A tool wear real-time monitoring and prediction method and system based on multi-sensor information fusion

Abstract

The present invention belongs to the technical field related to numerical control machining, and discloses a real-time monitoring and prediction method and system for tool wear based on multi-sensor information fusion. The method comprises: S1, collecting monitoring signals and tool wear images during a machining experiment; S2, segmenting the tool wear images using a wear region segmentation model and calculating the true wear value; S3, establishing a monitoring training sample set using the monitoring signals and true wear values ​​to train the wear monitoring model; establishing a prediction training sample set using a sequence of true wear values ​​to train the wear prediction model; S4, monitoring the tool wear value in real time using the real-time monitoring signals and the wear monitoring model; and online prediction of the tool wear value using the wear prediction model. The present invention can achieve real-time monitoring of the tool wear value based on the real-time monitoring signals and online wear value prediction, which can significantly improve production efficiency.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field related to numerical control machining, and more particularly to a tool wear real-time monitoring and prediction method and system based on multi-sensor information fusion. BACKGROUND

[0002] In mechanical machining, 70%-80% of part forming is completed by numerical control machine tools, and the tool, as the execution end of the numerical control machine tool, inevitably produces friction with the chip and the workpiece. Influenced by the cutting force and cutting heat, the tool will continuously wear and the wear state will gradually deteriorate over time and finally collapse, which seriously affects the machining efficiency, machining quality and machining cost. Previous studies have shown that 10%-40% of the total downtime of machine tools is caused by tool failure. In addition, data shows that about 75% of cutting failures are due to the use of severely worn cutting tools. Therefore, intelligent monitoring and prediction of tool wear are of great significance to establishing a reasonable tool changing strategy and ensuring the continuity and safety of the machining process.

[0003] Current tool wear monitoring and prediction methods are divided into direct and indirect methods. The direct method directly measures the geometric size of tool wear through a microscope, a CCD camera, etc., which is more intuitive and has higher precision. However, this method requires downtime monitoring, and it is difficult to monitor tool wear in real time due to the influence of the machining environment such as chips and cutting fluid. In contrast, the indirect method indirectly reflects the tool wear state by analyzing signals such as vibration, cutting force, acoustic emission, and current that are strongly related to tool wear. This indirect wear measurement method is more robust and can monitor in real time, so it is used more frequently. Due to the complexity and nonlinearity of tool wear in actual cutting processes, as well as the diversity of materials and processes, it is difficult to establish an accurate tool wear mechanism model. At present, the indirect method usually uses data-driven intelligent methods, i.e., through machine learning or even deep learning methods to establish the relationship between sensor signal features and tool state.

[0004] However, there are the following problems in existing research: In most studies, the label data of tool wear value is obtained by measurement under offline conditions when constructing the data set, which is time-consuming and laborious, cannot obtain a large number of samples, and hinders the research process. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a tool wear real-time monitoring and prediction method and system based on multi-sensor information fusion, which is used to solve the problem that the label data of tool wear value needs to be obtained by measurement under offline conditions when constructing the data set in the existing research on monitoring tool state by machine learning method, which is time-consuming and laborious, and has low efficiency.

[0006] To achieve the above object, according to one aspect of the present application, a tool wear real-time monitoring and prediction method based on multi-sensor information fusion is provided, comprising:

[0007] Offline stage:

[0008] S1, performing a machining experiment, collecting monitoring signals obtained by multiple sensors during the machining experiment and tool wear images obtained by a camera;

[0009] S2, performing wear area segmentation on the tool wear images by using a trained wear area segmentation model, and calculating a real wear value based on the segmented wear area;

[0010] S3, establishing a monitoring training sample set by using the monitoring signals and the corresponding real wear value, training a wear monitoring model by using the monitoring training sample set, and obtaining a trained wear monitoring model;

[0011] establishing a prediction training sample set by using a sequence of the real wear value, training a wear prediction model by using the prediction training sample set, and obtaining a trained wear prediction model;

[0012] Online stage:

[0013] S4, during the machining process, obtaining real-time monitoring signals by using multiple sensors; monitoring the wear value of the tool in real time by using the real-time monitoring signals and the trained wear monitoring model;

[0014] and predicting the wear value of the tool by using the trained wear prediction model.

[0015] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the present application, before the wear monitoring model is trained by using the monitoring training sample set in S3, it further comprises:

[0016] performing noise reduction processing and feature extraction processing on the monitoring signals to obtain preprocessed features; when the wear monitoring model is trained, the preprocessed features are used as the input of the wear monitoring model, and the output of the wear monitoring model is a wear value prediction value;

[0017] Among them, the noise reduction processing on the monitoring signals comprises:

[0018] at least one operation of average filtering, wavelet transform, inverse wavelet transform and no-load removal processing on the monitoring signals;

[0019] The feature extraction processing on the monitoring signals comprises:

[0020] The multi-domain features are extracted from the time domain, the frequency domain and the time-frequency domain of the monitoring signal after noise reduction processing; the strong correlation features of wear are screened from the multi-domain features by using the Pearson correlation coefficient method; and the high-dimensional features of the strong correlation features of wear are mapped to a low-dimensional space by using the KPCA method to obtain pre-processing features.

[0021] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, before the high-dimensional features of the strong correlation features of wear are mapped to a low-dimensional space by using the KPCA method in the feature extraction processing of the monitoring signal, the method further comprises the following steps:

[0022] The strong correlation features of wear are subjected to exponential filtering operation.

[0023] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the monitoring signal obtained by the plurality of sensors is subjected to noise reduction processing and feature extraction, and the pre-processing features are obtained, and the method further comprises the following steps:

[0024] The monitoring signals obtained by the plurality of sensors are subjected to noise reduction processing and multi-domain feature extraction respectively, then the multi-domain features corresponding to the monitoring signals obtained by the plurality of sensors are fused, and then the strong correlation features of wear are screened by using the Pearson correlation coefficient method based on the fused multi-domain features.

[0025] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the wear area segmentation model in S2 is a YOLOv8 segmentation model; the wear area segmentation model is trained by constructing a segmentation sample set through wear area labeling of the tool wear image obtained in S1.

