A Milling Tool Wear State Recognition Method, System, Medium and Device

Abstract

The present invention belongs to the field of intelligent monitoring of the milling process, and provides a method, a system, a medium and a device for identifying the wear state of a milling tool. Among them, the method for identifying the wear state of a milling tool includes obtaining any one of the cutting force signal and the acceleration signal during the milling process; using a tool wear state monitoring model to process the obtained signal to obtain the wear state of the milling tool; the process of using the tool wear state monitoring model to process the obtained signal is as follows: performing preliminary feature extraction and dimensionality reduction on the obtained signal to obtain a first local feature set; respectively performing feature extraction with three different depths on the first local feature set to correspondingly obtain a second local feature set, a third local feature set and a fourth local feature set; adaptively fusing the second local feature set, the third local feature set and the fourth local feature set to obtain a multi-scale feature set; determining the corresponding wear state of the milling tool based on the corresponding association relationship between the multi-scale feature set and the wear state of the milling tool.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent monitoring of the milling process, and particularly relates to a method, a system, a medium and a device for identifying the wear state of a milling tool. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of cutting machining, tool wear directly affects the machining accuracy and surface integrity of workpieces, and even leads to the scrapping of workpieces. Among them, the intelligent monitoring of the cutting process can ensure the safety and machining quality of automated machining equipment and is a key technology for realizing intelligent manufacturing. Traditional tool change strategies often rely on workers' experience or off-line measurement, but it is difficult to avoid the problem of reduced machining accuracy of workpieces caused by tool failure and waste caused by premature tool change.

[0004] In the prior art for realizing the intelligent monitoring of tool state during the machining process, considering the production cost and the complexity of the actual machining environment, the monitoring methods based on multi-sensor or single-sensor fusion have certain limitations: on the one hand, the information of multi-sensor signals is redundant, and the fusion of multi-sensor signals has limited improvement on the recognition accuracy of the model. More importantly, the increase in the number of sensors will cause the production cost to increase exponentially; on the other hand, the single-sensor signal has poor anti-interference ability and little effective information obtained; in addition, the installation of sensors is often limited by the workpiece size, cutting fluid, machine tool working range, etc. The replacement of sensor types in the actual machining environment is inevitable, but due to the large differences in the amplitude, statistical characteristics and change trends of different sensing signals, it is difficult to realize the generalization of single-channel sensing signals in tool wear state monitoring. These factors limit the popularization of the monitoring model in the actual machining environment.

[0005] In the existing system, due to the certain limitations in data transfer at different stages, the transfer and interaction of data between different software are difficult, and additional data conversion and interface development are required, resulting in a low degree of integration of the existing tool state monitoring system. It is difficult to directly map the cutting signal to the tool state and achieve cross-domain display. In addition, the existing monitoring systems often rely on manual experience and have cumbersome operation steps, which is not conducive to the construction of an unattended intelligent workshop system. Tool change often occurs before severe wear. Therefore, it can be considered that the recognition accuracy in the severe wear stage determines the quality of the monitoring model. However, the current method for identifying the tool wear state has poor recognition accuracy in the severe wear stage and cannot accurately guide tool change. These problems have become the key bottlenecks restricting the application and popularization of the tool state monitoring system in the milling process.

[0006] In summary, the existing tool wear state monitoring technology has problems such as poor generalization ability of single-channel sensing signals in tool wear state monitoring, low integration degree of the monitoring system, and difficulty in accurately guiding tool change. Summary of the Invention

[0007] To solve at least one of the technical problems existing in the above background technology, the present invention provides a milling tool wear state recognition method, system, medium and device, which has the advantages of sensing generalization ability and is applicable to tool wear state monitoring of single-channel and non-specific sensing signals.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] The first aspect of the present invention provides a milling tool wear state recognition method.

[0010] A milling tool wear state recognition method includes:

[0011] Obtaining any one of the cutting force signal and the acceleration signal during the milling process;

[0012] Processing the obtained signal by using a tool wear state monitoring model to obtain the milling tool wear state;

[0013] Among them, the process of processing the obtained signal by using the tool wear state monitoring model is:

[0014] Performing preliminary feature extraction and dimensionality reduction on the obtained signal to obtain a first local feature set;

[0015] Performing feature extraction on the first local feature set at three different depths respectively to obtain a second local feature set, a third local feature set and a fourth local feature set;

[0016] Adaptive fusion of the second local feature set, the third local feature set and the fourth local feature set to obtain a multi-scale feature set;

[0017] Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, the corresponding milling tool wear state is determined.

[0018] As an implementation manner of the first aspect of the present invention, during the training process of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal during the milling process is used as a sample set for training.

