Tool wear prediction method based on physical information self-supervised learning with less labels

Abstract

The tool wear prediction method based on physical information self-supervised learning under few labels belongs to the technical field of tool wear state prediction of numerical control machining centers. The method comprises the following steps: collecting monitoring signals such as cutting force, vibration and acoustic emission in the machining process; performing pretreatment operations such as intercepting and Z-score standardization on each sample; designing a mixed residual convolutional neural network tool wear monitoring model to perform information fusion and high-dimensional representation extraction on multi-source sensor signals; using an auxiliary task to pretrain the tool wear monitoring model, guiding the model to extract the invariance features of the signals; and under the condition of labeled data scarcity, the tool wear prediction method realizes high-precision prediction and excellent generalization ability of tool wear, can effectively reduce the fluctuation range of prediction error, and maintains the stability of the prediction result. In the model training, the tool wear prediction method effectively utilizes a small amount of labeled data, ensures the stability and reliability of tool wear monitoring, and meets the requirements of monitoring accuracy and adaptability in actual production.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tool wear state prediction of numerical control machining center, and particularly relates to a tool wear amount prediction method based on physical information self-supervised learning under few-marking. BACKGROUND

[0002] Manufacturing plays a vital role in the modern economy, not only as a key engine of economic growth, but also as an important pillar of innovation, employment and social development. With the close integration of manufacturing and new generation information technology, intelligent manufacturing is becoming a universal and key trend, triggering a profound change in manufacturing methods and becoming one of the core driving forces of the deep integration of informatization and industrialization. Among them, the state monitoring of key components in the metal cutting process is a key construction direction to promote the intelligentization of the manufacturing process. As the cutting execution element directly acting on the workpiece, the tool will inevitably be worn by the combined action of different load factors when removing the workpiece material, causing tool degradation and accompanying phenomena such as increased cutting force, increased processing area temperature, and intensified vibration. This will have a negative impact on the dimensional accuracy and surface quality of the product, increase the risk of part scrap, increase additional production costs, and even cause unexpected downtime of the machine tool, resulting in unnecessary economic losses. Therefore, timely and effective monitoring of the tool wear state is crucial to ensure the reliability of product quality, improve production efficiency and reduce safety risks, and provides important support for the development of intelligent manufacturing.

[0003] The invention patent with authorization number CN112757053B discloses a model fusion tool wear monitoring method and system based on power and vibration signals, which includes: S1: signal acquisition and screening, the power signal and vibration signal of the tool spindle are synchronously sampled, the PLC of the machine tool is synchronously sampled to obtain the real-time processing information of the machine tool, the processing information is used to screen the power signal and the vibration signal, and the power signal and the vibration signal of the tool to be monitored during processing are obtained; S2: signal noise reduction, threshold values are set for the power signal and the vibration signal of the processing section signal, and high-frequency region signals are removed by threshold filtering to reduce noise; S3: feature calculation, the features of the power signal and the vibration signal after noise reduction are calculated to obtain an original feature dataset; S4: feature dimension reduction, the importance scores of all features in the original feature dataset are calculated by using a random forest algorithm, the obtained scores are arranged in descending order, a rejection ratio is determined, and part of the features are rejected according to the rejection ratio to obtain a new feature set; S5: data division, the data of the new feature set is divided into three parts to serve as a training set, a validation set and a test set respectively; S6: model training, the data of the training set is input into a Catboost model and an extreme random tree model, parameters are adjusted, the tool wear monitoring model corresponding to the Catboost model and the extreme random tree model is trained respectively, a Stacking integrated algorithm is adopted, and the tool wear monitoring model corresponding to the Catboost model and the extreme random tree model is fused and learned; S7: model prediction, the features of the test set are predicted for tool wear loss according to the fused tool wear monitoring model; S8: synchronous update, the prediction result of the tool wear loss is subjected to Kalman filter noise reduction to obtain a tool wear prediction value.

