is the feature index, and 1≤j≤k; X is the predicted value of the cutting force signal sensitivity feature given by the cutting force model; X is the true value of the corresponding experimental cutting force signal sensitivity feature; λ is the regularization factor; w is the weight vector.

[0043] Preferably, the total loss function in S6 is a function with the tool wear value loss function and the sensitive feature loss function in S5 as independent variables.

[0044] Preferably, the sensitive feature described in S8 is the tool wear sensitive feature obtained in S2.

[0045] As an optional implementation, in S10, the cutting force physical model can be calculated using MATLAB programming to obtain the cutting force simulation signal.

[0046] Based on the above method, this invention provides a tool wear condition monitoring system that considers physical processes, comprising:

[0047] Signal acquisition module: configured to acquire the cutting force signal of the tool to be monitored;

[0048] Model training module: configured to optimize the weights and thresholds of the deep learning model by combining the cutting force physical model;

[0049] Prediction module: configured to input the cutting force signal of the tool into a trained deep learning model to obtain the predicted value of tool wear.

[0050] The specific implementation of the modules in this embodiment, as described above in the tool wear state monitoring method that considers physical processes, will not be described in detail here.

[0051] Compared with the prior art, the present invention has the following main advantages:

[0052] Compared with existing technologies, this invention addresses the problems of current deep learning models for predicting tool wear conditions, which do not consider the physical process of cutting and have high requirements for the quality and quantity of training data. It constructs a deep learning model that combines a physical model of tool cutting forces with the optimization of weights and thresholds during training. This significantly improves the interpretability of the prediction results, exhibits high correlation with tool geometry parameters and workpiece materials, and makes the results verifiable.

[0053] This method considers the physical process of cutting and can accurately monitor the tool wear state in real time based on the cutting force signal during the cutting process. The prediction results are highly interpretable and have low dependence on the quantity and quality of samples in the dataset. It can achieve accurate real-time monitoring of tool wear state based on cutting force signal even with a small number of samples.

[0054] Finally, this invention overcomes the limitations of the two existing methods and proposes a tool wear condition monitoring method based on digital-analog linkage. This method provides real-time and accurate understanding of the wear condition of specialized tools, enabling early warning of tool failures, improving machining efficiency, and maximizing tool life and utilization.

[0055] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 The flowchart and system architecture diagram of the tool wear condition monitoring method considering physical processes are shown in this invention.

[0058] Figure 2 This describes the interaction and linkage process between the deep learning model and the physical model in this invention.

[0059] Figure 3 This is the result of the Pearson correlation calculation between the flank wear value of the tool and the characteristic cutting force signal of the present invention.

[0060] Figure 4 This is the tool wear prediction result based on a deep learning model, which does not consider physical processes in this invention.

[0061] Figure 5 The tool wear prediction results based on a deep learning model take into account the physical processes of the present invention.

[0062] Figure 6 This paper compares the predictive performance of the four tool condition monitoring models of this invention based on two evaluation indicators, MAE and RMSE. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0064] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0065] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0066] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this application is in use. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In addition, the terms "first," "second," and "third," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0067] In the description of this application, it should also be noted that, unless otherwise expressly stated and limited, the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0068] 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 exemplary embodiments according to this application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. Without conflict, embodiments and features in the embodiments of this invention can be combined with each other.

[0069] Terminology Explanation:

[0070] Gradient helical cutting tools: These are cutting tools where the helix angle of any cutting edge is not a constant value, but rather gradually changes according to a certain pattern with the axial dimension of the tool. For details, please refer to the patents (publication numbers: CN116079124A and CN116140677A).

[0071] Tool wear value (VB): Since the wear on the flank face is relatively easy to measure, in metal cutting research, the wear standard is often formulated based on the maximum allowable average wear in the middle part of the wear band on the flank face (usually expressed as VB).

[0072] Cutting force: refers to the equal and opposite cutting forces generated during the cutting process that act on the workpiece and the cutting tool. In other words, it is the resistance generated by the workpiece material against the cutting action of the tool during machining.