[0019] As an implementation manner of the first aspect of the present invention, the milling tool wear state includes an initial running-in state, a normal wear state and a severe wear state.

[0020] As an implementation of the first aspect of the present invention, the tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer.

[0021] As an implementation of the first aspect of the present invention, the parallel residual network layer is composed of three parallel branches, each branch stacks a number of residual blocks, and a batch normalization layer is arranged behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks of the same branch have the same size, and the convolutional kernels of the three branches have different sizes.

[0022] As an implementation of the first aspect of the present invention, after obtaining any one of the cutting force signal and the acceleration signal during the milling process, it further includes:

[0023] Data preprocessing operation; the data preprocessing operation includes: removing in-cut and out-cut data, data downsampling, and data selection.

[0024] The second aspect of the present invention provides a milling tool wear state recognition system.

[0025] A milling tool wear state recognition system provided by the present invention includes:

[0026] A signal acquisition module, which is used to acquire any one of the cutting force signal and the acceleration signal during the milling process;

[0027] A state recognition module, which uses the tool wear state monitoring model to process the above-acquired signal to obtain the milling tool wear state;

[0028] Among them, the process of using the tool wear state monitoring model to process the above-acquired signal is:

[0029] Perform preliminary feature extraction and dimensionality reduction on the above-acquired signal to obtain a first local feature set;

[0030] Perform feature extraction on the first local feature set at three different depths respectively, and correspondingly obtain a second local feature set, a third local feature set, and a fourth local feature set;

[0031] Adaptively fuse the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set;

[0032] Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state.

[0033] As an implementation of the second aspect of the present invention, during the training process of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal during the milling process is used as a sample set for training.

[0034] As an implementation manner of the second aspect of the present invention, the milling tool wear state includes an initial running-in state, a normal wear state, and a severe wear state.

[0035] As an implementation manner of the second aspect of the present invention, the tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer; the parallel residual network layer is composed of three parallel branches, each branch stacks a number of residual blocks, and a batch normalization layer is arranged behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks of the same branch have the same size, and the convolutional kernels of the three branches have different sizes.

[0036] As an implementation manner of the second aspect of the present invention, after obtaining any one of the cutting force signal and the acceleration signal during the milling process, it further includes:

[0037] Data preprocessing operations; the data preprocessing operations include: removal of in-cut and out-cut data, data downsampling, and data selection.

[0038] The present invention provides another milling tool wear state recognition system, which includes:

[0039] A signal real-time acquisition and transmission module, a state recognition module, and a cloud visualization module;

[0040] The signal real-time acquisition and transmission module is used to acquire any one of the cutting force signal and the acceleration signal during the milling process and transmit it to the state recognition module;

[0041] The state recognition module is used to process the above-obtained signal by using the tool wear state monitoring model to obtain the milling tool wear state and transmit it to the cloud visualization module;

[0042] Among them, the process of processing the above-obtained signal by using the tool wear state monitoring model is:

[0043] Perform preliminary feature extraction and dimensionality reduction on the above-obtained signal to obtain a first local feature set;

[0044] Perform feature extraction on the first local feature set at three different depths respectively to obtain a second local feature set, a third local feature set, and a fourth local feature set;

[0045] Adaptive fusion of the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set;

[0046] Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state;

[0047] The cloud visualization module is used to display the wear state of the milling tool in real time.

[0048] The third aspect of the present invention provides a computer-readable storage medium.

[0049] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the milling tool wear state recognition method as described above are implemented.

[0050] The fourth aspect of the present invention provides an electronic device.

[0051] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in the milling tool wear state recognition method as described above are implemented.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] (1) The present invention utilizes the characteristic that the cutting force signal and the acceleration signal have similar change trend characteristics. For the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal in the milling process is used as a sample set for training, and any one of the cutting force signal and the acceleration signal input during the milling process can be processed to identify the wear state of the milling tool, making the generalization ability of the tool wear state monitoring model of the present invention stronger, enabling the tool wear state monitoring method of the present invention to efficiently guide tool change, being applicable to single-channel and non-specific sensing signals, solving the problem of non-specific sensing signal input during the monitoring process, and eliminating the need to consider prior knowledge in the multi-channel signal fusion process.

[0054] (2) The present invention also provides a milling tool wear state recognition system. This system integrates a signal real-time acquisition and transmission module, a state recognition module, and a cloud visualization module, enabling the cutting signal to be quickly transmitted among the modules without the need to additionally configure a data conversion interface and a communication interface, providing a guarantee for the development of an integrated tool wear state monitoring system.