[0004] The invention patent with authorization number CN111832432B discloses a tool wear real-time prediction method based on wavelet packet decomposition and deep learning, which includes: (1) synchronously collecting relevant sensor signals in the workpiece processing process, selecting a stable signal segment as a signal segment to be analyzed, and expanding the signal sample to be analyzed to increase the sample size; wavelet packet decomposition is performed on the signal to be analyzed to obtain a plurality of wavelet packet coefficient two-dimensional matrices; (2) the wavelet packet coefficient two-dimensional matrix is used as the input of a feature extraction CNN model block, and the one-dimensional feature matrix output by each feature extraction CNN model block is spliced into a longer one-dimensional matrix, and then feature fusion is performed and a two-layer fully connected network is established, thereby obtaining a convolutional neural network model; (3) inputting the signal data to be analyzed into the convolutional neural network model to predict the tool wear in real time.

[0005] The traditional method applied to tool residual life prediction generally needs feature extraction, feature selection and feature dimension reduction before model establishment, which all depend on certain experience and theory, and when the data preprocessing is not appropriate, important feature information will be lost, resulting in reduced accuracy of tool residual life prediction. SUMMARY

[0006] The purpose of the present application is to provide a tool wear prediction method based on physical information self-supervised learning with less labeling, which takes cutting force, vibration and acoustic emission signals as monitoring signals, realizes accurate prediction of tool wear in a two-stage self-supervised learning framework by combining hybrid residual convolutional neural network and fully connected regression layer, and overcomes the shortcomings in the prior art.

[0007] The technical scheme adopted by the present application is:

[0008] The tool wear prediction method based on physical information self-supervised learning with less labeling comprises the following steps:

[0009] S1. Collecting cutting force, vibration and acoustic emission signals during the machining process;

[0010] S2. Taking the collected signals as data samples, marking part of the samples, and performing preprocessing operations such as truncation and Z-score standardization on each sample;

[0011] S3. Designing a hybrid residual convolutional neural network tool wear monitoring model to perform information fusion and high-dimensional feature extraction on multi-source sensor signals;

[0012] S4. Based on the self-supervised learning framework, pre-training the tool wear monitoring model using auxiliary task one, guiding the model to extract the invariance features of the signals through enhanced view generation and twin network contrast loss calculation;

[0013] S5. Based on the self-supervised learning framework, pre-training the tool wear monitoring model using auxiliary task two, enhancing the physical consistency of the feature vectors by introducing physical constraints;

[0014] S6. Based on the self-supervised learning framework, gradually guiding the model to have both efficient feature extraction capability and physical law consistency by alternately optimizing auxiliary task one and auxiliary task two;

[0015] S7. Based on the self-supervised learning framework, fine-tuning the pre-trained tool wear monitoring model using a limited number of labeled samples, and obtaining the final wear value evaluation result;

[0016] S8. Evaluating the overall performance of the model in tool wear state monitoring task by calculating the model evaluation index.

[0017] Compared with the prior art, the present application has the following beneficial effects:

[0018] In the case of difficulty in obtaining tool wear value label marks, the present application makes full use of a large number of unlabeled tool cutting process monitoring signals to improve tool wear value prediction under the condition of wear mark limitation, so as to not only utilize the excellent nonlinear modeling capability of a deep neural network, but also reduce the dependence of the deep neural network on wear marks.

[0019] The present application introduces a regularization constraint based on physical information in a self-supervised learning framework, excludes deep network model parameters that do not conform to physical laws, effectively limits the feasible solution range of the model parameter space, improves the generalization capability of the pre-training features, and maintains the model generalization capability through partial parameter freezing in the fine-tuning process.

[0020] The present application realizes high-precision prediction and excellent generalization capability of tool wear in the case of scarce labeled data, can effectively reduce the fluctuation range of prediction error, and maintain the stability of the prediction result.

[0021] The present application effectively utilizes a small amount of labeled data in model training, ensures the stability and reliability of tool wear monitoring, and meets the requirements of monitoring accuracy and adaptability in actual production.

[0022] The present application has important significance for tool wear state monitoring and can provide highly adaptive and expandable prediction capability in actual application.