[0026] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the real wear value is calculated based on the segmented wear area in S2, and specifically comprises the following steps:

[0027] The pixel coordinates of the vertices of the wear area are obtained to form a vertex point set;

[0028] The equations of the line segments between two vertices are obtained by traversing the vertex point set; and the original cutting edge line segment of the tool is determined by screening the slope of the line segment and the number of vertices passing through;

[0029] The perpendicular distance between the remaining points of the vertex point set and the original cutting edge line segment of the tool is calculated, and the maximum value of the perpendicular distance is selected as the pixel number of the maximum wear width;

[0030] The actual value of the maximum wear width is obtained as the real wear value according to the pixel number of the maximum wear width and the size of the tool wear image.

[0031] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the plurality of sensors in S1 include multiple of a vibration sensor, a cutting force sensor, and an acoustic emission sensor; the wear monitoring model in S3 is a KAN deep learning model; and the wear prediction model in S3 is a Transformer model.

[0032] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the wear value of the tool is predicted by using the trained wear prediction model, and the method specifically comprises the following steps.

[0033] When the wear prediction model is predicted for the first time, the input is a monitoring value sequence of the tool wear value, and the output is a predicted value sequence of the tool wear value;

[0034] In each subsequent prediction of the wear prediction model, the input is a combination of the monitoring value of the tool wear value and the predicted value output by the previous prediction, or is a predicted value sequence output by the previous prediction, and the output is a predicted value sequence of the tool wear value, forming a lead prediction mode;

[0035] The monitoring value of the tool wear value is an output value of the wear monitoring model.

[0036] According to the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application, the lead prediction mode further comprises:

[0037] A timely correction strategy: comparing the predicted value sequence of the tool wear value output by the wear prediction model with the corresponding monitoring value sequence of the tool wear value to obtain the error between the predicted value and the monitoring value;

[0038] If the error is less than a preset threshold, the online prediction is continued in the lead prediction mode; if the error is greater than or equal to the preset threshold, the input is the monitoring value sequence of the tool wear value when the next prediction of the wear prediction model is performed, so as to correct the model input; and then the online prediction is continued in the lead prediction mode.

[0039] According to another aspect of the application, a tool wear real-time monitoring and prediction system based on multi-sensor information fusion is provided, which comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the tool wear real-time monitoring and prediction method based on multi-sensor information fusion according to any one of the above aspects.

[0040] Overall, compared with the prior art, the tool wear real-time monitoring and prediction method and system based on multi-sensor information fusion provided by the application have the following advantages:

[0041] 1. The trained wear area segmentation model can realize online segmentation of tool wear area, and further online calculation of wear value, without offline test to obtain wear value, which can significantly improve the acquisition speed of monitoring model and prediction model training sample set, and simplify the operation; further training of the wear monitoring model can realize real-time monitoring of the wear value of the tool according to the real-time monitoring signal, and further based on the wear prediction model trained by the wear value, online wear value prediction can be realized, which can realize real-time prediction without interfering with production, and this prediction method can greatly improve production efficiency;

[0042] 2. The most representative and explanatory features are obtained through multi-domain feature extraction and KPCA dimension reduction, which not only improves the prediction accuracy, but also effectively removes redundant information;

[0043] 3. The tool wear segmentation model based on YOLOv8 can efficiently identify the tool wear area and accurately calculate the tool wear value with an error of less than 5μm;

[0044] 4. The prediction accuracy of the KAN tool wear real-time monitoring model on the three sets of experimental data of the public data reached 100%, 99.99% and 99.98%, respectively, and reached 99.95% on the experimental data. At the same time, it has very low error on both public and experimental data;

[0045] 5. In the traditional prediction mode, the prediction accuracy of the Transformer tool wear prediction model on the three sets of experimental data of the public data reached 99.98%, 99.94% and 99.94%, respectively, and reached 99.94% on the experimental data; At the same time, it has very low error on both public and experimental data;

[0046] 6. In the advanced prediction mode, the tool wear prediction model has generally poor prediction effect under different T1 and T2 parameters. However, by adopting the timely correction strategy, under different T1 and T2 parameters, it has at least 99.31%, 97.00% and 99.14% on the three sets of experimental data of the public data, respectively; It has at least 99.38% accuracy on the experimental data; At the same time, it has relatively low error on both public and experimental data. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is the overall structure diagram of the tool wear real-time monitoring and prediction method based on multi-sensor information fusion provided by the application;

[0048] Figure 2 is the structure diagram of the YOLOv8 tool wear area segmentation model;

[0049] Figure 3 is the change of the training index of the segmentation model;

[0050] Figure 4 is a tool wear segmentation model inference effect diagram, wherein (a) is when the tool is not worn; (b) is at the initial stage of wear; (c) is at the middle stage of wear; and (d) is at the late stage of wear;

[0051] Figure 5 is a process of calculating the quantitative tool wear value, wherein (a) identifies the original cutting edge; (b) obtains the quantitative tool wear value; and (c) converts the pixel value into the actual length;

[0052] Figure 6 is a public data spectrum diagram, wherein (a) is a C1 data set force signal spectrum; (b) is a C4 data set force signal spectrum; (c) is a C6 data set force signal spectrum; (d) is a C1 data set vibration signal spectrum; (e) is a C4 data set vibration signal spectrum; and (f) is a C6 data set vibration signal spectrum;

[0053] Figure 7 is a feature map before feature post-processing and before dimensionality reduction of the feature;

[0054] Figure 8 is a feature map after feature post-processing and before dimensionality reduction of the feature;

[0055] Figure 9 is a schematic diagram of a lead prediction mode. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0057] Please refer to Figure 1 The embodiment provides a tool wear real-time monitoring and prediction method based on multi-sensor information fusion, which comprises the following steps:

[0058] Offline stage:

[0059] S1, performing a machining experiment, collecting monitoring signals obtained by a plurality of sensors during the machining experiment and tool wear images obtained by a camera;

[0060] S2, segmenting a wear area of the tool wear image by using a trained wear area segmentation model, and calculating a real wear value based on the segmented wear area;

[0061] S3, a monitoring training sample set is established by using the monitoring signal and the corresponding real wear value, a wear monitoring model is trained by using the monitoring training sample set, and a trained wear monitoring model is obtained;

[0062] A prediction training sample set is established by using the sequence of real wear values, a wear prediction model is trained by using the prediction training sample set, and a trained wear prediction model is obtained;

[0063] In the online phase:

[0064] S4, in the machining process, real-time monitoring signals are obtained by using a plurality of sensors; the real-time monitoring signals and the trained wear monitoring model are used to monitor the wear value of the tool in real time;

[0065] And the trained wear prediction model is used to predict the wear value of the tool.

