Double-metal milling cutter wear monitoring method and system considering material attribute difference

Abstract

The invention discloses a bimetallic milling cutter wear monitoring method and system considering material attribute differences, and relates to the technical field of bimetallic material machining, the method comprises the following steps: collecting milling force and vibration signals of a bimetallic vertical milling cutter in a milling process in real time; the collected milling force and vibration signals are input into an unsupervised material recognition model, and after feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering; and inputting the collected milling force, the vibration signal and the corresponding metal material category into a physical guidance neural network model based on a wear mechanism, and outputting the wear value of the rear cutter surface of the bimetallic vertical milling cutter. According to the method, the difference between the two metal materials is considered, the abrasion mechanism guidance is introduced, the monitoring model with higher accuracy and better robustness is obtained through an abrasion physical-data fusion monitoring mode, and the accuracy of monitoring the abrasion value of the rear tool face of the tool in the bimetallic end milling scene is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bimetallic material processing, and in particular to a bimetallic milling cutter wear monitoring method and system taking into account material property differences. Background Art

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

[0003] Bimetallic parts are often superior to parts made from each of their two components separately, so bimetallic materials play an important role in many engineering fields such as aircraft, automobiles, power generation and construction equipment. In order to meet the design and performance standards, bimetallic blanks require extremely precise cutting processes to be processed into bimetallic parts, and the tool is the key to achieving high-speed cutting. During high-speed cutting, the cutting temperature rises sharply, the impact and vibration of the tool gradually increase, and the tool wear accelerates, resulting in a decrease in the surface quality of the workpiece, causing fractures and chipping. Existing research shows that tool monitoring technology can increase processing efficiency by 10% to 50% and reduce production costs by 10% to 40%. Therefore, it is of great significance to accurately monitor tool wear.

[0004] In the actual bimetal cutting tool wear scenario, the existing wear monitoring research still faces many difficulties, such as: First, in the modeling process of bimetal parts cutting tool wear status monitoring, the difference between the two metal materials is not considered, and they are usually directly processed as a single material, resulting in a large error in the final wear monitoring; secondly, although the wear monitoring method based on the pure physical model is less affected by the abnormal points in the training set, its monitoring error is large. Compared with the pure physical model, the pure data-driven method has better accuracy, but this method is easily affected by abnormal points, which in turn affects the final monitoring accuracy; in addition, for the data-driven method, the data set is scarce, and the existing data set only focuses on single metal cutting, but lacks attention to bimetal cutting tool wear. There is currently no public data set suitable for bimetal end milling tool wear.

[0005] In recent years, with the continuous development of computer computing power, the end milling tool wear monitoring method based on neural network has been widely used in regression tasks. Although this method has made a lot of achievements in the field of general tool wear monitoring, most of the existing bimetal milling wear monitoring based on neural network is a simple migration product of ordinary single metal tool wear monitoring, and no reasonable solutions are proposed for the various difficulties existing in the above bimetal milling wear monitoring. Summary of the invention

[0006] In order to address the deficiencies of the above-mentioned prior art, the present invention provides a bimetallic milling cutter wear monitoring method and system taking into account the difference in material properties. The differences between the two metal materials are taken into account and wear mechanism guidance is introduced. A monitoring model with higher accuracy and better robustness is obtained through wear physics-data fusion monitoring. The problem that the existing bimetallic end milling cutter wear monitoring method does not consider the impact of different material properties on wear monitoring and only performs wear monitoring through a pure data-driven method, which results in a large error in the final wear monitoring, is solved. The accuracy of tool back face wear value monitoring in the bimetallic end milling scenario is effectively improved.

[0007] In a first aspect, the present invention provides a method for monitoring wear of a bimetallic milling cutter taking into account differences in material properties.

[0008] A wear monitoring method for a bimetallic milling cutter considering differences in material properties, comprising:

[0009] Real-time acquisition of milling force and vibration signals of bimetal end milling cutters during milling;

[0010] The collected milling force and vibration signals are input into the unsupervised material recognition model. After feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering.

[0011] The collected milling force, vibration signal and corresponding metal material category are input into the physics-guided neural network model based on wear mechanism, and the back face wear value of the bimetal end milling tool is output. During the training process of the physics-guided neural network model, the loss between the actual measured wear value and the predicted wear value is used as the first loss, and the loss between the predicted wear rate based on the physical formula of wear rate and the actual wear rate is used as the second loss. The first and second losses are combined as the loss function.