[0023] The model training of the present application is divided into two stages: in the first stage, a mixed residual convolutional neural network is pre-trained through representation learning of unlabeled data to optimize the signal feature extraction capability; in the second stage, the model is fine-tuned using limited labeled data to realize high-precision prediction of tool wear value through mean square error loss. Experimental results show that this method performs well in wear monitoring and residual life prediction, and has significant industrial application value. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of the present application;

[0025] Figure 2 is a network architecture diagram of the tool wear monitoring model of the present application;

[0026] Figure 3 is a network structure diagram of the auxiliary task one of the present application;

[0027] Figure 4 is a network structure diagram of the auxiliary task two of the present application;

[0028] Figure 5 is a tool wear monitoring model architecture and fine-tuning strategy diagram of the present application;

[0029] Figure 6 is a data preprocessing schematic diagram of a milling data set involved in the examples of the present application;

[0030] Figure 7 is a Pearson correlation analysis schematic diagram of tool wear monitoring involved in the examples of the present application;

[0031] Figure 8 is a labeled sample tool wear prediction diagram under different degrees of scarcity involved in the examples of the present application;

[0032] Figure 9 is a schematic diagram showing the change of evaluation indexes of six tool wear monitoring methods (ResNet, LSTM, XGBoost, SVR, KNN and PISSL) with the increase of training data percentage. DETAILED DESCRIPTION

[0033] In order to better understand the purpose, structure and function of the present application, the present application will be further described in detail below in combination with the drawings.

[0034] Deep learning can utilize massive unprocessed data to learn features autonomously through deep networks. With the wide application of tool condition monitoring technology, the tool condition monitoring data acquired during machining has been greatly improved. Under the background of massive tool condition monitoring data, applying deep learning methods to tool residual life prediction can help overcome the dependence on specific prior knowledge and professional knowledge, and has more obvious advantages compared with traditional methods.

[0035] As shown in Figure 1 The present application provides a tool wear amount prediction method based on physical information self-supervised learning under few labels, which comprises the following steps:

[0036] Step S1: collecting cutting force, vibration and acoustic emission monitoring signals and the like during machining;

[0037] Step S2: taking the collected signals as data samples, marking part of the samples, and performing preprocessing operations such as truncation and Z-score standardization on each sample;

[0038] This step S2 specifically comprises:

[0039] In the experiment, each tool was subjected to 315 cutting cycles with a cutting length of 108 mm. The sensor data collected in each cutting cycle was regarded as a data sample. A total of 1890 data samples were generated in the experiment. In order to obtain the true wear state of the samples, a small number of tools were randomly selected for tool wear marking during the service life, and a data set composed of a small number of marked samples and a large number of unmarked samples was constructed. Each sample is composed of 7-channel sensor signals at a certain time step, including three-direction cutting force signals, three-direction vibration signals and acoustic emission signals. Specifically, after the tool completes a cutting cycle, the relief wear value of each cutting edge is measured as the state label of the sample. Accordingly, the data set is divided into marked samples (including C1, C4 and C6 tools) and unmarked samples (including C2, C3 and C5 tools);

[0040] Each sample in the data set is cut along the time step dimension. During the cutting of the sample along the time step dimension, first, 5% of the sample length of data points at the beginning and end of the sample are discarded to exclude the cutting-in and cutting-out stages. Then, considering that the marking work is carried out after the cutting cycle, the remaining 80000 sample points at the end of the data are retained as the initial sample to maximize the wear marking;

[0041] The sample data is subjected to z-score standardization, and the expression is:

[0042]

[0043] Wherein, x represents the value of a sampling point on a sensor in the sample;

[0044] μ represents the mean value of all sampling points of the x sensor data;

[0045] σ represents the standard deviation of all sampling points of the x sensor data.