[0066] The embodiment further considers that the existing research is less developed while considering the tool wear monitoring and prediction. The combination of the two can combine short-term real-time monitoring and long-term planning and decision-making, significantly improve production efficiency, ensure product quality, and prolong tool life.

[0067] In addition, the existing tool wear prediction model belongs to an interpolation type prediction, and the model is based on limited historical wear data to make a prediction for a limited time step in the future. This prediction mode does not have the ability to perceive the wear trend of a super-long time step, and has certain limitations in actual application. In view of the deficiencies of the prior art, the main solution of the present application is:

[0068] An automatic data acquisition platform is established to automatically collect multi-sensor information and tool wear images; multi-domain feature information of multi-sensor data is extracted, and through fusion and post-processing, the most representative and explanatory monitoring model input features are obtained through effective dimension reduction methods; a machine vision model is constructed by using the tool wear image, the tool wear area is automatically identified, and the quantitative tool wear value is further calculated to provide label data for the data set; a tool wear monitoring model is constructed to realize real-time mapping of multi-domain feature data of the sensor to the tool wear value; a tool wear prediction model is constructed to realize the mapping relationship between the historical and future tool wear sequences; the traditional prediction mode is improved, and a timely correction strategy is combined to realize the prediction of a super-long time step while ensuring the prediction accuracy.

[0069] Specifically, an automated data acquisition platform is built to automatically acquire multi-sensor data and tool wear images to provide support for feature and label data. When acquiring tool wear images, a safe shooting point is set inside the machine tool, and two industrial cameras and lenses are installed below and to the side to acquire the bottom and side images of the tool. To facilitate the acquisition of images of each side edge of the tool, the M100CN instruction is developed to orient the machine tool spindle at any angle, where N is the angle of rotation with the encoder zero point of the machine tool as the reference point. In addition, the camera software provides a function to take pictures with shortcut keys, combined with the FOCAS communication protocol, to automatically acquire tool bottom and side edge wear images. When acquiring multi-sensor information, vibration sensors, force sensors, acoustic emission sensors, etc. are installed on the machine tool workbench, combined with data acquisition cards connected to the PC, and the Devsoft software provides a shortcut key to record data, which can also automatically acquire multi-sensor data.

[0070] After obtaining the tool wear images, the annotation software is used to label the wear area to generate a label file to construct a data set, and the data set is divided into a training set and a test set according to a certain proportion. A deep learning instance segmentation model is built, and the wear images and their label files are used as model inputs to iteratively train the model, allowing it to learn key features and determine the optimal parameters. Finally, the model has the ability to segment tool wear areas, and can intelligently segment tool wear areas at the pixel level. After training, the performance of the model is evaluated on the test set. In order to obtain quantitative tool wear values, the tool wear area is segmented based on the morphological calculation of the quantitative wear value. This value provides a label value for the subsequent tool wear real-time monitoring and prediction data set.

[0071] After obtaining multi-sensor information, it is first preprocessed by filtering, wavelet transform, and removing empty load. Second, multi-domain features are extracted from the time domain, frequency domain, and time-frequency domain, and the most representative features are obtained through feature post-processing, feature fusion dimension reduction, and other operations. The Pearson correlation coefficient method is used to screen features that are strongly correlated with wear. Finally, the KPCA method is used to map high-dimensional features to low-dimensional space, allowing the model to focus on truly valuable features. This feature provides feature values for the subsequent tool wear real-time monitoring and prediction data set.

[0072] A KAN (Kolmogorov-Arnold Networks) tool wear real-time monitoring model is established, and a neural network is used to establish a nonlinear relationship between multi-sensor multi-domain features and quantified tool wear values, and to complete real-time mapping of multi-sensor multi-domain features to tool wear values. First, the data set is divided into a training set and a test set to evaluate the performance of the model during the training process. Second, the input features are transmitted to the neurons of the hidden layer. Each hidden layer neuron calculates the univariate function output of each input variable and sums it up, and then performs a nonlinear transformation. Third, the outputs of all hidden layer neurons are weighted and summed to obtain the final output. This process realizes the nonlinear combination of high-dimensional input through univariate functions to approximate the output of the target function. Then, in the training stage, all parameters are randomly initialized, the difference between the model output and the true label is calculated to obtain the loss function. Finally, the backpropagation algorithm is used to update all parameters of the model to obtain the optimal performance monitoring model. The output sequence is used as the input of the Transformer prediction model to calculate the future tool wear sequence.

[0073] A Transformer tool wear prediction model is established, and a neural network is used to establish a nonlinear relationship between historical tool wear sequences and future tool wear sequences to complete the prediction of future tool wear values from historical tool wear values. First, the input feature data and label data are appropriately processed and formatted, and divided into a training set and a test set. Second, the input is reshaped into a three-dimensional format [batch size, sequence length, features]. This is because the Transformer model is designed based on sequence processing, which requires a clear time step and feature dimension of the sequence. Then, the model is constructed with input layer, linear embedding layer, multi-head attention layer, feedforward neural network layer, global average pooling layer, dropout layer and output layer, the model output is calculated and the loss is calculated with the label. Finally, the parameters of the model are updated iteratively by the backpropagation algorithm to obtain the optimal parameter prediction model.

[0074] In view of the lack of long-time step wear trend perception ability of the traditional interpolation type prediction mode, a lead prediction mode is proposed. In this mode, the model takes the previous time step prediction output as input each time, so it has unlimited time step prediction ability. At the same time, the accuracy decline caused by the accumulation of prediction errors is more and more serious, so the invention proposes a timely correction strategy. The timely correction strategy can compensate for the error accumulation problem in the lead prediction mode and correct the model input in time. The combination of the lead prediction mode and the timely correction strategy not only enables the model to have long-time step prediction ability, but also ensures the prediction accuracy of the model.

[0075] In another embodiment, a real-time monitoring and prediction system for tool wear based on multi-sensor information fusion is also provided. The system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it executes the real-time monitoring and prediction method for tool wear based on multi-sensor information fusion described in any embodiment.