[0012] In a further technical solution, the training process of the unsupervised material recognition model and the physical guided neural network model is:

[0013] Collecting milling force and vibration signals of a bimetal end milling cutter during milling under several different milling parameters; the different milling parameters include different spindle speeds, feeds per tooth, and cutting depths;

[0014] The collected milling force and vibration signals are segmented into effective signals, and each segment of the signal is labeled with the metal material category label and the end mill tool flank wear value label to construct a training set;

[0015] Constructing an unsupervised material recognition model and a physical guided neural network model; wherein the unsupervised material recognition model includes a feature extraction module, a feature dimension reduction module and a clustering module, and the physical guided neural network model includes a feature extraction module, a feature fusion module and a monitoring module;

[0016] The two constructed models are trained separately using the training set to obtain the trained unsupervised material recognition model and the physics-guided neural network model.

[0017] A further technical solution is to use the training set to train the physics-guided neural network model:

[0018] The physical formula of wear rate in the wear degradation process of end milling cutter is selected, and the unknown coefficients in the physical formula are fitted according to the actual measured wear rate;

[0019] The loss between the actual measured wear value and the predicted wear value is taken as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is taken as the second loss. The first and second losses are combined to construct a loss function for constraining the model output.

[0020] Based on the loss function constructed with physical guidance, the model is iteratively trained using the training set until the maximum number of iterations is reached and the model training is completed.

[0021] In a second aspect, the present invention provides a bimetallic milling cutter wear monitoring system that takes into account differences in material properties.

[0022] A bimetallic milling cutter wear monitoring system taking into account material property differences, comprising:

[0023] Signal acquisition module, used for real-time acquisition of milling force and vibration signals of bimetal end milling cutters during milling process;

[0024] The bimetal milling cutter wear material identification module is used to input the collected milling force and vibration signals into the unsupervised material identification model, and after feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering;

[0025] The bimetallic milling cutter wear detection module is used to input the collected milling force, vibration signal and corresponding metal material category into the physical guidance neural network model based on the wear mechanism, and output the back face wear value of the bimetallic end milling cutter; wherein, during the training process of the physical guidance neural network model, the loss between the actual measured wear value and the predicted wear value is used as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is used as the second loss, and the first and second losses are combined as the loss function.

[0026] In a third aspect, the present invention further provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned bimetallic milling cutter wear monitoring method taking into account material property differences when executing the executable instructions stored in the memory.

[0027] In a fourth aspect, the present invention further provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned bimetallic milling cutter wear monitoring method that takes into account the difference in material properties.

[0028] In a fifth aspect, the present invention also provides a computer program product, which includes executable instructions, and the executable instructions are stored in a computer-readable storage medium; wherein, when the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the above-mentioned bimetal milling cutter wear monitoring method that considers the difference in material properties is implemented.

[0029] One or more of the above technical solutions have the following beneficial effects:

[0030] The present invention provides a bimetallic milling cutter wear monitoring method and system considering the difference in material properties, considering the difference between two metal materials, introducing wear mechanism guidance, and obtaining a monitoring model with higher accuracy and better robustness through wear physics-data fusion monitoring; in the present invention, the milling force and vibration signals of the bimetallic end milling cutter are collected to construct a benchmark data set for bimetallic end milling cutter wear monitoring, and at the same time, an unsupervised material recognition model and a physical guidance network model suitable for bimetallic end milling cutter wear monitoring are constructed, and then the benchmark data set is used to train the unsupervised material recognition model and the physical guidance network model to obtain trained weights and biases, and finally the milling force and vibration signals of the bimetallic end milling cutter collected in real time are sequentially input into the above-mentioned material recognition model and the physical guidance network model, and the monitored real-time milling material category and accurate back tool wear value are output; the present invention solves the problem that the existing bimetallic end milling cutter wear monitoring method does not consider the influence of different material properties on wear monitoring, and only performs wear monitoring in a pure data-driven manner, thereby resulting in a large error in the final wear monitoring, and effectively improves the accuracy of tool back tool wear value monitoring in the bimetallic end milling scenario.