[0055] The advantages of the additional aspects of the present invention will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0057] Figure 1 It is a flowchart of the tool wear state monitoring method in the embodiment of the present invention;

[0058] Figure 2 This is the framework diagram of the intelligent monitoring system for the milling tool state in the embodiment of the present invention;

[0059] Figure 3 This is the structural schematic diagram of the residual block in the embodiment of the present invention;

[0060] Figure 4 This is the schematic diagram of the change point detection algorithm in the embodiment of the present invention;

[0061] Figure 5 This is the schematic diagram of the sliding average downsampling in the embodiment of the present invention;

[0062] Figure 6 This is the structural schematic diagram of the improved parallel residual network in the embodiment of the present invention;

[0063] Figure 7(a) is the confusion matrix of the acceleration signal recognition result in the embodiment of the present invention;

[0064] Figure 7(b) is the confusion matrix of the cutting force signal recognition result in the embodiment of the present invention. Detailed implementation manners

[0065] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0066] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0067] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0068] The existing tool wear state monitoring system has a low degree of integration, considering three aspects: First, data transmission is restricted. In the existing system, data transmission at different stages has certain limitations. For example, signal acquisition usually uses LabVIEW for real-time data acquisition, while signal processing may require other software such as MATLAB for further processing. This leads to difficulties in data transmission and interaction between different software, and additional data conversion and interface development are required. In addition, software compatibility issues: Compatibility between different software may also become an obstacle to integration. For example, LabVIEW and MATLAB may use different data formats or data interfaces, which requires additional development work to achieve smooth data transmission and sharing. Such compatibility issues result in a low degree of integration and increase the development and maintenance costs of the system. Finally, the need for cloud visualization. In modern monitoring systems, cloud visualization has become increasingly common. However, implementing cloud visualization usually requires additional cloud servers and data transmission costs.

[0069] Example 1

[0070] This embodiment provides a method for identifying the wear state of a milling tool, which specifically includes the following steps:

[0071] Step 1: Obtain any one of the cutting force signal and the acceleration signal during the milling process.

[0072] In some specific implementation processes, after obtaining any one of the cutting force signal and the acceleration signal during the milling process, it further includes:

[0073] Data preprocessing operations; the data preprocessing operations include: removing in-cut and out-cut data, data downsampling, and data selection.

[0074] Specifically, unstable signals during in-cut and out-cut are removed by a window-based change point detection method. The implementation of the window-based change point detection function is as follows:

[0075] It includes calculating the difference between two adjacent windows sliding along the signal y. For a given cost function C(·), the difference between two sub-signals is:

[0076] d(y a.t ,y t.b )=c(y a.b )-c(y a.t )-c(y t.b ) (1)

[0077] In the formula, 1≤a<t<b≤T, t represents the moment when the signal changes, a and b represent the starting and ending moments of the sampled sub-signal, and T represents the complete sampling time.

[0078] When two windows cover different time periods and the numerical values in the two windows differ by more than 10 times, resulting in a peak. Then, calculate the complete difference curve to obtain the index t of the change point.

[0079] In the specific implementation process, intercept the processed data and select the stable cutting data for a certain time period. Among them, in milling machining, stable cutting data refers to the data where the cutting parameters (such as cutting speed, feed rate, cutting depth, etc.) change while remaining in a relatively stable state during the milling of a workpiece. During the stable cutting process, the changes in cutting parameters often show a certain period, mainly to distinguish from the machine idle data here. It can be said that the data other than the machine idle data and the unstable data during the cutting-in and cutting-out is the stable cutting data.

[0080] It can be understood here that in other embodiments, those skilled in the art can also use other existing methods to remove the cutting-in and cutting-out data, which will not be elaborated here.

[0081] In the operations of data downsampling and data selection, for example, select a sliding average window with a length of 5 to downsample the stable cutting data, while reducing the influence of noise. Obtain the preprocessed data sample as the input of the state monitoring model.

[0082] It should be noted here that the length of the sliding average window can be specifically set by those skilled in the art according to the actual situation, which will not be elaborated here.

[0083] Step 2: Use the tool wear state monitoring model to process the above-obtained signal to obtain the milling tool wear state.

[0084] Among them, the process of using the tool wear state monitoring model to process the above-obtained signal is as follows:

[0085] Step 2.1: Perform preliminary feature extraction and dimensionality reduction on the above-obtained signal to obtain the first local feature set;

[0086] Step 2.2: Perform feature extraction on the first local feature set at three different depths, respectively obtaining the second local feature set, the third local feature set, and the fourth local feature set;

[0087] Step 2.3: Adaptively fuse the second local feature set, the third local feature set, and the fourth local feature set to obtain the multi-scale feature set;

[0088] Step 2.4: Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state.