[0076] In some specific embodiments, the present invention provides a method for real-time monitoring and advance prediction of tool wear based on multi-sensor information fusion, including a fully automated data acquisition platform, a tool wear segmentation model, a tool wear quantification method, multi-domain feature extraction of multi-sensor information, a real-time monitoring model for tool wear, a tool wear prediction model, an advance prediction mode for tool wear, and a timely correction strategy, such as Figure 1 shown.

[0077] Example 1

[0078] Build the fully automated data acquisition platform; set up safe photo-taking points inside the machine tool. Photo-taking points can be set up outside the machine tool processing area. Two industrial cameras are installed below and to the side of the photo-taking points to respectively capture images of the bottom and side wear of the tool;

[0079] To automatically capture images of each cutting edge, an M100CN command was developed to orient the machine tool spindle to any angle, where N is the angle of spindle rotation. Communication between the machine tool and the PC was established using the TCP / IP network protocol, and key information such as FANUC CNC system variables, status, and parameters was obtained based on the FOCAS communication protocol. Furthermore, the camera's accompanying software includes the ability to take and store images using shortcut keys, making it possible to automate image acquisition using C# scripts.

[0080] On the other hand, by installing vibration sensors and dynamometers on the machine tool workbench, combined with a data acquisition card and PC, multi-sensor information can be collected. The Devsoft software supporting the data acquisition card has the function of storing data using shortcut keys, which also facilitates the automatic acquisition of multi-sensor data using C# scripts.

[0081] The monitoring signal collected in S1 has a corresponding relationship with the tool wear image. At the end of each acquisition of the monitoring signal, the tool can be controlled to move to the image acquisition point, the bottom wear image of the tool is acquired by the industrial camera below, the side edge wear image of the tool is acquired by the industrial camera on the side, and the tool is controlled to rotate at the image acquisition point to acquire the wear image of different side edges of the tool; wherein the bottom wear image of the tool and the side edge wear image of the tool form the tool wear image. That is, the tool wear image corresponding to each monitoring signal can be acquired, and then the tool wear value corresponding to each monitoring signal can be acquired. For example, the monitoring signal can be collected during the machining of each tool, and then the tool wear image is collected when the tool machining is completed. Then, by marking the wear area of the tool wear image acquired in S1, the construction of the segmentation sample set is performed, and the wear area segmentation model is trained by using the segmentation sample set.

[0082] Overall, in the machine tool machining process, first, the multi-sensor data is automatically and real-time acquired by using C# script; second, the machine tool NC code subroutine is developed, so that the tool is automatically moved to the shooting point after a certain number of cutting and the machine tool is paused; then, the wear images of the bottom surface and side edges of the tool are automatically acquired by C# script in cooperation with the NC code; finally, the machining process is continued, and the full-automatic data acquisition is realized by means of FOCAS communication and C# script, and the specific steps are shown in Table 1.

[0083] Table 1 Tool wear online measurement method

[0084]

[0085]

[0086] Example 2

[0087] The tool wear area segmentation model is a YOLOv8 segmentation model; the tool wear segmentation model is constructed by using the YOLOv8 open source project; in order to identify the tool wear area, the tool wear area is segmented by machine vision; the wear area in the image is manually marked by using a marking software, and the obtained 2176 samples are sequentially divided into a training set, a validation set and a test set in a ratio of 1546:546:84; the training set is used to train the YOLOv8 instance segmentation model, and the validation set and the test set are used to evaluate the instance segmentation performance of the YOLOv8 model;

[0088] As shown in Figure 2The YOLOv8 model architecture shown performs instance segmentation on the wear area of the tool; the model mainly consists of basic modules such as Conv, C2f, SPPF, Upsample, Concat, etc.; the Conv module performs convolution operation on the feature map, which is used to extract the local features of the image; the C2f module realizes efficient calculation by splitting, processing and fusing features; the SPPF module realizes multi-scale extraction and fusion of features through multi-level spatial pyramid pooling; the Upsample module uses deconvolution to restore the resolution of the feature map, thereby improving the positioning accuracy of the target; the Concat module is used for feature fusion and skip connection; these basic modules constitute the YOLOv8 network architecture according to the connection mode on the right side of the figure, and different Segment layers are used to identify targets of three scales in the input image; the design of the anchor box with different sizes in the output layer enables the model to realize multi-scale feature extraction and fusion, ensuring good detection of multi-scale targets;

[0089] After preparing the tool wear image and the corresponding label file, the PC configures the YOLOv8 running environment and starts iterative training; after 200 rounds of iterative training, the index change of the YOLOv8 model is as shown in Figure 3 ; it can be seen that the loss gradually decreases to a small value during the entire training process; in addition, the precision and recall rate of BoundingBox and Mask both reach 100%; and the mAP50, mAP50:95 of BoundingBox and the mAP50, mAP50:95 of Mask are 0.995, 0.995, 0.995 and 0.913 respectively; obviously, the model has very superior instance segmentation performance;

[0090] As Figure 4 shown, the results of randomly selected pictures in each stage of the entire life cycle of the tool after instance segmentation by the model, it can be seen that the model can effectively identify the wear area in any stage of tool wear, providing a basis for subsequent tool wear measurement.

[0091] Embodiment 3

[0092] In S2, the real wear value is calculated and obtained based on the segmented wear area, specifically including:

[0093] Obtain the pixel coordinates of the vertices of the wear area to form a vertex point set;

[0094] Iterate through the vertex point set to obtain the equations of the line segments between two vertices; and determine the tool original cutting edge line segment by screening the slope of the line segment and the number of vertices passing through;

[0095] Calculate the perpendicular distance between the remaining points of the vertex point set and the tool original cutting edge line segment, and select the maximum value as the pixel number of the maximum wear width.

[0096] According to the number of pixels of the maximum wear width and the size of the tool wear image, the actual value of the maximum wear width is obtained as the real wear value.