[0031] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] Figure 1 A schematic flow chart of a bimetallic milling cutter wear monitoring method considering material property differences proposed in an embodiment of the present invention;

[0034] Figure 2A schematic diagram of a segmentation signal valid interval in an embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of extracting multi-domain features from segmented multi-source valid signals in an embodiment of the present invention;

[0036] Figure 4 Schematic diagram of features after t-SNE dimension reduction in an embodiment of the present invention;

[0037] Figure 5 A schematic diagram of Kmeans clustering and confusion matrix in an embodiment of the present invention;

[0038] Figure 6 This is a schematic diagram of Kmeans clustering when there are few samples in an embodiment of the present invention;

[0039] Figure 7 A schematic diagram of a characteristic separation of bimetal in an embodiment of the present invention;

[0040] Figure 8 The wear-related thermodynamic diagram of the bimetallic feature before and after separation in the embodiment of the present invention;

[0041] Fig. 9 The wear-related thermal diagram before and after the weighted fusion of the bimetallic features in the embodiment of the present invention;

[0042] Fig.10 Schematic diagram of the weighted fusion process of cast iron (CI) No. 4 domain features in an embodiment of the present invention;

[0043] Fig.11 This is a schematic diagram of fitting the physical formula of wear rate in an embodiment of the present invention;

[0044] Fig.12 This is a schematic diagram of the wear monitoring effect with and without differentiating material differences according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that the following detailed descriptions are exemplary only, are intended to describe specific embodiments, are intended to provide further explanation of the present invention, and are not intended to limit exemplary embodiments according to the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art to which the present invention belongs. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0046] Embodiment 1

[0047] This embodiment provides a bimetallic milling cutter wear monitoring method taking into account the difference in material properties, such as Figure 1As shown, the method specifically comprises the following steps:

[0048] Step S101: collecting the milling force and vibration signals of the bimetal end milling cutter during the milling process in real time.

[0049] In this embodiment, the milling force and milling vibration signals of the end milling tool during the milling of the bimetallic material are collected in real time by the sensor.

[0050] Step S102: input the collected milling force and vibration signals into an unsupervised material recognition model, extract features and reduce feature dimensions, and output the metal material category of the current signal through clustering.

[0051] Step S103, input the collected milling force, vibration signal and corresponding metal material category into the physical guidance neural network model based on wear mechanism, and output the flank wear value of the bimetal end milling cutter. In the training process of the physical guidance neural network model, the loss between the actual measured wear value and the predicted wear value is taken as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is taken as the second loss, and the first and second losses are combined as the loss function.

[0052] In this embodiment, an unsupervised material recognition model and a physically guided neural network model based on wear mechanism are pre-built and trained, and then wear detection is performed through the above steps S102 and S103, that is, the unsupervised material recognition model is first used to identify the metal material category corresponding to the signal, and then, based on the identified metal material type, combined with the collected milling force and vibration signals, a more accurate wear monitoring result is output through the physically guided neural network model. Specifically, the training process of the above model is:

[0053] Firstly, the milling force and vibration signals of the bimetal end milling cutter during the milling process under several different milling parameters are collected; among which, different milling parameters include different spindle speeds, feeds per tooth and cutting depths, that is, the collected data covers the diverse milling forces and vibration signals under different spindle speeds, feeds per tooth and cutting depths.

[0054] Secondly, the collected milling force and vibration signals are segmented, and each segment of the signal is labeled with the metal material category label and the flank wear value label of the end milling tool to construct a training set.

[0055] Afterwards, an unsupervised material recognition model and a physics-guided neural network model are constructed.

[0056] Finally, the two constructed models were trained using the training set to obtain the trained unsupervised material recognition model and physics-guided neural network model.

[0057] Furthermore, the above unsupervised material recognition model includes a feature extraction module, a feature dimension reduction module and a clustering module, wherein:

[0058] (1) The feature extraction module is used to identify the effective interval of the input milling force and vibration signals and divide them into multiple effective segments, extract the time domain, frequency domain and time-frequency domain features of each effective segment in the signal, and generate a multi-domain feature matrix.

[0059] In the feature extraction module, such as Figure 2 As shown in FIG, the threshold method is used to identify and extract the effective interval of the input original signal, which is: a threshold is preset based on experience, and the tool entry point is determined based on the preset threshold, that is, when a value reaching this threshold is detected in the signal, the tool entry point can be determined; further, considering that the length of each knife signal used in the experiment is the same, the signal of a fixed length after the entry point is taken as the effective cutting interval (i.e., the effective interval); as shown in FIG. Figure 3 As shown, the effective interval in the original signal is extracted, and the effective interval of the original signal is divided into multiple effective segments, that is, the segments are divided into segments according to a time period of a fixed length, and the time domain, frequency domain and time-frequency domain features of each effective segment are extracted to generate a multi-domain feature matrix.