[0089] In the specific implementation process, during the training of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal in the milling process is used as a sample set for training. The tool wear state monitoring model of this embodiment has the ability of perceptual generalization.

[0090] The tool wear state monitoring model is trained by any one of the cutting force signal and the acceleration signal. After training is completed, there is no need to use the other signal for training, and the other signal can be directly used as the input of the model to identify the tool wear state.

[0091] During the milling process, tool wear will affect the changes in the acceleration signal and the cutting force signal. Generally speaking, as the tool wear deepens, both the cutting force signal and the acceleration signal will show corresponding changes.

[0092] Cutting force signal: In an ideal state, during the tool wear process, the cutting force will gradually increase. Because when the tool starts to wear, its cutting efficiency will decrease, and more force is required to complete the same cutting work. This change will be reflected in the amplitude of the cutting force signal, causing the amplitude to gradually increase.

[0093] Acceleration signal: The acceleration signal is usually used to monitor the vibration of the tool. When the tool wears, its cutting performance will decline, which may lead to greater vibration. This vibration will be transmitted through the acceleration signal, causing the amplitude of the acceleration signal to increase.

[0094] From the above analysis, it can be seen that there is a close relationship between the cutting force signal and the acceleration signal, and they both reflect the tool wear state. When the tool starts to wear, its cutting efficiency will decrease, so more cutting force is required to complete the same work, causing the amplitude of the cutting force signal to gradually increase. At the same time, tool wear will also cause greater vibration, which will be transmitted to the acceleration signal, increasing its amplitude. As the tool wear deepens, the amplitudes of both signals will gradually increase, showing a similar trend of change.

[0095] The proposed tool wear monitoring model exactly utilizes this correlation. The model is designed as an end-to-end system that does not need to know the specific type of the input signal, but only needs to focus on whether the signal formats are consistent. During the training process, the model learns the mapping relationship between the characteristics of the acceleration signal and tool wear. Therefore, when in use, even if the input is the cutting force signal, due to the similar characteristics presented by these two signals in representing tool wear, the model can still accurately identify the tool wear state. This design effectively utilizes the generalization ability of the deep learning model, enabling it to handle different types of input signals.

[0096] In this embodiment, the wear states of the milling cutter include the initial running-in state, the normal wear state, and the severe wear state. The above-mentioned tool wear states are divided according to the tool wear change law: State 1 - the initial running-in stage, which appears with a rapid wear rate; State 2 - the normal wear stage, which has a uniform wear rate and a steady-state wear zone; State 3 - the severe wear stage, which again appears with a rapid wear rate.

[0097] It should be noted here that in other embodiments, those skilled in the art can specifically divide the categories of the wear states of the milling cutter according to the actual situation.

[0098] In this embodiment, the tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer.

[0099] Among them, the parallel residual network layer is composed of three parallel branches, and each branch stacks a number of residual blocks. Figure 3 The structure of the residual block is shown. The residual block adopts a skip connection, and the target mapping can be obtained through formula 2. The reason for adopting the skip connection: as the network deepens, it is difficult to achieve the identity transformation of features, and the model convergence speed and recognition accuracy will decline. By weakening the strong connection between each layer, the problems of gradient disappearance and explosion are avoided. It has been confirmed that the skip connection can improve the model training speed. In addition, the skip connection can make up for the information loss caused by downsampling and strengthen the information dependence relationship between layers.

[0100] The residual network indirectly obtains the target mapping by learning the residual, which is easier to implement compared with directly obtaining the target mapping. A batch normalization layer is set behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks in the same branch have the same size, and the convolutional kernels of the three branches have different sizes.

[0101] H(x) = F(x) + x (2)

[0102] In the formula, H(x) is the target mapping, F(x) is the residual function, and x is the input of the considered layer.

[0103] For example: Use a 7×7 wide convolutional kernel and a 2×2 max-pooling layer to perform preliminary feature extraction and dimensionality reduction on the preprocessed signal to obtain the first local feature set f1.

[0104] Take f1 as the input of the parallel residual network. In the parallel structure, three branches are designed, and three residual blocks are stacked in each branch. The number of channels of the data in the residual blocks is set to 64, 128, and 256 respectively. Different-sized convolutional kernels are used in each branch, which are 3×3, 5×5, and 7×7 respectively. The convolutional kernels use dilated convolution with dilation rates set to 1, 2, and 5 respectively. A batch normalization layer (BN) is set after the convolutional layer in each residual block to improve the generalization ability of the network. Thus, a local feature set containing three branches is obtained, and adaptive pooling operation is used for dimensionality reduction to obtain the second local feature set f2, the third local feature set f3, and the fourth local feature set f4.