[0097] That is, a tool wear quantification method is proposed, and the tool wear value is further calculated from the wear area obtained by segmentation; the international standardization organization stipulates that the wear standard of the carbide tool in rough machining is that, under the normal wear condition of the cutting tool, the average wear width VB average = 0.3 or the maximum wear width VB max = 0.5 is the wear standard; the maximum wear width VB max is used as the wear standard for subsequent research;

[0098] The wear area obtained by example segmentation is an irregular polygon, and the vertex coordinates of the area can be used to further calculate VB max ; first, the equations of the line segments formed by two points are obtained by traversing the vertices of the polygon; that is, for the irregular polygon area obtained after segmentation, the target set T = {t1, t2,..., t n}, wherein each target t i is a group of two-dimensional points on the polygon area, and a point P = (x, y) is a point on a two-dimensional plane, and a line segment represents a straight line segment connecting points p and q; the length d(p, q) of the line segment is obtained by using the Euclidean distance and calculating the slope;

[0099] On this basis, the line segments whose slope and the number of points passing through the vertices of the polygon do not meet the conditions are filtered out to determine the original cutting edge of the tool, such as (a) in Figure 5 ; all point pairs (p, q) in the target set are specifically traversed, the distance and the slope are calculated, and the line segments that do not meet the conditions are filtered out; if the number of points passing through the line segment is greater than or equal to the threshold value (the threshold value set by the present application is 20), the longest distance and the line segment are updated, and the longest line segment obtained after traversing all the vertices of the polygon is the original cutting edge of the tool when the tool is not worn;

[0100] The perpendicular distances of other vertices from the original cutting edge are calculated in turn, that is, the distance of a given point r = (x3, y3) from the line segment is calculated; the maximum distance at this time represents the pixel value number of the maximum wear width VB max , such as (b) in Figure 5 ; that is, after the boundary of the original cutting edge of the tool is obtained, the point on the wear area farthest from the boundary needs to be determined, and the perpendicular distance of the point from the original cutting edge is the maximum wear width VB max of the tool flank face;

[0101] The maximum wear width VB can be calculated by calculating the distance from the point to the line max The pixel value is further converted based on the actual length and width of the image to obtain VB max The camera software is equipped with basic functions such as measuring area and distance, which can measure the actual width and height of the wear image. Since the image resolution is fixed at 3840*2160, VB can be easily obtained after conversion. max The true value of Figure 5 (c) in the.

[0102] Example 4

[0103] The multi-domain feature extraction part of multi-sensor information includes two parts: preprocessing and feature extraction; that is, before using the monitoring training sample set to train the wear monitoring model in S3, it also includes: performing noise reduction processing and feature extraction processing on the monitoring signal to obtain preprocessing features; when training the wear monitoring model, the preprocessing features are used as the input of the wear monitoring model, and the output of the wear monitoring model is the wear value prediction value.

[0104] Correspondingly, in S4, the real-time monitoring of the tool wear value is performed using the real-time monitoring signal and the trained wear monitoring model, including: performing noise reduction processing and feature extraction processing on the real-time monitoring signal to obtain real-time preprocessing features, inputting the real-time preprocessing features into the trained wear monitoring model to obtain the real-time monitoring value of the tool wear value.

[0105] The multiple sensors in S1 include multiple vibration sensors, cutting force sensors, and acoustic emission sensors; the sensors can be installed on the machine tool workbench to collect processing monitoring signals during operation. During preprocessing, since the cutting force, vibration, and acoustic emission data collected by the sensors have a large number of data points and contain interference information from factors such as the environment and equipment vibration, this noise will affect the accurate prediction of the tool wear state. Therefore, it is necessary to perform noise reduction preprocessing on the raw data of multiple sensors; the noise reduction processing of the monitoring signals includes: performing at least one of average filtering, wavelet transform, inverse wavelet transform, and de-noising processing on the monitoring signals; for example:

[0106] First, a filter of size 7 is used to implement average filtering through a sliding window; as shown in formula (1), where S(n+k) represents the value of the original signal at position n+k, K(k) represents the filter, w represents the filter window size, and F(n) represents the filtered signal; it can be seen that the filtered signal is equal to the average value of the signal before filtering after the convolution filter in a window;

[0107] Secondly, the db4 wavelet is used to carry out wavelet transform, and the signal is decomposed into detail and approximate parts of different scales; then the wavelet coefficients after threshold processing are used to carry out inverse wavelet transform to obtain the reconstructed signal; as shown in formula (2), wherein DWT represents discrete wavelet transform, c j,k represents approximate part coefficients, d j,k represents detail part coefficients, φ j,k (t) represents a scale function, and ψ i,k (t) represents a wavelet function; it is worth noting that only the detail coefficients are processed when the threshold is processed, and then the approximate coefficients and the detail coefficients after threshold processing are used to reconstruct the signal;

[0108] Finally, since the sensor data collected when the tool is cut in and cut out (empty load) has no effect on tool wear and has no characterization ability, the first and last 5% of data in each tool path are removed.

[0109]

[0110] The feature extraction processing of the monitoring signal includes: extracting multi-domain features from the time domain, frequency domain and time-frequency domain of the monitoring signal after noise reduction processing; using the Pearson correlation coefficient method to screen out strong correlation features (correlation coefficient greater than 0.9) from the multi-domain features; using the KPCA method to map the high-dimensional features of the strong correlation features to a low-dimensional space to obtain preprocessed features. In the feature extraction processing of the monitoring signal, before using the KPCA method to map the high-dimensional features of the strong correlation features to a low-dimensional space, it further includes: performing an exponential filtering operation on the strong correlation features, and using an exponential filtering with a smoothing coefficient of 0.02 to effectively alleviate the frequent sudden changes of the strong correlation features.

[0111] The noise reduction processing and feature extraction processing of the monitoring signal to obtain preprocessed features further include: respectively performing noise reduction processing on the monitoring signals obtained by the plurality of sensors and extracting multi-domain features, then fusing the multi-domain features corresponding to the monitoring signals obtained by the plurality of sensors, and then using the Pearson correlation coefficient method to screen out strong correlation features based on the fused multi-domain features.