[0073] The physical model of cutting force is a mathematical model that combines the physical process of tool cutting to describe the mechanical relationships in the cutting process and is used to predict the cutting force generated during the cutting process.

[0074] Deep learning models: computational models based on machine learning and artificial neural networks that can simulate the neural network structure and learning methods of the human brain, and are used to solve various complex data analysis and pattern recognition problems.

[0075] The features and performance of this application will be further described in detail below with reference to the embodiments.

[0076] Example 1

[0077] like Figure 1-2 To address the poor simulability of purely physical models and the issues of existing deep learning models requiring large quantities and high-quality data samples with poor interpretability, this embodiment proposes a tool wear condition monitoring method that considers physical processes. This method can efficiently and accurately monitor tool wear conditions in real time with relatively small cutting data samples, and specifically includes the following steps:

[0078] S1: Using the same cutting parameters, collect the flank wear value VB of multiple tools during each wear process, and simultaneously collect the corresponding cutting force signal to construct a cutting dataset;

[0079] S2: Calculate the correlation between cutting force signal and tool wear value, and screen tool wear sensitive features in time domain and frequency domain;

[0080] S3: Construct a physical model of cutting forces that takes into account tool wear;

[0081] S4: Building a deep learning model;

[0082] S5: Construct the tool wear value loss function and the sensitive feature loss function;

[0083] S6: Construct a total loss function that includes the two loss functions from S5;

[0084] S7: Input the cutting force signal from S1 into the deep learning model constructed in step S4 to obtain the tool wear prediction value;

[0085] S8: Extract the sensitive features of the cutting force signal input in S7;

[0086] S9: Calculate the loss function between the predicted tool wear value in S7 and the actual tool wear value in S1;

[0087] S10: Combining the cutting parameters in S1 and the tool wear prediction values ​​obtained in S4, the cutting force simulation signal is obtained by considering the cutting force physical model of tool wear, and the sensitive features in the cutting force simulation signal are extracted.

[0088] S11: Calculate the loss function between the sensitive features in S8 and S10;

[0089] S12: Calculate the total loss function;

[0090] S13: Repeat steps S7 to S12 using 70% of the cutting dataset from S1 until the model converges to obtain the optimal tool wear condition monitoring model.

[0091] S14: Input the cutting force signal to be monitored into the optimal tool wear condition monitoring model to obtain the tool wear prediction value that takes into account the physical process.

[0092] In step S1, the cutting parameters are set as follows: cutting speed of 131.88 m / min, feed per tooth of 0.02 mm / z, axial depth of cut of 20 mm, radial depth of cut of 0.1 mm, multiple tools are set as 3 solid carbide graduated spiral end mills, and the workpiece is set as a weakly rigid part such as a titanium alloy (Ti-6Al-4V) frame beam. The specific process of obtaining the tool flank wear value includes: using a Daheng M231 industrial camera including a CCD industrial camera, focusing lens, adjustable LED aperture, camera bracket and power supply to photograph the tool flank wear, and collecting the maximum value VB of the flank wear of each tool at different cutting distances in the initial wear, normal wear and rapid wear stages.

[0093] The cutting force signal includes two directions, X and Y, which are sensitive to weakly rigid parts: cutting force F x F y The specific equipment for acquiring cutting force signals includes a hardware system and a software system. The hardware system specifically includes a KISTLER 9139A triaxial piezoelectric force gauge and a DH5922 data acquisition box. The software system specifically refers to Kistler's DynoWare 3.2.2.0-1.0 software. The cutting force signal acquisition frequency is 10kHz. The cutting force signal F is constructed... x F y A cutting dataset with a one-to-one correspondence between the tool wear value VB, totaling 60 sets of cutting data;

[0094] The correlation calculation in step S2 is set to use the Pearson correlation coefficient method to screen the sensitive feature variables of tool wear in the cutting force signal. The closer the absolute value of the Pearson correlation coefficient is to 1, the stronger the correlation between the two variables; the closer it is to 0, the weaker the correlation. The Pearson correlation coefficient method for screening feature variables first involves dimensionless initialization of the data:

[0095]

[0096] In the formula, i represents different feature variables; k represents the parameter values ​​of different feature variables;

[0097] The Pearson correlation coefficient between variables X and Y is expressed as follows:

[0098]

[0099] In the formula, l xx Let X be the sum of squared deviations from the mean; yy Let Y be the sum of squared deviations from the mean; xy Let X be the sum of the products of the deviations from the mean of variables X and Y.