[0060] Preferably, in this embodiment, for each channel of each signal, 13 time domains, 11 frequency domains and 8 time-frequency domains obtained by wavelet decomposition are extracted, and a total of 32 domain features are extracted, and then a total of 32×6=192-dimensional feature matrices of 6 channels of the two signals of milling force and vibration signal are extracted. It should be noted that the NASA data set has only 2 single-channel signals. Among them, the extracted time domain features include maximum value, minimum value, average value, peak value, variance, standard deviation, kurtosis, skewness, root mean square, waveform factor, peak factor, pulse factor, margin factor; the extracted frequency domain features include frequency domain amplitude average value, frequency domain amplitude skewness, frequency domain amplitude kurtosis, centroid frequency, mean square frequency, frequency variance, root mean square frequency, frequency amplitude variance, frequency standard deviation, frequency band energy, relative power spectrum entropy; the extracted time-frequency domain features include the mean of 8 components obtained after the signal is decomposed by 3 layers of wavelet packets.

[0061] (2) The feature dimensionality reduction module is used to perform unsupervised dimensionality reduction on the extracted multi-domain feature matrix to generate a low-dimensional feature matrix.

[0062] In the feature dimensionality reduction module, an unsupervised dimensionality reduction method based on t-SNE is used to reduce the dimensionality of the extracted multi-domain high-dimensional feature matrix. In this embodiment, the dimensionality is reduced to 2 dimensions through t-SNE unsupervised dimensionality reduction to obtain a 2-dimensional feature matrix. Figure 4 As shown, each point represents the 2D feature of a valid segment, and each point has a corresponding metal material category label, so as to obtain the 2D features of all valid segments in the input signal after dimensionality reduction.

[0063] (3) The clustering module is used to identify and output the metal material category based on the low-dimensional feature matrix after dimensionality reduction.

[0064] In the clustering module, based on the low-dimensional feature matrix of each valid segment, the Kmeans clustering method is used to automatically identify the metal material category of each valid segment in the input original signal, such as Figure 5 As shown, after clustering recognition, the categories are divided into two categories: metal 1 and metal 2. Preferably, according to the category division results of clustering recognition, a confusion matrix can be constructed, which is a matrix that marks the number of correct and incorrect classification results for each category. Figure 5 It can be seen from the confusion matrix shown that the clustering method adopted in this embodiment has a higher category recognition accuracy.

[0065] Specifically, the 2D feature matrix after dimension reduction is input into the Kmeans classifier, and the clustering is specified as 2 categories for training. The clusterer is used to obtain the material category of each fragment, where the clustering process is:

[0066] (3.1) Initialization: Select K initial cluster centers; in this embodiment, K=2;

[0067] (3.2) Assignment, assign each data point to the nearest cluster center: For each data point in the data set, calculate its distance to each cluster center and assign it to the cluster center with the nearest distance;

[0068] (3.3) Update, recalculate the center of each cluster;

[0069] (3.4) Iteration: Repeat steps (3.2) and (3.3) until the maximum number of iterations is reached and output the final clustering result.

[0070] It should be noted that the milling force and vibration data, material category and wear label in this embodiment are obtained by segmenting the data in the NASA milling data set. Metal 1 is cast iron (CI) and metal 2 is steel (Steel). In the actual milling tool wear monitoring process, in order to ensure real-time performance, it is impossible to always cut off at the material boundary when dividing the segments. Therefore, there is always a segment that contains both a part of the metal 1 signal and a part of the metal 2 signal, which is called a transition segment. Figure 6 As shown in , when the amount of data is small, clustering can easily divide the transition segment, but for most features, the feature value of the transition segment is between the segments of the other two classes. In other words, Figure 5As shown, when the sample size is large, clustering cannot easily separate the transition segments, and since the number of transition segments is much smaller than that of the other two categories of segments, the above method adopted in this embodiment automatically classifies the transition segments into one of the two categories of metal 1 and metal 2 by the clustering model, and will not cause a large error in the recognition result.

[0071] Furthermore, the physical guided neural network model includes a feature extraction module, a feature fusion module and a monitoring module, wherein:

[0072] (4) Feature extraction module: This module is the same as the feature extraction module in the unsupervised material recognition model and will not be described in detail here.