[0105] All convolutional operations are one-dimensional convolutions, and the convolution operation is as follows:

[0106]

[0107] In the formula, is the feature output by the j-th convolutional layer, is the input of the i-th layer, f(·) is the activation function, and are the weights and biases of the convolutional layer respectively.

[0108] The pooling operation is as follows:

[0109]

[0110] In the formula, is the feature output by the j-th pooling layer, is the input of the i-th layer, and pool(·) is the pooling operation.

[0111] Adaptive fusion is performed on the f2, f3, and f4 local feature sets to obtain the fused multi-scale feature set f5. Deep, multi-scale, and large-perception features ensure that the model has the ability of perceptual generalization. The fused feature set is used as the input of the fully connected layer, and the tool wear state is output through the Softmax function. At the same time, the error is backpropagated, and thus the trained tool wear state monitoring model is obtained.

[0112] The expression of the cross-entropy loss function of the Softmax classifier is as follows:

[0113]

[0114] In the formula, p(x) is the expected output and q(x) is the actual output.

[0115] In the process of training a tool wear state monitoring model, a model trained with acceleration signals as input is, during application, verified using another set of acceleration signals and cutting force signals as input respectively. The final result is the confusion matrix of the recognition results of the two signals. Based on the confusion matrix of the recognition results, the recall rates and precision rates of the three categories are calculated respectively. Then, different weights are assigned to each category, weakening the weight of the initial wear stage and strengthening the weight of the severe wear stage. The weighted average value of the three stages is used as the final recognition recall rate and precision rate. The final evaluation index is obtained by comprehensively considering the calculation results of the recall rate and precision rate. Here, the subscripts I, N, and S represent the initial wear stage, normal wear stage, and severe wear stage respectively.

[0116] Taking the calculation of the precision rate and recall rate in the initial wear stage as an example, the recall rate can be expressed as:

[0117]

[0118] The precision rate can be expressed as:

[0119]

[0120] In the formula, TP I represents the number of samples that are recognized as positive samples from positive samples, FN I represents the number of samples that are recognized as samples of other categories from positive samples, and FP I represents the number of samples that are recognized as positive samples from negative samples.

[0121] Different weights w are assigned to the recall rates and precision rates of different categories to obtain the weighted precision rate PWA and recall rate RWA. The weighted recall rate is expressed as:

[0122] RWA = w1R I + w2R N + w3R S (8)

[0123] The weighted precision rate is expressed as:

[0124] PWA = w1P I + w2P N + w3P S (9)

[0125] In the formula, w is the weight factor, and w1, w2, and w3 are set to 0.2, 0.3, and 0.5 respectively.

[0126] Among them, the distribution principle of the weight factor is to consider the importance of each category in the problem. Like most machining scenarios, tool change often occurs before severe wear. Therefore, it can be considered that the accuracy of identifying the severe wear stage determines the quality of the monitoring model. Accurately identifying the severe wear stage is crucial for problem-solving. Therefore, a higher weight is assigned to it to ensure that the model pays more attention to the prediction results of this category. On the contrary, tool change rarely occurs in the initial wear stage of the tool, and the accuracy of identifying this stage has little impact on the selection of the final tool change timing. Therefore, a lower weight is assigned.

[0127] The comprehensive evaluation method of the model performance proposed in this embodiment takes into account the tool change timing and wear data distribution in the actual machining environment. The proposed method calculates the recall rate and precision rate of the recognition results, avoiding the defect of only using the accuracy rate to evaluate unbalanced data. In addition, different weights are assigned to each category, weakening the weight of the initial wear stage and strengthening the weight of the severe wear stage, and introducing a fusion factor to comprehensively consider the recall rate and precision rate, ensuring the recognition performance of the severe wear stage and effectively guiding the selection of tool change timing.

[0128] To verify the feasibility of the milling tool wear state monitoring method of this embodiment, a milling experiment on Ti-6Al-4V thin-walled parts was carried out through a DMU70V five-axis CNC machining center. The size of the workpiece is 100*150*5mm, the tool is a two-flute end mill with a diameter of 14mm, the spindle speed is 8000r / min, the feed rate is 1280mm / min, the radial depth of cut is 0.2mm, the axial depth of cut is 4mm, and the feed per tooth is 0.08mm / r.