[0100] Sum of squared deviations of X from the mean l xx satisfy:

[0101]

[0102] Sum of squared deviations of Y yy satisfy:

[0103]

[0104] The sum of the products of the deviations from the mean of variables X and Y xy satisfy:

[0105]

[0106] The above methods yield Pearson correlations between 23 commonly used features in both the time and frequency domains and the tool flank wear value. The results are as follows: Figure 3 As shown in a)~v).

[0107] Through correlation analysis, features with a correlation greater than 0.8 were extracted as sensitive features, including a total of 5 features: 4 time-domain features, namely a) mean, c) peak value, e) root mean square value, and g) waveform factor; and 1 frequency-domain feature, o) energy spectrum.

[0108] The feature domain, name, mathematical expression, and physical meaning of the tool wear sensitivity characteristics calculated based on the Pearson correlation coefficient are shown in the table below, where x i This represents the cutting force signal collected during a certain time period in the cutting process.

[0109]

[0110] In the above technical solution, the cutting force physical model considering tool wear in step S3 refers to the cutting force physical model that incorporates a frictional effect force model. The shearing force generated by the shearing action of the rake face and the frictional force and pressure generated by the wear of the flank face together constitute the cutting force of the worn tool. Among them, the frictional force and clamping force generated by the wear of the flank face are collectively referred to as the frictional effect force, and the expression is as follows:

[0111]

[0112] In the formula, dF t To account for the resultant force of the tangential infinitesimal element after tool wear, dF r To consider the radial resultant force of the infinitesimal element after tool wear; where the tangential frictional force dF tw and radial infinitesimal pressure dF rw The calculation method is as follows:

[0113]

[0114] In the formula, F tw (VB) is the tangential frictional force per unit cutting length on the flank face, F rw (VB) is the radial pressure per unit width of the flank face, which is a function of the flank face wear value VB;

[0115] Tangential frictional force F per unit cutting edge length on the back face tw (VB) and radial pressure F rw (VB) is as follows:

[0116]

[0117] Where x is the blade length distance, VB * Let τ0 be the fixed width of the elastic contact region, and σ0 be the tangential stress and normal stress, respectively.

[0118] In the above technical solution, the deep learning model is set as a CNN-Transformer model, combining the CNN model with the Transformer model. The CNN model consists of three layers of 1D-CNN, used to efficiently extract tool wear-sensitive features from the cutting force signal. Each layer of the CNN model has a convolution kernel size K=5, a padding loop number P=2, and a stride S=2. The number of convolution kernels in the first layer is F=64; in the second layer, F=128; and in the third layer, F=256. The convolution kernels slide upwards in a one-dimensional time series with a certain stride, thereby extracting local features of the cutting force signal. The length N of the convolutional sequence can be calculated as follows:

[0119]

[0120] The Transformer model is based on a multi-head attention mechanism and is a model with efficient global feature capture and strong parallel processing capabilities. The input of the model is the sensitive features output by the CNN model, and the output is the tool wear prediction value. The multi-head attention mechanism can effectively solve the attention bias problem that may exist in a single self-attention mechanism. It includes Matmul, Score, SoftMax and Matmul. The model principle is based on equations (10) to (13):

[0121] MultiHead(Q,K,V)=concat(head1,head2,…,head n W o (10)

[0122] head i =Attention(Q) i K i V i (11)

[0123]

[0124]

[0125] In the formula, Q, K, and V represent the three weight matrices for query, key, and value, respectively; X is the input matrix; i is the index of the attention heads, satisfying 1≤i≤m, where m is the number of heads in the multi-head attention mechanism, and in this paper, m=8. W i Q W i K and W i V Let be the trainable parameter matrices of the i-th head, and satisfy . d is a dimension and d k =d v =d model / n.