[0073] (5) The feature fusion module is used to split the multi-domain feature matrix of the extracted input signal according to the metal material category of each valid segment in the identified input signal, and perform weighted fusion on the two split feature matrices to obtain a low-dimensional fused feature matrix.

[0074] First, if Figure 7 As shown, according to the clustering module, the metal material category label of each valid segment in the signal is identified and obtained. At this time, it can be determined which metal material the end mill is cutting at a certain time, and then the bimetallic feature can be separated or split into feature 1 when cutting metal 1 and feature 2 when cutting metal 2 according to the category label, and then feature matrix 1 and feature matrix 2 can be obtained. Specifically, a clusterer is used to obtain the metal material category of each segment, so as to distinguish the values ​​belonging to different materials in each feature of the input signal, and obtain feature matrix 1 for cutting metal 1 and feature matrix 2 for cutting metal 2, so that they can be input into the wear value monitoring model for monitoring later. Figure 8 The wear-related heat map shown in the figure, in which each matrix represents a feature, and the depth of the feature color is determined by the Pearson correlation coefficient between the feature and wear. For most of the correlation coefficients between features and wear, the correlation coefficients of the two separated feature matrices are larger, that is, the latest feature matrix is ​​more correlated with the milling cutter wear value.

[0075] Secondly, for the two feature matrices obtained above, such as Fig. 9 As shown in the figure, weighted fusion is performed on each channel and domain to obtain two new low-dimensional feature matrices. The weight of each feature is calculated based on the Pearson correlation coefficient of each feature and the wear value on each channel and domain, the weighting factor, and the inverse Tanh weighting function. All features are weighted and fused into fusion features in each domain, that is, a low-dimensional fusion feature matrix is ​​obtained.

[0076] Specifically, Fig. 9As shown, the features are weighted and fused on each channel and domain. The signal of each channel extracts the features of 3 domains, and weighted fusion is performed in each domain to obtain 3 fused features. Since 6 channels are set in this embodiment, 6×3=18 fused features are finally obtained. In this way, two 192-dimensional feature matrices are fused to obtain two 18-dimensional feature matrices. It should be noted that the NASA data set is only 6-dimensional. The above feature fusion process is as follows Fig.10 As shown, the size of the weight is determined according to the correlation coefficient x between each feature and wear in the same domain, and the inverse tanh function is used to determine the size relationship of each weight and to convert negatively correlated features into positively correlated features. The weighting factor a is an adjustable positive odd number. The optimal point is found through iteration, that is, the weighting factor a is randomly taken. In this embodiment, a=1 is taken, and it is iterated and added from a=1 until the correlation coefficient of the fusion feature is maximized, and the value of the weighting factor a is determined; on this basis, the weight is determined according to the correlation coefficient of each feature and wear in the current domain, and the inverse tanh function and the weighting factor are combined to perform weighted fusion to obtain the fusion feature corresponding to the domain. Compared with the attention mechanism, which is a weighting method, the weighted fusion method proposed in this embodiment can assign weights to each feature, and compared with the filtering method, the method proposed in this embodiment can retain more effective components.

[0077] (6) The monitoring module is used to output the monitored tool wear value based on the fused low-dimensional fusion feature matrix.

[0078] Furthermore, in this embodiment, on the basis of considering the difference between the two metal materials, the wear mechanism guidance is also introduced, and a monitoring model with higher accuracy and better robustness is obtained through wear physics-data fusion monitoring. Fig.11 As shown in Figure 2, the training process of the physical guided neural network model is:

[0079] Firstly, the physical formula of wear rate in the wear degradation process of end milling cutter is selected, and the unknown coefficients in the physical formula are fitted according to the actual measured wear rate.

[0080] In this embodiment, the physical formula of wear rate is first selected as:

[0081]

[0082] In the above formula, t represents time, a, b, c are variable parameters, i.e., unknown coefficients; VB represents the wear value.

[0083] The coefficients to be determined in the above physical formula are determined by fitting it with the wear rate obtained based on the actual measured wear value. Specifically, the actual measured wear value is completed by cubic spline interpolation, and the wear rate is obtained after differentiation, and then the coefficients to be determined in the selected wear rate formula are determined by least square fitting. Among them, the wear rate formula selected in this embodiment can accurately describe the three stages of wear, that is, the wear rate satisfies the law of changing from large to small and then large, thereby ensuring the accuracy of the final wear monitoring.

