Cutting tool wear monitoring method

Abstract

The invention relates to a cutting tool wear monitoring method, which comprises the following steps that: signals # imgabs0 # in multiple directions in the cutting process of a tool are collected, and the signals comprise cutting force, vibration and sound signals; the abrasion loss # imgabs1 # of the tool after cutting is completed is collected, the signal # imgabs2 # and the abrasion loss # imgabs3 # are spliced, and a state matrix x and an abrasion target value vector c are obtained; performing data interception on the state matrix x by adopting a sliding window to obtain M data segments # imgabs4 #; the time domain feature of the data segment # imgabs5 # is calculated, and # imgabs6 # is obtained; interpolation is carried out on the target value vector c, and the abrasion loss # imgabs7 # corresponding to each data segment is obtained; and a # imgabs8 # and a # imgabs9 # are used for training a neural network, and a tool wear monitoring model is constructed. According to the method, the wear of the cutter can be monitored only by inputting the current cutting data and the prior distribution parameters of the wear loss, and the interpretability of the model is improved by adopting a codec structure.

Description

Technical Field

[0001] The invention belongs to the technical field of metal cutting, and in particular relates to a cutting tool wear monitoring method. Background Art

[0002] Tool wear in metal cutting machining refers to the gradual degradation of the drill, end mill and insert of the tool as it interacts with the workpiece during machining. Tool wear monitoring is a key aspect of modern machining processes and plays a key role in ensuring production efficiency, product quality and cost-effectiveness. By monitoring tool wear, tool conditions can be evaluated dynamically and in real time, tool life can be extended, and machining parameters and tool change schedules can be optimized.

[0003] At present, tool wear monitoring is mainly divided into direct method and indirect method. The direct method refers to the use of optical instruments to collect tool wear images for wear monitoring. Although this method can provide higher monitoring accuracy, it requires interrupting the processing process for image acquisition. At the same time, the processing environment will cause a lot of interference and noise, and continuous monitoring cannot be achieved. The indirect method refers to evaluating the wear of the tool by analyzing various sensor signals. Sensor signals include: cutting force, vibration, sound, motor power, etc. The indirect method can continuously and real-time collect signals during the processing process without interrupting the processing process.

[0004] The current indirect method mainly includes three stages: feature extraction, feature selection and model construction. For example, the Chinese invention patent with publication number CN116638374A discloses a tool wear state recognition method based on feature selection and sensor combination, which includes the following steps: combining three sensors that generate cutting force, vibration and acoustic emission signals to obtain seven sensor combinations; collecting multi-sensor signals during processing; intercepting the signal of the intermediate stable milling stage, and processing the abnormal value and environmental interference noise signal to obtain the feature subset ①; eliminating the features with weak correlation with tool wear, and analyzing the correlation between the features to obtain the feature subset ③ that eliminates redundant features; dividing into training set and test set, respectively input into the forest to construct a tool wear state recognition model, and identify the tool wear state. However, this method has the following disadvantages: the number of tool wear value samples in actual operation is limited, and the tool wear value increases with the use time. The above method ignores the relationship between tool wear values ​​at different time nodes, resulting in inaccurate trained models. Summary of the invention

[0005] In view of the above-mentioned technical problems, the present invention provides a cutting tool wear monitoring method, aiming to solve the problem of low accuracy in tool wear monitoring.

[0006] To solve the above problems, the present invention provides a cutting tool wear monitoring method, which comprises the following steps: Step S1: Collect signals from multiple directions during tool cutting , the signals include cutting force, vibration and sound signals; the directions are x-axis, y-axis and z-axis directions based on the machining coordinate system; Step S2: Collect the wear amount of the tool after cutting is completed , the signal and wear Perform splicing to obtain the state matrix x and the wear target value vector c; Step S3: Use a sliding window to intercept the state matrix x and obtain M data segments ; Calculate the data segment The time domain characteristics are obtained ; Interpolate the target value vector c to obtain the wear amount corresponding to each data segment ; Step S4: Use , Train a neural network and build a tool wear monitoring model.

[0007] Furthermore, the Data for seven channels ,in, is the cutting force signal, is the vibration signal, For sound signal.

[0008] Furthermore, the The initial value of is 0.

[0009] Furthermore, the calculation of the data segment The time domain characteristics are obtained Specifically: vertically calculate the data statistics of each channel and get the maximum value , mean , covariance ,in The index value of the channel.

[0010] Furthermore, the use , The specific calculation formula for training the neural network is: ,in is the tool wear monitoring value at the predicted time t, is a neural network, Q is the model encoding length, The amount of wear The prior distribution parameters of .

[0011] Furthermore, the wear amount The prior distribution parameters of are as follows: in, The amount of wear The mean of The amount of wear The covariance of .

[0012] Furthermore, the tool wear monitoring model includes: a gated residual network GRN, a feature selection network VSN, a static skew variance vector encoder SCE, a cyclic codec LSTMEnDe, and a time fusion decoder TFD.