[0112] Specifically, the features are extracted as shown in Table 2; in the time domain, the mean, root mean square, standard deviation, signals of three directions of cutting force, signals of three directions of vibration and acoustic emission data are extracted and fused to form 21-dimensional features; in the frequency domain, the center frequency, frequency variance, mean frequency, signals of three directions of cutting force, signals of three directions of vibration and acoustic emission data are extracted and fused to form 21-dimensional features; taking the 160th pass data of the public data as an example, the three-direction force and vibration data spectrum obtained by Fourier transform are as shown in FIG. 8; Figure 6 It can be seen that the frequency of the signal is mainly concentrated in the frequency band of 1-10Khz, and the resolution of each sub-band can be determined to be not less than 518Hz through the spectrum; therefore, in the time-frequency domain, the frequency band of 1-12.5Khz is divided into 32 sub-bands, and the energy of the original signal is calculated in each sub-band, and 224-dimensional features are obtained;

[0113]

[0114] However, the high-dimensional features extracted manually have a lot of redundant information, which will significantly consume the model training time and resources; therefore, the Pearson correlation coefficient method shown in formula (5) is used to select 35-dimensional features with a correlation coefficient greater than 0.9 with the tool wear as strong correlation features from the 21+21+224-dimensional feature signals;

[0115]

[0116] By observing the figure of feature extraction and screening in Figure 7 It is found that the first 34 feature signals and the 35th wear signal have a positive correlation, but the feature signals frequently appear to drop after dimensionality reduction; there can be many reasons for this phenomenon; one is that the average filter window size is not reasonable, another possibility is that the signal itself contains some rapidly changing events or spikes, or the selected filter is not suitable for processing the sudden drop in the signal; no matter what the reason is, in the face of this situation, targeted feature post-processing is needed to avoid affecting the subsequent model prediction performance; the exponential filtering method with a decay factor of 0.02 is used in the present application to eliminate the frequent drop in the signal, as shown in (a) of FIG. 9; Figure 8

[0117] ​The post-processed features can be highly correlated with each other, meaning that they carry repetitive information; in order to reduce the computational overhead and extract more representative features, the kernel principal component analysis (KPCA) method is used to further map the 35-dimensional high-dimensional features to a low-dimensional space, so that the subsequent model can focus on the truly valuable features in the data, and finally obtain 6-dimensional features, as shown in (b) of FIG. 6; the specific implementation steps of the KPCA feature dimension reduction method are shown in Table 3; Figure 8

[0118] Finally, the samples are constructed in units of 3 time steps, and the feature and label data are constructed, and the training set and test set are divided in a ratio of 7:3 for subsequent monitoring and prediction models.

[0119] Table 2 Multi-domain feature extraction

[0120]

[0121]

[0122] Table 3 KPCA specific implementation steps

[0123]

[0124] Example 5

[0125] The tool wear monitoring model is a KAN deep learning model; the KAN tool wear real-time monitoring model completes the real-time mapping of multi-sensor multi-domain features to tool wear values; first, the data set is divided into a training set and a test set, so as to evaluate the performance of the model during the training process; second, the input features are transmitted to each neuron of the hidden layer; each hidden layer neuron calculates the univariate function output of each input variable and sums it up, and then performs a nonlinear transformation; third, the outputs of all hidden layer neurons are weighted and summed to obtain the final output; this process realizes the nonlinear combination of high-dimensional input through univariate functions to approximate the output of the target function; then, in the training stage, all parameters are randomly initialized, the difference between the model output and the true label is calculated, and thus the loss function is obtained; finally, the back propagation algorithm is used to update all parameters of the model to obtain the optimal performance monitoring model.

[0126] The tool wear monitoring model shown in formula (6) is used to establish a real-time nonlinear mapping between multi-sensor information and tool wear values, and its output provides reliable input for the subsequent prediction model; the present application uses the KAN model to realize this process;

[0127] y t = M(x t ) (6) ​

[0128] The K-N theorem shown in formula (7) reveals that any multivariate continuous function can be represented as the superposition of several univariate continuous functions, which provides a new way of thinking and direction for the deep learning framework; the traditional MLP uses linear transformation and fixed activation function combination on the neuron nodes to complete the nonlinear mapping; while KAN directly realizes the mapping between nodes through learnable nonlinear activation function on the neuron edges, and replaces the linear weight calculation with parameterized spline univariate function. This innovative way, although more complex than MLP in calculating the relationship between nodes, completely eliminates the problem of large number of parameters caused by the fully connected structure of MLP;

[0129]

[0130] Let n i represent the number of neurons in layer i th , let (l,i) represent the i th th neuron in layer l l,i , and its activated value is represented by x l ; then, among the n l+1 neurons between layers l and l+1, the activation function connecting (l,i) and (l+1,j) can be represented as formula (8); therefore, the relationship between the output and input of layer l th can be represented as formula (9);

[0131] φ l,j,i , l=0,···,L-1, i=1,···,n l , j=1,···,n l+1 (8)

[0132]

[0133] Let the coefficient matrix of x l be represented as Φ l , then the KAN network with L layers can be represented as formula (10); at this time, by comparing with the L-layer MLP structure of formula (11), it can be seen that KAN replaces the nonlinear mapping of learnable W and fixed σ of MLP with learnable Φ; moreover, MLP calculates the linear combination of each neuron of the previous layer on the neuron edge and performs nonlinear activation at the node; in contrast, KAN directly performs nonlinear activation on the neuron edge and then performs linear combination, thus greatly reducing the computational complexity and saving resources;

[0134]

[0135] Control Φ using formula (12-14), where the spline(x) function is parameterized as ci linear combination of B-splines; since the parameters are trainable and the transformation of KAN(x) between layers is differentiable, the KAN model can be trained by a backpropagation algorithm;

[0136] φ(x) = w(b(x) + spline(x)) (12)

[0137] b(x) = silu(x / (1+e -x )) (13)

[0138]

[0139] To verify the effectiveness of the KAN model, the KAN model is compared with other advanced models, as shown in Tables 4 and 5, respectively; it can be seen that the KAN tool wear real-time monitoring model has smaller error and higher accuracy than other advanced models. C1, C4 and C6 are three existing public data sets. The experimental data set is a data set constructed by determining the wear value based on the wear area segmentation model provided by the application.

[0140] Table 4: Prediction indicators of the monitoring model and other advanced models on the public data test set

[0141]

[0142] Table 5: Prediction indicators of the monitoring model and other advanced models on the experimental data test set

[0143]

[0144]

[0145] Example 6

[0146] The tool wear prediction model is a Transformer model; the Transformer tool wear prediction model completes the mapping of the historical to future tool wear sequence; first, the input features and label data are appropriately processed and formatted, and are divided into a training set and a test set; second, the input data is reshaped into a three-dimensional format [batch size, sequence length, features]; this is because the Transformer model is designed based on sequence processing, and the time step and feature dimension of the sequence need to be specified; then, the model is constructed with an input layer, a linear embedding layer, a multi-head attention layer, a feedforward neural network layer, a global average pooling layer, a dropout layer and an output layer, the model output is calculated and the loss is calculated with the label; finally, the parameters of the model are updated iteratively through a backpropagation algorithm to obtain an optimal parameter prediction model.