[0129] During the cutting process, a three-axis acceleration sensor is installed on the back of the workpiece. A Kistler wireless rotary cutting dynamometer (RCD) and NIPXIE-4464 are used to collect cutting force and acceleration signals. The sampling frequency is set to 5kHz. The flank wear amount at 1 / 2 axial depth of each tooth of the tool is measured offline by a portable microscope. To reduce downtime, the tool wear is measured every ten cuts. A total of 10 wear values are obtained for each tool, and the remaining wear values are supplemented by spline interpolation to obtain complete wear labels.

[0130] Three tools of the same specification are selected to carry out the milling experiment. Each tool is cut 100 times, and the length of each pass is equal to the width of the workpiece. The three tools are cut 300 times in total, and the total cutting length is 4.5 meters. Three datasets including cutting signals and wear amounts are generated, denoted as T1, T2, and T3. When the flank wear of the milling cutter reaches the international failure standard, that is, VB≥300μm or VB max ≥500μm, stop further cutting.

[0131] Train a monitoring model using the acceleration signals and data labels in the x - direction from the two datasets T1 and T2. Use the T3 dataset to verify the feasibility of the milling tool wear state monitoring method and system based on the improved parallel residual network in the present invention.

[0132] First, obtain the acceleration in the x - direction and the cutting force signal during each cutting; pre - process the two collected cutting signals respectively, set the change - point detection interval, as Figure 4 shown, and successively complete the invalid data processing of the signal and the interception of the stable signal segment. And perform down - sampling through a sliding average window, as Figure 5 shown. Based on this, obtain the one - dimensional acceleration and cutting force signals in the stable cutting state as the input matrix of the monitoring model.

[0133] In this embodiment, use the T1 and T2 datasets as training data to establish and train the tool wear state monitoring model. Figure 6 shows the structure of the model. The initial learning rate, batch size, and number of epochs are set to 0.15, 64, and 200 respectively. The learning rate decay method is step - decay, and the decay rate is set to 0.1. It should be noted that the model structures for the two signal inputs are the same, avoiding the problem of repeated model training caused by sensor replacement in the actual machining environment. Obtain the confusion matrix of the output results of the T3 dataset, as shown in Figures 7(a) and 7(b).

[0134] Then, based on the recall rates and precision rates of the three categories, assign different weights to the recall rates and precision rates of different categories, weaken the weight in the initial wear stage, and strengthen the weight in the severe wear stage. At the same time, introduce a fusion factor to obtain an evaluation index CWA that comprehensively considers the recall rate and precision rate. After calculation, for the T3 dataset, for the recognition of acceleration and cutting force, CWA is 94.3% and 95.7% respectively. While ensuring high recognition accuracy, the precision difference between the two signals is 1.4%, verifying the model's perception generalization ability. This evaluation method focuses on the recognition in the severe wear stage and can accurately select the tool - changing time.

[0135] Embodiment 2

[0136] This embodiment provides a milling tool wear state recognition system, which includes:

[0137] A signal acquisition module, which is used to acquire either the cutting force signal or the acceleration signal during the milling process;

[0138] A state recognition module, which uses the tool wear state monitoring model to process the above - acquired signal to obtain the milling tool wear state;

[0139] Among them, the process of using the tool wear state monitoring model to process the above - acquired signal is:

[0140] Perform preliminary feature extraction and dimensionality reduction on the obtained signals above to obtain a first local feature set;

[0141] Perform feature extraction on the first local feature set at three different depths respectively, and correspondingly obtain a second local feature set, a third local feature set, and a fourth local feature set;

[0142] Adaptively fuse the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set;

[0143] Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state.

[0144] Among them, during the training process of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal in the milling process is used as a sample set for training.

[0145] Specifically, the milling tool wear state includes an initial running-in state, a normal wear state, and a severe wear state.

[0146] Among them, the tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer; the parallel residual network layer is composed of three parallel branches, each branch stacks a number of residual blocks, and a batch normalization layer is set behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks in the same branch have the same size, and the convolutional kernels of the three branches are of different sizes.

[0147] After obtaining any one of the cutting force signal and the acceleration signal in the milling process, it further includes:

[0148] Data preprocessing operations; the data preprocessing operations include: removal of cut-in and cut-out data, data downsampling, and data selection.

[0149] It should be noted here that each module in this embodiment corresponds one by one to each step in Embodiment 1, and its specific implementation process is the same, so it will not be repeated here.