[0013] Furthermore, the wear The prior distribution parameter s is passed through the feature selection network VSN to obtain the selected features , wear amount The prior distribution parameter s is passed through the static oblique variance vector encoder SCE to obtain the output encoded feature vector , , and .

[0014] Furthermore, the input feature of the recurrent encoder-decoder LSTMEnDe is and Features and , output the encoded time series features .

[0015] Furthermore, the time fusion decoder TFD is used to decode the encoded time series features Combination Perform data enhancement; use the time-domain self-attention multi-head mechanism to calculate attention and weights, and then perform feature calculation and fusion according to the weights; perform gated calculation, skip-layer connection and normalization on the fused features, and perform element-level feedforward calculation through the gated residual network GRN, and finally output the tool wear monitoring value.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) The method proposed in this patent does not require previous wear data. It only needs to input the current cutting data and the prior distribution parameters of the wear amount to monitor the wear of the tool.

[0017] (2) The method proposed in this patent uses a codec structure to improve the interpretability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0019] Figure 1 is a block diagram of the cutting tool wear monitoring method of the present invention; Figure 2 It is a feature extraction process block diagram of the present invention; Figure 3 This is a block diagram of the feature selection process of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the specific embodiments described herein are only used to explain the present invention and are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0021] Example 1

[0022] like Figure 1-Figure 3 As shown, an embodiment of the present invention relates to a cutting tool wear monitoring method, the method comprising: The present invention provides a cutting tool wear monitoring method, the method comprising the following steps: Step S1: Collect signals from multiple directions during tool cutting , the signals include cutting force, vibration and sound signals; the directions are x-axis, y-axis and z-axis directions based on the machining coordinate system; Step S2: Collect the wear amount of the tool after cutting is completed , the signal and wear Perform splicing to obtain the state matrix x and the wear target value vector c; Step S3: Use a sliding window to intercept the state matrix x and obtain M data segments ; Calculate the data segment The time domain characteristics are obtained ; Interpolate the target value vector c to obtain the wear amount corresponding to each data segment ; To ensure the accuracy of the model, feature extraction and feature selection are crucial stages. Figure 2-Figure 3As shown in the figure, during the tool cutting process, sensors are used to collect cutting force signals in three directions. , vibration signal , sound signal , thus forming a 7-channel signal , Corresponding to the N sets of data collected, that is, The tool performs multiple cuts, and after each cut, the wear of the tool edge is measured. During the K cutting processes, K groups of data are collected. ,Right now .

[0023] The entire life cycle of the tool is regarded as a complete time series, and K groups of collected data are and wear , spliced ​​along the time dimension to obtain the state matrix x and the wear target value vector c.

[0024] Using a sliding window, the state matrix x is intercepted according to the interval wid, and a total of M data segments are intercepted. . Interpolate the target value vector c to obtain the wear amount corresponding to each data segment Because a new tool is used at the beginning of cutting, the initial value of the wear interpolation is 0. Thus, all the processed data segments are obtained. and wear The corresponding relationship between . Each data segment after processing , contains wid pieces of 7-channel data. Calculate the data statistics of each channel vertically and get the maximum value , mean , covariance ,in is the index value of the channel. After the above feature selection, the data segment is transformed into a 3*7 channel data , corresponding to the wear amount , forming a new corresponding relationship, namely .

[0025] Step S4: Use , Train a neural network and build a tool wear monitoring model.

[0026] This step is the model construction process, including: gated residual network GRN, feature selection network VSN, static oblique variance vector encoder SCE, cyclic codec LSTMEnDe, and time fusion decoder TFD. The gated residual network GRN is the basic unit of the entire network. It controls the data on and off in the neural network and the degree of nonlinearity of the model by selectively allowing and blocking data to pass. The feature selection network VSN is established based on the gated residual network GRN, and is used to dynamically select the most relevant features in each time step. The static oblique variance vector encoder SCE is established based on the gated residual network GRN to process time-independent features. The cyclic codec LSTMEnDe is used to calculate and characterize the changing characteristics of data within a local time period. The model implementation process includes the main steps of generating context information, generating cyclic codec features, and generating time fusion features, as follows: Generate context information, wear amount The prior distribution parameter s is passed through the feature selection network VSN to obtain the selected features , wear amount The prior distribution parameter s is passed through the static oblique variance vector encoder SCE to obtain the output encoded feature vector , , and .