[0147] The tool wear prediction model shown in equation (15) is used to mine the data y of the historical T1 time series provided by the monitoring model (t-T1)：t The relationship between the future T2 time series data y t.(t+T2) ; this process is implemented using a Transformer model;

[0148] y t:(t+T2) = P(y (t-T1):t ) (15)

[0149] First, assume that the model input tensor is where T represents the sequence length, N represents the batch size, and d represents the input feature dimension; since the purpose is to realize the mapping between historical and future wear data, the input feature dimension input = 1; and the Transformer usually needs high-dimensional input to better capture the complex relationship between features, so the X input to the Transformer needs to be mapped to a high-dimensional space first; after the change shown in equation (16), the output tensor X0 is

[0150]

[0151] Then X0 is input into the Transformer encoder composed of multi-head attention mechanism and feedforward neural network; when performing multi-head attention calculation, for the lth layer, the self-attention calculation of each head h is as shown in equation (17), where WQ, WK, and WV represent the transformation matrices of query, key, and value, respectively;

[0152]

[0153] Each Transformer encoder includes a feedforward neural network as shown in equation (18), and the output of the lth layer can be represented as shown in equation (19); after L layers, the output of the encoder can be represented as shown in equation (20); in the output X L of the Transformer encoder, the representation of the last time step T is selected, i.e., X L [-1], and the prediction result can be obtained by regression through the decoder, as shown in equation (21);

[0154] FFN(X) = ReLU(XW1 + b1)W2 + b2 (18)

[0155] X l+1 = LayerNorm(FFN(Attention(X l )) + X l (19)

[0156] XL = TransformerEncoder(X0) (20)

[0157]

[0158] To verify the effectiveness of the Transformer tool wear prediction model, first verify its performance in the traditional prediction mode. Its prediction indicators with other advanced models on the test set of public and experimental data are shown in Tables 6 and 7 respectively; it can be seen that the Transformer prediction model has better prediction performance.

[0159] Table 6 Prediction indicators of prediction models and other advanced models on the test set of public data

[0160]

[0161] Table 7 Prediction indicators of prediction models and other advanced models on the test set of experimental data

[0162]

[0163] Example 7

[0164] The advanced prediction mode; Figure 9 The difference between the traditional tool wear prediction and the advanced prediction mode is shown; the traditional tool wear prediction mode relies on historical true data, that is, the monitoring model needs to provide historical true wear values for each prediction; in this mode, the wear values at T1 time points need to be calculated to serve as the input of the prediction model; therefore, the monitoring model needs to provide T1*N historical wear data for N times of prediction of the prediction model, which seriously affects the real-time performance of the prediction model;

[0165] In contrast, the advanced prediction mode feeds back the prediction model output to the input end and fuses it with the historical true wear value data to update the model input for the next prediction, thereby making continuous predictions; in this mode, the model can theoretically make infinite recursive predictions based on a set of model inputs, and can make long-term predictions; however, as the number of predictions increases, the cumulative error cannot be ignored, so it is necessary to use the historical tool wear value to correct the model input in time after a certain number of predictions.

[0166] That is, using the trained wear prediction model to predict the wear value of the tool specifically includes:

[0167] When the wear prediction model is predicted for the first time, the input is the monitoring value sequence of the tool wear value, and the output is the predicted value sequence of the tool wear value;

[0168] The monitoring value of the tool wear value is the output value of the wear monitoring model.

[0169] The monitoring value of the tool wear value is the output value of the wear monitoring model.

[0170] In order to verify the effectiveness of the prediction model in the leading mode, the prediction was carried out on the public and experimental data test sets under different conditions at T1 and T2, and the results are shown in Tables 8 and 9, respectively.

[0171] Table 8 Prediction index of the prediction model in the leading mode on the public data test set

[0172]

[0173]

[0174] Table 9 Prediction index of the prediction model in the leading mode on the experimental data test set

[0175]

[0176] Example 8

[0177] The leading prediction mode further comprises:

[0178] The timely correction strategy is to compare the prediction value sequence of the tool wear value output by the wear prediction model with the corresponding monitoring value sequence of the tool wear value, and obtain the error between the prediction value and the monitoring value.

[0179] If the error is less than the preset threshold, the online prediction is continued in the leading prediction mode; if the error is greater than or equal to the preset threshold, the monitoring value sequence of the tool wear value is input into the wear prediction model for the next prediction to correct the model input; then the online prediction is continued in the leading prediction mode. In the machining process, the input of the wear prediction model can be corrected once every preset time interval or every preset number of tools.

[0180] Specifically, the timely correction strategy is that the accumulation of prediction errors inevitably leads to the decline of prediction accuracy in the leading prediction mode; therefore, the timely correction strategy is proposed, and the principle is shown in Table 10. i T The real historical sequence at the i-th time point is used as the model input for prediction. i P The output of the previous prediction is spliced with the historical sequence as the model input for prediction at the i-th time point. irepresents the prediction result output at the i-th time point, P represents the prediction model, and △ represents the MAE value between the prediction result at the i-th time point and the real wear sequence; when T1 = 3 and T2 = 1, the first prediction occurs at the fourth time point, at which time the model obtains Y4' according to the historical sequence S4 T Y4' is obtained by prediction; in subsequent prediction, prediction can be performed in a look-ahead prediction mode, and then the MAE between the previous prediction result and the real value is calculated every interval preset time or interval preset tool number, if less than the error threshold δ, then the input of the next prediction uses the last prediction result to fuse the historical data as the input; on the contrary, if the MAE of the last prediction is greater than δ, the real historical wear data is used as the input for the next prediction, so as to avoid the accumulation of errors in time; wherein the real historical sequence of wear value can be the output value of the wear monitoring model.

[0181] Table 10 and the principle of the timely correction strategy

[0182]

[0183] Specifically, in the online prediction process, prediction can be performed every interval time, and each prediction can be performed according to the tool number or the machining time. Taking the prediction according to the tool number as an example: the input of the wear prediction model at each prediction is the wear value monitoring value sequence corresponding to the machining of a plurality of tools before the current time, the wear value monitoring value and prediction value fusion sequence or the wear value prediction value sequence, and the output is the wear value prediction value corresponding to the machining of at least one tool after the current time.