[0150] Embodiment 3

[0151] As Figure 1 and Figure 2 shown, this embodiment provides a milling tool wear state recognition system, which includes:

[0152] A signal real-time acquisition and transmission module, a state recognition module, and a cloud visualization module;

[0153] The signal real-time acquisition and transmission module is used to acquire any one of the cutting force signal and the acceleration signal during the milling process and transmit it to the state recognition module;

[0154] The state recognition module is used to process the acquired signal by using the tool wear state monitoring model, obtain the milling tool wear state and transmit it to the cloud visualization module;

[0155] Among them, the process of processing the acquired signal by using the tool wear state monitoring model is as follows:

[0156] Perform preliminary feature extraction and dimensionality reduction on the acquired signal to obtain the first local feature set;

[0157] Perform feature extraction on the first local feature set at three different depths respectively, and correspondingly obtain the second local feature set, the third local feature set and the fourth local feature set;

[0158] Adaptively fuse the second local feature set, the third local feature set and the fourth local feature set to obtain a multi-scale feature set;

[0159] Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state;

[0160] The cloud visualization module is used to display the milling tool wear state in real time.

[0161] According to Figure 1 and Figure 2 , the hardware devices include vibration sensors, microphones, and NI acquisition boxes. An intelligent milling processing platform is built by using the hardware devices and a CNC milling machine.

[0162] In the signal real-time acquisition and transmission module, the real-time acquisition of the cutting signal is realized by using Labview software, and the cutting signal is transmitted to the MySQL cloud database through the ODBC data source to obtain the original data set.

[0163] The milling tool wear state recognition system provided in this embodiment further includes:

[0164] A signal preprocessing module.

[0165] In the data preprocessing module, the original data in the database is retrieved through the ODBC data source, and the change point detection interval is set. The invalid data during cutting in and cutting out is removed by using the mutation point detection algorithm. The implementation of the change point detection function based on the window is as follows:

[0166] It includes calculating the difference between two adjacent windows sliding along the signal y. For a given cost function C(·), the difference between two sub-signals is:

[0167] d(ya.t , y t.b ) = c(y a.b ) - c(y a.t ) - c(y t.b ) (1)

[0168] Where 1 ≤ a < t < b ≤ T, t represents the moment when the signal changes, a and b represent the start and end moments of the sampled sub-signal, and T represents the complete sampling time.

[0169] When the two windows cover different time periods, the difference reaches a relatively large value, resulting in a peak. Then calculate the complete difference curve to obtain the index t of the change point.

[0170] Then intercept the data after removing the cut-in and cut-out signals, select the signal of a certain time period, thereby completing the preprocessing of the original signal and obtaining the input matrix of the state monitoring model.

[0171] Based on the processed historical data, establish a tool wear state monitoring model based on an improved parallel residual network.

[0172] The preprocessed cutting signal X = (x1, x2, x3,..., x n ) is input into the monitoring model, and the tool wear state Y = (y1, y2, y3,..., y n ) is output. If y n is in the severe wear stage, the system gives an early warning and replaces the milling tool.

[0173] Where x n represents the nth signal sample input into the model, and y n represents the nth tool wear state recognition result.

[0174] In the cloud visualization module, the original signal and the output tool state are transmitted to the cloud database through the ODBC data source to achieve cross-regional monitoring visualization. At the same time, this module also realizes the management of machine tool information, sensor information, tool information, and working conditions, providing a reference value for realizing intelligent monitoring of the CNC machining process in the industrial environment.

[0175] Example 4

[0176] This example provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the milling tool wear state recognition method described in Example 1 above.

[0177] Example 5

[0178] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in the milling tool wear state recognition method described in the above-mentioned Embodiment 1 are implemented.

[0179] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0180] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying the wear state of a milling tool, characterized in that, It has the ability of perceptual generalization and is applicable to the tool wear state monitoring of single-channel and non-specific perceptual signals, including: Obtaining any one of the cutting force signal and the acceleration signal during the milling process; Using the tool wear state monitoring model to process the signal obtained above to obtain the tool wear state of the milling tool; Among them, the process of using the tool wear state monitoring model to process the signal obtained above is: Performing preliminary feature extraction and dimensionality reduction on the signal obtained above to obtain a first local feature set; Performing feature extraction with three different depths on the first local feature set respectively, and correspondingly obtaining a second local feature set, a third local feature set, and a fourth local feature set; Adaptive fusion of the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set; Based on the corresponding association relationship between the multi-scale feature set and the tool wear state of the milling tool, determining the corresponding tool wear state of the milling tool; The tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer; The parallel residual network layer is composed of three parallel branches, each branch stacks a number of residual blocks, and a batch normalization layer is set behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks of the same branch have the same size, and the convolutional kernels of the three branches have different sizes; the residual blocks adopt skip connections, and the target mapping is obtained through the following formula: ； In the formula, H(x) is the target mapping, F(x) is the residual function, x is the input of the layer under consideration; The preprocessed signal is subjected to preliminary feature extraction and dimensionality reduction using a 7×7 wide convolution kernel and a 2×2 max pooling layer to obtain a first local feature set f 1; Take f 1 as the input of the parallel residual network; among them, three branches are designed in the parallel structure, and three residual blocks are stacked in each branch. The number of channels of the data in the residual blocks is set to 64, 128, and 256 respectively; different sizes of convolutional kernels are used in each branch, which are 3×3, 5×5, and 7×7 respectively; the convolutional kernels use dilated convolution, and the dilation rates are set to 1, 2, and 5 respectively; a batch normalization layer is set after the convolutional layer in each residual block to improve the generalization ability of the network; thus, a local feature set containing three branches is obtained, and adaptive pooling operation is used for dimensionality reduction to obtain the second local feature set f 2. The third local feature set f 3. The fourth local feature set f 4; Will f 2、 f 3、 f Adaptively fuse the 4 local feature sets to obtain the fused multi-scale feature set f 5; The fused feature set is used as the input of the fully connected layer, and the tool wear state is output through the Softmax function; The expression of the cross-entropy loss function of the Softmax classifier is as follows: ； In the formula, p(x) is the expected output, and q(x) is the actual output.