[0126] The input feature matrix X undergoes a linear transformation as shown in equation (10) above, resulting in a parameter matrix generated by matrix multiplication. The resulting trainable parameter matrices Q, K, and V are then input into a multi-head attention mechanism to enhance the model's fitting ability. Q and K are processed by MatMul to generate a similarity matrix. Each element of the similarity matrix is ​​divided by d. k The square root of the result is then normalized using the Softmax function, making each value a weight coefficient greater than 0 and less than 1.

[0127] In step S4, the Adam optimizer is used to minimize the loss function, and L2 regularization is introduced to obtain the following formula for the loss function between the predicted and actual tool wear values:

[0128]

[0129] In the formula, m is the signal length; y is the predicted value of tool flank wear; i λ is the actual value of the tool's flank wear; λ is the regularization factor; w is the weight vector.

[0130] The loss function obtained from the cutting force model regarding the sensitive characteristics of the simulated cutting force signal and the experimental cutting force signal is as follows:

[0131]

[0132] In the formula, j is the feature index, and 1≤j≤5; X represents the predicted value of the cutting force signal sensitivity feature given by the cutting force model; X represents the true value of the corresponding experimental cutting force signal sensitivity feature. λ is the regularization factor; w is the weight vector.

[0133] Preferably, in step S5, the total loss function is a function with the tool wear value loss function and the sensitive feature loss function from S4 as independent variables; let ξ be the weight of the loss function between the tool wear value and the true value. Let (1-ξ) be the weight of the loss function between the milling force simulation signal sensitive feature and the experimental signal sensitive feature. Therefore, the total loss function of the tool wear state monitoring model considering the physical process can be obtained as follows:

[0134] Loss=ξ*loss1+(1-ξ)*loss2 (16)

[0135] In the formula, ξ is the weight coefficient of the tool wear value loss function, and 0 < j < 1; loss1 is the loss function between the predicted value and the actual value of tool flank wear; loss2 is the comprehensive loss function between sensitive features related to tool flank wear value.

[0136] Preferably, the sensitive features mentioned in S7 are the five tool wear sensitive features obtained in S2;

[0137] In S9, the physical model for cutting force is set as the cutting force model for a gradually increasing helical tool, and the calculation formula is as follows:

[0138]

[0139] In the formula, F x F y F zThese represent the cutting forces in the X, Y, and Z directions during the cutting process of a graduated helical tool; l1 is the tool cutting edge length, N is the number of tool teeth / number of cutting edges; the helix angle of the tool side edge is β={β1, β2, …, β i , ..., β N}, where 1≤i≤N, and the helix angle of the i-th cutting edge is β. i And β imin ≤βi≤β imax , where β imin With β imax These are the helix start angle and helix end angle of the gradual helix angle of the i-th cutting edge; the axial depth of cut is a. p ;β iap For the tool axis a p The helix angle of the i-th cutting edge at point φ; i Let dF be the position angle of the i-th cutting edge element at time t; ti dF ri dF ai They are respectively micro elements u ij Tangential, radial and axial milling forces, N; K tci K rci K aci They are respectively micro elements u ij The tangential, radial, and axial shear force coefficients; K tei K rei K aei They are respectively micro elements u ij The tangential, radial, and axial plowing force coefficients; h D (φ i ) represents the cutting thickness of the i-th cutting edge element at time t, in mm; g(φ) i ) is a window function for determining whether to perform a cut;

[0140] In the above technical solution, the tool axis a p The helix angle β of the i-th cutting edge at point i iap satisfy:

[0141]

[0142] In the above technical solution, the instantaneous cutting thickness h of the j-th cutting micro-element on the i-th cutting edge D (φ i The length of ) can be expressed as:

[0143]

[0144] In the formula, D is the tool diameter (mm); the tool tooth angle is... is the angle between the i-th cutting edge and the (i-1)-th cutting edge (i.e., the i-th tooth angle); f is the feed rate, mm / r; n is the rotational speed, r / min;

[0145] In the above technical solution, the window function g(φ) i To determine whether a micro-element participates in cutting, the following expression is used:

[0146]

[0147] Where: φ st The milling approach angle is expressed in rad; φ ex The milling cut angle is given by φ in rad; and the two satisfy the relationship: 0 ≤ φ st <φ ex ≤π; Angle of approach φ st , tangent angle φ ex It can be represented as follows:

[0148]

[0149] In the formula: a e The cutting width is in mm.

[0150] In the above technical solution, the tangential, radial, and axial shear force coefficients K tci K rci K aci Satisfy the following formula:

[0151]

[0152] In the formula, τ s η is the shear yield strength, φ0 is the shear angle, α0 is the tool rake angle, ζ0 is the normal friction angle, and η is the shear yield strength. c This is the chip flow angle.

[0153] In the above technical solution, the tangential, radial, and axial plowing force coefficients K tei K rei K aei Satisfy the following formula:

[0154]

[0155] In the formula, r e Let θ be the radius of the blade tip circle. f Let θ be the splitting angle, and θ f =ζ0.

[0156] In the above technical solution, the physical model of the cutting force is set to obtain the cutting force simulation signal by MATLAB programming calculation.

[0157] During model training in steps S6 to S12, the parameters are set as follows: maximum number of iterations is 600, number of layers is 2, Dropout ratio is 0.25, and learning rate is 0.0001. Data is input into the model for iterative training. After each iteration, the model's predicted wear is input into the physical model to generate a simulated cutting force signal, which is then compared with the actual wear to calculate the loss function loss1. Simultaneously, the feature loss function loss2, which combines the sensitive features extracted from the simulated cutting force signal with those extracted from the real cutting force signal, is calculated. This total loss function loss is then used to update and optimize the weight thresholds of the CNN-Transformer model until the required number of iterations is reached and the loss function meets the requirements.

[0158] Regarding the weighting coefficients of the tool wear loss function, this invention uses ξ = 0.1 to 0.9 (incrementing by 0.1 each time) for trial calculations, and selects the optimal solution ξ = 0.6. The trained CNN-Transformer model can predict tool wear based on the cutting force signal. The cutting force model for a graduated spiral tool that considers tool wear is based on the cutting physical process of the graduated spiral tool, and has a high correlation with tool geometry parameters, workpiece material, etc., and the prediction results are highly interpretable.

[0159] This embodiment uses a pre-built gradient helical tool cutting dataset to verify the effectiveness of the tool wear condition monitoring model based on digital-analog linkage. To demonstrate the effectiveness of the F-CNN-Transformer model in tool wear prediction, its performance is compared with three other models: the traditional machine learning method Support Vector Regression (SVR), the deep learning method (Convolutional Neural Network (CNN)), and the CNN-Transformer hybrid model. To evaluate the performance differences between different networks and measure the model's prediction effect, Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used as performance evaluation metrics for this model, as shown in the following formulas:

[0160]

[0161]

[0162] For comparative analysis, the hyperparameter settings of the three models (the traditional machine learning method Support Vector Regression (SVR), the deep learning method (Convolutional Neural Network (CNN)), and the CNN-Transformer hybrid model) are the same as those of the F-CNN-TRANSFORMER model. All three models use MSE as the loss function and Adam as the optimization function, and are trained and validated in the same computing environment. The training sets are S1 and S2 from the established gradient helical tool cutting dataset, and the model performance is validated using S3.