[0084] Secondly, the loss between the actual measured wear value and the predicted wear value is taken as the first loss, and the loss between the predicted wear rate based on the physical formula of wear rate and the actual wear rate is taken as the second loss. The first and second losses are combined to construct the loss function of the constraint model output.

[0085] Specifically, the loss between the predicted wear rate y1 and the actual wear rate y (i.e., |y1-y|) obtained according to the above wear rate physical formula is taken as a physical loss term and added to the loss function of the ANN network (the loss function is the loss related to the wear amount) to obtain a PINN model guided by the wear mechanism.

[0086] Finally, based on the loss function constructed under physical guidance, the model is iteratively trained using the training set until the maximum number of iterations is reached and the model training is completed.

[0087] like Fig.12 As shown, the wear monitoring effect of distinguishing material differences proposed in this embodiment is compared with that of not distinguishing material differences. The comparison of the monitoring results and errors of the wear values ​​further proves the great improvement in accuracy of the method proposed in this embodiment and the accuracy of PINN. At the same time, due to the use of a monitoring method that combines physics and data, PINN has better robustness.

[0088] Embodiment 2

[0089] This embodiment provides a bimetallic milling cutter wear monitoring system taking into account material property differences, specifically comprising:

[0090] Signal acquisition module, used for real-time acquisition of milling force and vibration signals of bimetal end milling cutters during milling process;

[0091] The bimetal milling cutter wear material identification module is used to input the collected milling force and vibration signals into the unsupervised material identification model, and after feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering;

[0092] The bimetallic milling cutter wear detection module is used to input the collected milling force, vibration signal and corresponding metal material category into the physical guidance neural network model based on the wear mechanism, and output the back face wear value of the bimetallic end milling cutter; wherein, during the training process of the physical guidance neural network model, the loss between the actual measured wear value and the predicted wear value is used as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is used as the second loss, and the first and second losses are combined as the loss function.

[0093] Embodiment 3

[0094] This embodiment provides an electronic device, including: a memory, used to store executable instructions; and a processor, used to implement the above method provided in this embodiment when executing the executable instructions stored in the memory.

[0095] Embodiment 4

[0096] This embodiment also provides a computer-readable storage medium storing executable instructions. When the executable instructions are executed by a processor, the processor will be caused to execute the above method provided in this embodiment.

[0097] Embodiment 5

[0098] This embodiment provides a computer program product, which includes an executable instruction, which is a computer instruction; the executable instruction is stored in a computer-readable storage medium. When a processor of an electronic device reads the executable instruction from the computer-readable storage medium and the processor executes the executable instruction, the electronic device executes the above method provided in this embodiment.

[0099] The steps involved in the above embodiments 2 to 5 correspond to the method embodiment 1. For the specific implementation, please refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0100] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0101] The above description is only a preferred embodiment of the present invention. Although the specific implementation mode of the present invention is described in conjunction with the accompanying drawings, it is not a limitation of the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the protection scope of the present invention.

Claims

1. A bimetallic milling cutter wear monitoring method considering material property differences, characterized in that: include: Real-time acquisition of milling force and vibration signals of bimetal end milling cutters during milling; The collected milling force and vibration signals are input into the unsupervised material recognition model. After feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering. The collected milling force, vibration signal and corresponding metal material category are input into the physics-guided neural network model based on wear mechanism, and the back face wear value of the bimetal end milling tool is output. During the training process of the physics-guided neural network model, the loss between the actual measured wear value and the predicted wear value is used as the first loss, and the loss between the predicted wear rate based on the physical formula of wear rate and the actual wear rate is used as the second loss. The first and second losses are combined as the loss function.

2. A bimetallic milling cutter wear monitoring method considering material property differences as claimed in claim 1, characterized in that: The training process of the unsupervised material recognition model and the physical guided neural network model is: Collecting milling force and vibration signals of a bimetal end milling cutter during milling under several different milling parameters; the different milling parameters include different spindle speeds, feeds per tooth, and cutting depths; The collected milling force and vibration signals are segmented into effective signals, and each segment of the signal is labeled with the metal material category label and the end mill tool flank wear value label to construct a training set; Constructing an unsupervised material recognition model and a physical guided neural network model; wherein the unsupervised material recognition model includes a feature extraction module, a feature dimension reduction module and a clustering module, and the physical guided neural network model includes a feature extraction module, a feature fusion module and a monitoring module; The two constructed models are trained separately using the training set to obtain the trained unsupervised material recognition model and the physics-guided neural network model.