2. The milling cutter wear state recognition method according to claim 1, wherein During the training process of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal during the milling process is used as a sample set for training; Or The tool wear state of the milling tool includes the initial running-in state, the normal wear state, and the severe wear state.

3. The milling cutter wear state recognition method according to claim 1, characterized in that After obtaining any one of the cutting force signal and the acceleration signal during the milling process, it further includes: Data preprocessing operations; the data preprocessing operations include: removing the in-cut and out-cut data, data downsampling, and data selection.

4. A milling cutter wear state recognition system, which is used to implement the milling cutter wear state recognition method described in any one of claims 1-3, and is characterized in that, Including: A signal acquisition module, which is used to obtain any one of the cutting force signal and the acceleration signal during the milling process; A state recognition module, which uses the tool wear state monitoring model to process the signal obtained above to obtain the tool wear state of the milling tool; Among them, the process of using the tool wear state monitoring model to process the signal obtained above is: Performing preliminary feature extraction and dimensionality reduction on the signal obtained above to obtain a first local feature set; Performing feature extraction with three different depths on the first local feature set respectively, and correspondingly obtaining a second local feature set, a third local feature set, and a fourth local feature set; Adaptive fusion of the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set; Based on the corresponding association relationship between the multi-scale feature set and the tool wear state of the milling tool, determining the corresponding tool wear state of the milling tool.

5. The milling cutter wear state recognition system according to claim 4, wherein During the training process of the tool wear state monitoring model, any one of the cutting force signal and the acceleration signal during the milling process is used as a sample set for training; Or The tool wear state of the milling tool includes the initial running-in state, the normal wear state, and the severe wear state; Or The tool wear state monitoring model includes a preliminary feature extraction layer, a parallel residual network layer, an adaptive fusion layer, and a fully connected layer; the parallel residual network layer is composed of three parallel branches, each branch stacks a number of residual blocks, and a batch normalization layer is arranged behind the convolutional layer in each residual block; the convolutional kernels in the residual blocks of the same branch have the same size, and the convolutional kernels of the three branches have different sizes; or After obtaining any one of the cutting force signal and the acceleration signal in the milling process, it further includes: Data preprocessing operations; the data preprocessing operations include: removal of cut-in and cut-out data, data downsampling, and data selection.

6. A milling cutter wear state recognition system, which is used to implement the milling cutter wear state recognition method described in any one of claims 1-3, characterized in that It includes: A signal real-time acquisition and transmission module, a state recognition module, and a cloud visualization module; The signal real-time acquisition and transmission module is used to acquire any one of the cutting force signal and the acceleration signal in the milling process and transmit it to the state recognition module; The state recognition module is used to process the acquired signal by using the tool wear state monitoring model to obtain the milling tool wear state and transmit it to the cloud visualization module; Among them, the process of processing the above-acquired signal by using the tool wear state monitoring model is: Perform preliminary feature extraction and dimensionality reduction on the above-acquired signal to obtain a first local feature set; Perform feature extraction with three different depths on the first local feature set respectively to obtain a second local feature set, a third local feature set, and a fourth local feature set; Adaptively fuse the second local feature set, the third local feature set, and the fourth local feature set to obtain a multi-scale feature set; Based on the corresponding association relationship between the multi-scale feature set and the milling tool wear state, determine the corresponding milling tool wear state; The cloud visualization module is used to display the milling tool wear state in real time.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the milling tool wear state recognition method described in any one of claims 1-3.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the milling tool wear state recognition method described in any one of claims 1-3.

Images