[0163] All four models can capture the evolution trend of the wear process of the gradually helical tool, but the prediction results differ significantly. The prediction results of the better CNN-Transformer hybrid model among the four models are compared with those of this invention, such as... Figure 4 The figure shows the prediction results of the CNN-Transformer hybrid model. Due to the multi-head attention mechanism, the dimension of the input sequence is divided into multiple heads, and a different weight matrix is ​​defined for each head. This maps the input vector into multiple different subspaces, allowing the model to learn from more perspectives and acquire richer feature information. Long-distance dependencies can be obtained. As can be seen from the figure, the local fluctuations in the prediction results are relatively smooth, and the prediction accuracy is quite ideal.

[0164] After introducing a cutting force physical model to optimize the weights and thresholds, the prediction results are as follows: Figure 5 As shown, the prediction accuracy of the model is further improved after introducing the physical model. The fluctuation is significantly better than before. Figure 4 The small prediction results of the CNN-Transformer hybrid model demonstrate that incorporating the physical model into the deep model plays an important role in improving the accuracy of tool wear prediction.

[0165] The predictive performance of the above four tool condition monitoring models based on MAE and RMSE evaluation metrics is compared as follows: Figure 6 As shown in the figure, the F-CNN-Transformer model with the cutting force physical model exhibits the lowest MAE and RMSE errors, at 6.68% and 7.12%, respectively. This demonstrates that optimizing the weights and thresholds of the deep learning model based on the physical process by incorporating the cutting force physical model plays a crucial role in further improving the accuracy of tool wear prediction.

[0166] The present invention further cross-validated the F-CNN-Transformer model established in this invention on test sets S1 and S2, using S2 and S3 as the training set and S1 as the test set. The prediction errors of the four models are shown in Table 1. As can be seen from the data in Table 1, the F-CNN-Transformer model outperforms the other models in both MAE and RMSE metrics, demonstrating the good generalization ability of the proposed F-CNN-Transformer-based tool wear condition monitoring model.

[0167] Table 1 Comparison of Predictive Performance of Four Models

[0168]

[0169] In summary, the proposed F-CNN-Transformer deep learning model, which considers physical processes, can effectively obtain the correlation between cutting force signals and tool flank wear values ​​based on small sample experimental data. It directly learns locally and globally sensitive tool wear characteristics from preprocessed data, achieving high-precision tool wear state monitoring in the environment of tapered helical tools cutting beam-type parts. When training and test samples change, the model consistently demonstrates stable high-precision predictions and significant robustness.

[0170] Example 2

[0171] This embodiment provides a tool wear condition monitoring system that considers physical processes, including:

[0172] Signal acquisition module: configured to acquire the cutting force signal of the tool to be monitored;

[0173] Model training module: configured to optimize the weights and thresholds of the deep learning model by combining the cutting force physical model;

[0174] Prediction module: configured to input the cutting force signal of the tool into a trained deep learning model to obtain the predicted value of tool wear.

[0175] The implementation of the specific modules in this embodiment refers to the steps of the tool wear state monitoring method considering physical processes described in Embodiment 1, and will not be described in detail here.

[0176] Example 3

[0177] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the tool wear condition monitoring method considering physical processes described in Embodiment 1.

[0178] Example 4

[0179] This embodiment provides a computer-readable storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the tool wear state monitoring method considering physical processes described in Embodiment 1.