3. A bimetallic milling cutter wear monitoring method considering material property differences as claimed in claim 2, characterized in that: The feature extraction module is used to identify the effective interval of the input milling force and vibration signal and divide it into multiple effective segments, extract the time domain, frequency domain and time-frequency domain features of each effective segment in the signal, and generate a multi-domain feature matrix; wherein, in the feature extraction module, a threshold method is used to identify and extract the effective interval of the input original signal, the effective interval of the original signal is divided into multiple effective segments, and the time domain, frequency domain and time-frequency domain features of each effective segment are extracted to generate a multi-domain feature matrix; The feature dimension reduction module is used to perform unsupervised dimension reduction on the extracted multi-domain feature matrix to generate a low-dimensional feature matrix; wherein, in the feature dimension reduction module, the extracted multi-domain high-dimensional feature matrix is ​​subjected to t-SNE unsupervised dimension reduction to obtain a low-dimensional feature matrix; The clustering module is used to identify and output the metal material category based on the low-dimensional feature matrix after dimensionality reduction; wherein, in the clustering module, based on the low-dimensional feature matrix of each effective segment, the Kmeans clustering method is used to automatically identify the metal material category of each effective segment in the input original signal.

4. A bimetallic milling cutter wear monitoring method considering material property differences as claimed in claim 2, characterized in that: The feature fusion module is used to split the multi-domain feature matrix of the extracted input signal according to the metal material category of each valid segment in the identified input signal, and perform weighted fusion on the two split feature matrices to obtain a low-dimensional fused feature matrix; The monitoring module is used to output the monitored tool wear value according to the fused low-dimensional fusion feature matrix.

5. A bimetallic milling cutter wear monitoring method considering material property differences as claimed in claim 4, characterized in that: In the feature fusion module, the signal features are first split into feature matrix 1 when cutting metal 1 and feature matrix 2 when cutting metal 2 according to the metal material category with each small fragment in the identified signal; Then, according to the Pearson correlation coefficient between each feature and the wear value in each channel and domain, as well as the weighting factor and the inverse Tanh weighting function, the weight of each feature is calculated, and all the features are weighted and fused into fusion features in each domain, that is, a low-dimensional fusion feature matrix is ​​obtained.

6. A bimetallic milling cutter wear monitoring method considering material property differences as claimed in claim 1, characterized in that: The process of training the physics-guided neural network model using the training set is: The physical formula of wear rate in the wear degradation process of end milling cutter is selected, and the unknown coefficients in the physical formula are fitted according to the actual measured wear rate; The loss between the actual measured wear value and the predicted wear value is taken as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is taken as the second loss. The first and second losses are combined to construct a loss function for constraining the model output. Based on the loss function constructed with physical guidance, the model is iteratively trained using the training set until the maximum number of iterations is reached and the model training is completed.

7. A bimetallic milling cutter wear monitoring system taking into account material property differences, characterized in that: include: Signal acquisition module, used for real-time acquisition of milling force and vibration signals of bimetal end milling cutters during milling process; The bimetal milling cutter wear material identification module is used to input the collected milling force and vibration signals into the unsupervised material identification model, and after feature extraction and feature dimension reduction, the metal material category of the current signal is output through clustering; The bimetallic milling cutter wear detection module is used to input the collected milling force, vibration signal and corresponding metal material category into the physical guidance neural network model based on the wear mechanism, and output the back face wear value of the bimetallic end milling cutter; wherein, during the training process of the physical guidance neural network model, the loss between the actual measured wear value and the predicted wear value is used as the first loss, and the loss between the predicted wear rate based on the physical formula of the wear rate and the actual wear rate is used as the second loss, and the first and second losses are combined as the loss function.

8. An electronic device, characterized in that: include: A memory for storing executable instructions; The processor is used to implement a bimetallic milling cutter wear monitoring method taking into account material property differences as described in any one of claims 1 to 6 when executing executable instructions stored in the memory.

9. A computer-readable storage medium, characterized in that: Executable instructions are stored, which are used to cause the processor to execute the executable instructions to implement the bimetallic milling cutter wear monitoring method taking into account material property differences as described in any one of claims 1-6.

10. A computer program product, characterized in that The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the bimetal milling cutter wear monitoring method considering material property differences as described in any one of claims 1 to 6 is implemented.

Images