[0027] Generate cyclic encoder and decoder features, cyclic encoder and decoder LSTMEnDe input features and Features and , output the encoded time series features Specifically: ① The sensor collects the original data signal; ② Perform feature extraction and feature selection to obtain the processed data signal ; ③ According to the set model encoding length , input data segment sequence: ④Before Each data segment is processed in turn by its corresponding feature selection network VSN, combined with , output features: ⑤The data segment that needs to be monitored currently , processed by the corresponding feature selection network VSN, and the output feature ⑥Will and As the first data segment feature The initial memory state of the LSTM unit, input the first data segment feature , calculated by LSTM unit, and the output unit outputs data and memory data; ⑦ The memory data output from the previous step is used as the second data segment feature The initial memory state of the LSTM unit, and then input the second data segment feature , calculated by LSTM unit, and the output unit outputs data and memorize data;⑧ Calculate in sequence until the output Data segment features The output data of the LSTM unit ; ⑨ Output data of each LSTM unit , through gated calculation, skip-layer connection data fusion and normalization processing, the processed data is output , the process is expressed by the formula: ⑩ This is the generated cyclic encoding and decoding feature.

[0028] Generate time fusion features, and the time fusion decoder TFD is used to encode the time series features Combination Perform data enhancement; use the time-domain self-attention multi-head mechanism to calculate attention and weights, and then perform feature calculation and fusion based on the weights; perform gated calculation, skip-layer connection and normalization on the fused features, and perform element-level feedforward calculation through the gated residual network GRN, and finally output the tool wear monitoring value. Specifically: ① Cyclic encoding and decoding features Through the corresponding gated residual network GRN in the sequence, combined with contextual information , perform data enhancement; ② Use the time-domain self-attention multi-head mechanism to perform attention calculation and weight calculation on features of different step lengths; ③ Perform feature calculation and fusion according to the weights; ④ Process data through gated calculation, skip layer connection and normalization; ⑤ Perform element-level feedforward calculation through gated residual network GRN; ⑥ Process data through gated calculation, skip layer connection and normalization; ⑦ Fully connected layer regression outputs the current tool wear monitoring value The above process can be summarized as follows: Model building and training of neural network , that is, input wear amount The prior distribution parameters of and , output the predicted wear amount at time t Among them, Q is the coding length of the model prediction, and the wear amount The prior distribution parameters of Including model encoding length Q, wear amount The mean ,variance and index number. ,in The method proposed in this patent does not require previous wear data. It only needs to input the current cutting data and the prior distribution parameters of the wear to monitor the wear of the tool, and uses a codec structure to improve the interpretability of the model.

[0029] The above embodiments only express the preferred implementation mode of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A cutting tool wear monitoring method, characterized in that: The following steps are involved: Step S1: Collect signals from multiple directions during tool cutting , the signals include cutting force, vibration and sound signals; the directions are x-axis, y-axis and z-axis directions based on the machining coordinate system; Step S2: Collect the wear amount of the tool after cutting is completed , the signal and wear Perform splicing to obtain the state matrix x and the wear target value vector c; Step S3: Use a sliding window to intercept the state matrix x and obtain M data segments ; Calculate the data segment The time domain characteristics are obtained ; Interpolate the target value vector c to obtain the wear amount corresponding to each data segment ; Step S4: Use , Train a neural network and build a tool wear monitoring model.

2. The cutting tool wear monitoring method according to claim 1, characterized in that: Said Data for seven channels ,in, is the cutting force signal, is the vibration signal, For sound signal.

3. The cutting tool wear monitoring method according to claim 1, characterized in that: Said The initial value of is 0.

4. The cutting tool wear monitoring method according to claim 2, characterized in that: The calculating of the data segment The time domain characteristics are obtained Specifically: vertically calculate the data statistics of each channel and get the maximum value , mean , covariance ,in The index value of the channel.

5. The cutting tool wear monitoring method according to claim 4, characterized in that: The use , The specific calculation formula for training the neural network is: ,in is the tool wear monitoring value at the predicted time t, is a neural network, Q is the model encoding length, The amount of wear The prior distribution parameters of .

6. The cutting tool wear monitoring method according to claim 5, characterized in that: The amount of wear The prior distribution parameters of are as follows: in, The amount of wear The mean of The amount of wear The covariance of .

7. The cutting tool wear monitoring method according to claim 6, characterized in that: The tool wear monitoring model includes: a gated residual network GRN, a feature selection network VSN, a static oblique variance vector encoder SCE, a cyclic codec LSTMEnDe, and a time fusion decoder TFD.

8. The cutting tool wear monitoring method according to claim 7, characterized in that: Wear The prior distribution parameter s is passed through the feature selection network VSN to obtain the selected features , wear amount The prior distribution parameter s is passed through the static oblique variance vector encoder SCE to obtain the output encoded feature vector , , and .

9. The cutting tool wear monitoring method according to claim 8, characterized in that: Recurrent encoder-decoder LSTMEnDe input features and Features and , output the encoded time series features .

10. The cutting tool wear monitoring method according to claim 9, characterized in that: The time fusion decoder TFD is used to decode the encoded time series features. Combination Perform data enhancement; use the time-domain self-attention multi-head mechanism to calculate attention and weights, and then perform feature calculation and fusion according to the weights; perform gated calculation, skip-layer connection and normalization on the fused features, and perform element-level feedforward calculation through the gated residual network GRN, and finally output the tool wear monitoring value.

Images