[0184] The wear value monitoring value can be obtained based on the wear monitoring model according to the monitoring signal of the collected sensor after the corresponding tool machining. The wear value monitoring value used for error judgment with the wear value prediction value in the timely correction can also be obtained in the same way, that is, it can be obtained based on the wear monitoring model according to the monitoring signal collected online when the machining tool number corresponding to the wear value prediction value.

[0185] In order to verify the effectiveness of the timely correction strategy in the look-ahead mode, the correction strategy is used for prediction on the public and experimental data test sets respectively under different conditions of T1 and T2, and the results are shown in Tables 11 and 12 respectively; it can be seen that after using the timely correction strategy, both the error and the accuracy are obviously improved.

[0186] Table 11 Prediction index of the prediction model using the correction strategy in the look-ahead mode on the public data test set

[0187]

[0188] Table 12 Prediction index of the prediction model using the correction strategy in the look-ahead mode on the experimental data test set

[0189]

[0190] It is to be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will be apparent to those of skill in the art upon reading and understanding the above description. The scope of the application should, therefore, be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. A real-time monitoring and prediction method for tool wear based on multi-sensor information fusion, characterized in that: include: Offline stage: S1, conduct a machining experiment and collect monitoring signals obtained by multiple sensors and tool wear images obtained by a camera during the machining experiment; S2, using a trained wear area segmentation model to segment the wear area of ​​the tool wear image, and calculating and obtaining a true wear value based on the segmented wear area; S3, establishing a monitoring training sample set using the monitoring signal and the corresponding real wear value, and training a wear monitoring model using the monitoring training sample set to obtain a trained wear monitoring model; Using the sequence of real wear values ​​to establish a prediction training sample set, using the prediction training sample set to train a wear prediction model, and obtaining a trained wear prediction model; Online stage: S4, during the machining process, using multiple sensors to obtain real-time monitoring signals; using the real-time monitoring signals and the trained wear monitoring model to monitor the wear value of the tool in real time; and using the trained wear prediction model to predict the wear value of the tool; The wear prediction model that has been trained is used to predict the wear value of the tool, specifically including: The wear prediction model inputs a monitoring value sequence of tool wear values ​​during the first prediction, and outputs a prediction value sequence of tool wear values; Then, the wear prediction model inputs a combination of the monitoring value of the tool wear value and the predicted value output from the previous prediction, or a sequence of predicted values ​​output from the previous prediction, and outputs a sequence of predicted values ​​of the tool wear value, forming an advanced prediction mode. Wherein, the monitoring value of the tool wear value is the output value of the wear monitoring model; The advanced forecasting model also includes: Timely correction strategy: comparing the predicted value sequence of tool wear values ​​output by the wear prediction model with the corresponding monitored value sequence of tool wear values ​​to obtain the error between the predicted value and the monitored value; If the error is less than the preset threshold, online prediction will continue in the advance prediction mode; if the error is greater than or equal to the preset threshold, when the wear prediction model makes the next prediction, the monitoring value sequence of the tool wear value will be input to correct the model input; thereafter, online prediction will continue in the advance prediction mode.

2. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to claim 1, characterized in that: Before training the wear monitoring model using the monitoring training sample set in S3, the following steps are also included: Performing noise reduction and feature extraction on the monitoring signal to obtain preprocessing features; when training a wear monitoring model, the preprocessing features serve as input to the wear monitoring model, and the output of the wear monitoring model is a wear value prediction value; The noise reduction process on the monitoring signal includes: performing at least one of average filtering, wavelet transform, inverse wavelet transform, and dead-load removal processing on the monitoring signal; Performing feature extraction processing on the monitoring signal includes: Multi-domain features are extracted from the monitoring signal after noise reduction processing from the time domain, frequency domain, and time-frequency domain; the wear-strong correlation features are screened out from the multi-domain features using the Pearson correlation coefficient method; the high-dimensional features of the wear-strong correlation features are mapped to a low-dimensional space using the KPCA method to obtain preprocessing features.

3. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to claim 2, characterized in that: In the feature extraction process for the monitoring signal, before mapping the high-dimensional features of the wear-strong correlation features to a low-dimensional space using the KPCA method, the method further includes: An exponential filtering operation is performed on the wear-strongly correlated features.

4. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to claim 2, characterized in that: Performing noise reduction and feature extraction on the monitoring signal to obtain preprocessing features also includes: The monitoring signals obtained by multiple sensors are subjected to noise reduction processing and multi-domain feature extraction respectively, and then the multi-domain features corresponding to the monitoring signals obtained by multiple sensors are fused, and then the Pearson correlation coefficient method is used based on the fused multi-domain features to screen out the wear-strong correlation features.

5. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to any one of claims 1 to 4, characterized in that: The wear area segmentation model in S2 is a YOLOv8 segmentation model; by labeling the wear area of ​​the tool wear image obtained in S1, a segmentation sample set is constructed, and the wear area segmentation model is trained using the segmentation sample set.

6. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to any one of claims 1 to 4, characterized in that: In S2, the actual wear value is calculated based on the segmented wear area, specifically including: Obtain pixel coordinates of vertices in the wear area to form a vertex point set; Traversing the vertex point set to obtain equations of line segments between any two vertices; and determining the original cutting edge line segment of the tool by screening the slope of the line segment and the number of vertices it passes through; Calculate the vertical distances between the remaining points of the vertex point set and the original cutting edge line segment of the tool, and select the value with the largest vertical distance as the number of pixels of the maximum wear width; According to the number of pixels of the maximum wear width and the size of the tool wear image, the actual value of the maximum wear width is obtained as the true wear value.

7. The method for real-time monitoring and prediction of tool wear based on multi-sensor information fusion according to any one of claims 1 to 4, characterized in that: The multiple sensors in S1 include multiple vibration sensors, cutting force sensors and acoustic emission sensors; the wear monitoring model in S3 is a KAN deep learning model; and the wear prediction model in S3 is a Transformer model.

8. A real-time monitoring and prediction system for tool wear based on multi-sensor information fusion, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it executes the real-time monitoring and prediction method for tool wear based on multi-sensor information fusion as described in any one of claims 1 to 7.

Images