[0180] The embodiments described above are some, but not all, of the embodiments of this application. The detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. A tool wear state monitoring method considering a physical process, characterized by Includes the following steps: S1: Using the same cutting parameters, collect the actual wear values ​​of the flank face of multiple tools during each wear process, and simultaneously collect the corresponding cutting force signals to construct a cutting dataset; S2: Calculate the correlation between cutting force signal and tool wear value, and screen tool wear sensitive features in time domain and frequency domain; S3: Constructing a physical model of cutting forces considering tool wear: A cutting force model for a tapered helical tool considering tool wear is adopted, expressed as follows: In the formula, F x , F y , F z During the side milling process with a gradient helical tool X , Y , Z Directional cutting force; D The diameter of the cutting tool; N The tool's number of teeth / number of cutting edges; the tool's side helix angle is... β ={ β 1, β 2, ..., β i , ..., β N }, where 1≤ i ≤ N And the first i The helix angle of the cutting edge is β i and β imin ≤ β i ≤ β imax ,in β imin and β imax The first i The helix start angle and helix end angle of the gradually changing helix angle of the cutting edge; the axial depth of cut is a p ; β iap For the tool axis a p The first i Helix angle of the cutting edge; i for t Time of the first i The position angle of the micro-element of the cutting edge; K tci , K rci , K aci These are the tangential, radial, and axial shear force coefficients of the cutting micro-element, respectively; K tei , K rei , K aei These are the tangential, radial, and axial plowing force coefficients of the cutting micro-element, respectively; h D ( i )for t Time of the first i The cutting thickness of the micro-element of the cutting edge, in mm; g ( i () is a window function for determining whether to perform a cut; u ( β i () represents the function governing the gradual change of the helix angle of a gradually changing helical cutting tool; S4: Building a deep learning model; S5: Construct the tool wear value loss function and the sensitive feature loss function; S6: Construct a total loss function that includes the two loss functions from S5; S7: Input the cutting force signal from S1 into the deep learning model constructed in step S4 to obtain the tool wear prediction value; S8: Extract the sensitive features of the cutting force signal input in S7; S9: Calculate the loss function between the predicted tool wear value in S7 and the actual wear value in S1; S10: Combining the cutting parameters in S1 with the tool wear prediction value obtained in S7, the cutting force simulation signal is obtained through the cutting force physical model that considers tool wear, and the sensitive features in the cutting force simulation signal are extracted. S11: Calculate the loss function between the sensitive features in S8 and S10; S12: Calculate the total loss function; S13: Repeat steps S7 to S12 using 70% of the cutting dataset from S1, until the physical model of cutting force considering tool wear converges, and obtain the optimal tool wear state monitoring model. S14: Input the cutting force signal to be monitored into the optimal tool wear condition monitoring model to obtain the tool wear prediction value that takes into account the physical process.

2. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... The wear value of the tool's flank face is obtained by taking pictures with an industrial camera and marking them.

3. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... The correlation calculation in step S2 can use the Pearson correlation calculation method to screen the sensitive feature variables of tool wear in the cutting force signal.

4. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... The deep learning model in step S4 is one of CNN, LSTM, GRU, or BiGRU, or a deep learning model that combines two or more models, such as CNN-Transformer.

5. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... In step S5, the tool wear loss function refers to the loss function between the predicted tool wear value and the actual tool wear value. The Adam optimizer is used to minimize the loss function, and L2 regularization is introduced. The expression is as follows: In the formula, m The signal length; This is the predicted value for the wear of the tool's flank face; λ represents the actual wear value of the tool's flank face; λ is the regularization factor. This is the weight vector; The sensitive feature loss function refers to the loss function between the sensitive features of the simulated cutting force signal and the experimental cutting force signal obtained by the cutting force model. The expression is as follows: In the formula, m The signal length; k The number of sensitive features; j The feature number is 1 ≤ j ≤ k ; The predicted values ​​of the cutting force signal sensitivity features given by the cutting force model; X This represents the true value of the sensitive characteristic of the corresponding experimental cutting force signal; λ is the regularization factor; This is the weight vector.

6. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... The total loss function mentioned in step S6 is a function with the tool wear value loss function and the sensitive feature loss function in S5 as independent variables.

7. The tool wear condition monitoring method considering physical processes according to claim 1, characterized in that... The sensitive feature mentioned in step S8 is the tool wear sensitive feature obtained in S2.

8. A tool wear condition monitoring system that considers physical processes, comprising: Signal acquisition module: configured to acquire the cutting force signal of the tool to be monitored; Model training module: configured to optimize the weights and thresholds of the deep learning model by combining the cutting force physical model; Prediction module: configured to input the tool cutting force signal into a trained deep learning model to obtain tool wear prediction values; Each module is used to implement the steps of the tool wear condition monitoring method considering physical processes as described in any one of claims 1-7.

Images