[0046] Step S3: design a hybrid residual convolutional neural network tool wear monitoring model to perform information fusion and high-dimensional representation extraction on multi-source sensor signals (is to fuse and extract features of cutting force, vibration and acoustic emission through the designed network model);

[0047] The hybrid residual convolutional neural network tool wear monitoring model includes a hybrid residual convolutional neural network and a fully connected regression layer, and the tool wear monitoring value is output through the fully connected regression layer;

[0048] The hybrid residual convolutional neural network tool wear monitoring model combines shallow one-dimensional convolution, deep two-dimensional convolution and residual connection, aiming to convert the original data collected by multiple sensors into efficient representation signals. The model input is a multi-channel one-dimensional time series signal, including cutting force, vibration and acoustic emission signals. First, the input data passes through multiple one-dimensional convolution modules, which are composed of convolution layers, batch normalization layers and linear rectifier activation functions, independently extracting the shallow features of each sensor signal and increasing the number of channels. Subsequently, these shallow features are fused through a two-dimensional convolution module. The two-dimensional convolution module is composed of multiple residual basic blocks, each of which contains multiple convolution layers and their residual connections, extracting deep features of multi-source signals. After passing through the deep residual network, the output features are dimensionally reduced through the adaptive average pooling layer and converted to one-dimensional vectors through the flattening layer. Subsequently, these one-dimensional vectors are input into the fully connected regression layer. The fully connected layer is composed of a two-layer perceptron (MLP), the first layer maps the 512-dimensional input to 128-dimensional, and the second layer maps the 128-dimensional representation to a single tool wear value, with LeakyReLU and Softplus activation functions. Through the above process, the tool wear value is predicted. Taking single-channel signal input as an example, the overall architecture parameter settings of the hybrid residual convolutional neural network are shown in Table 1. The network architecture diagram of the hybrid residual convolutional neural network is shown in Figure 2 .

[0049] Table 1 Overall architecture parameter settings of hybrid residual convolutional neural network

[0050]

[0051]

[0052] Step S4: Based on the self-supervised learning framework, the tool wear monitoring model is pre-trained based on auxiliary task one, and through means such as enhanced view generation and twin network comparison loss calculation, the model is guided to extract the invariance features of the signal. The network structure diagram of auxiliary task one is shown in Figure 3 .

[0053] This step S4 specifically includes:

[0054] Auxiliary task one is a data-driven pre-assistant task based on a twin network, which is composed of two hybrid residual convolutional neural networks sharing weights, constructing two branches of the twin network;

[0055] The auxiliary task one generates two different views x1 and x2 by randomly combining different data enhancement methods including random cropping, reverse processing, amplitude inversion, noise adding and filtering processing operations, and inputs them into two branches of the twin network to obtain the latent tool wear representations z1 and z2 under different views; subsequently, a predictor is introduced in the upper branch of the twin network to generate a prediction vector p1, and a cosine similarity comparison is performed between the prediction vector p1 and the representation z2 of the lower branch to guide the mixed residual convolutional neural network to learn the invariant representation. The cosine similarity is used to measure the similarity between them, and is defined as follows:

[0056]

[0057] where ‖·‖2 represents the norm.

[0058] The gradient stopping operation is used in the lower branch to exchange the enhanced views to calculate the loss function in a symmetrical manner:

[0059]

[0060] where sg(·) represents the gradient stopping operation on the input variable.

[0061] Step S5: Based on the self-supervised learning framework, the tool wear monitoring model is pre-trained by using the auxiliary task two, the physical consistency of the representation vector is enhanced by introducing the physical constraint, and the network structure diagram of the auxiliary task two is as shown in Figure 4 .

[0062] The auxiliary task two is a regularization constraint auxiliary task based on physical properties,

[0063] The auxiliary task two proposes two distance measurement methods that fuse physical knowledge by combining the idea of triple loss, which together constitute the regularization constraint loss function. The triple loss is calculated by the degradation representations of three different samples; in this sample triple, the reference input is usually defined as the anchor sample x A , which corresponds to the cutting process monitoring data at t0 moment, and is denoted by symbol A; the similar input close to the reference input is defined as the positive sample x P , which corresponds to t1 moment, and is denoted by symbol P; and the input with large difference from the reference input is defined as the negative sample x N , which corresponds to t2 moment, and is denoted by symbol N. For a given triple sample {x A , x P , x N}, the embedded representations obtained by the backbone network are represented by {z A , z P , z N}, wherein, The first distance measurement method of the two distance measurement methods is constrained to the Euclidean distance in the embedding space, requiring that the anchor sample x A is closer to the positive sample x P in the Euclidean distance in the embedding space than to the negative sample x N . Thus, the constraint condition can be obtained as:

[0064]

[0065] where δ is the margin of the forced condition between the sample pairs, represents the square of the norm. For the i-th set of triple samples, the loss function of the Euclidean distance constraint can be obtained as:

[0066]

[0067] The second distance measurement method of the two distance measurement methods is constrained to the development trend of the representation, requiring that the vector is as consistent as possible in direction with the vector . The cosine similarity S c is used as an angle measurement index of the representation vector, and for the i-th set of triple samples, the loss function of the vector angle constraint can be obtained as:

[0068]

[0069] Thus, the loss function of the constraint representation Euclidean distance is and the loss function of the constraint representation vector angle is The two are combined by weighted summation to form the loss function of the task 2 representation measurement constraint as:

[0070]

[0071] where λ1 and λ2 are weighting coefficients.

[0072] Step S6: Based on the self-supervised learning framework, the auxiliary task 1 and the auxiliary task 2 are alternately optimized to gradually guide the model to have efficient feature extraction capability and physical law consistency.

[0073] This step S6 specifically includes:

[0074] ​In the pre-training stage, all collected samples are used to complete the self-supervised auxiliary task. The twin encoder part adopts a cosine decay schedule dynamic strategy to update the network parameters, and the learning rate gradually decreases to zero with the training round number, which takes into account the convergence speed in the early stage of the training process and the optimization stability in the later stage. The training of the predictor is carried out in a fixed learning rate manner. In the loss function of the representation metric constraint, the weight and are set to 0.001 and 1.

[0075] The specific settings of the pre-training parameters are shown in Table 2.

[0076] Table 2 Pre-training parameter settings

[0077]

[0078]

[0079] Step S7: Based on the self-supervised learning framework, the pre-trained tool wear monitoring model is fine-tuned using a limited scale of labeled samples, and the final wear value evaluation result is obtained; as shown in Figure 5 .

[0080] This step S7 specifically includes:

[0081] The shallow convolutional network parameters of the pre-trained model are frozen, and only the deep residual module is adjusted;

[0082] The optimization process of the deep residual module and the fully connected regression layer is as follows: first, the model error is obtained by calculating the loss function, then the gradient of the loss function with respect to the model parameters is calculated using backpropagation, and the parameters of the deep residual module and the fully connected regression layer are optimized through gradient descent method, finally the model training is completed with the goal of minimizing the mean square error (MSE). The optimization objective can be expressed as:

[0083]

[0084] In the formula, n is the number of samples in the labeled data set D L .

[0085] Finally, the model obtains the data representation through the fine-tuned hybrid residual convolutional neural network, and after average pooling and flattening layer, the predicted wear value is output through the optimized fully connected regression layer;

[0086] In the target task of tool wear value monitoring, a 5-fold cross-validation method was used for repeated experiments. The labeled samples were randomly divided into 5 equal parts. In each experiment, one of them was selected as the test set and the remaining 4 were used as the training set. In order to verify the accuracy of the tool wear value evaluation method under the condition of limited labels, only part of the data in the training set was used as the input sample to complete the training of the tool wear value monitoring model. In order to more comprehensively compare different degrees of label scarcity, three different degrees of limited label scenarios were set, which were 10%, 20%, 30%, 50% and 100% of the total number of training set samples. The tool wear monitoring model uses a multi-layer perception network as the backbone network followed by a fully connected layer. The Softplus activation function is used to ensure the non-negativity of the output, and the final wear value evaluation result is obtained. The optimizer parameters are the same as the pre-training. During the training process, the shallow convolutional parameters are frozen, the deep residual block parameters are fine-tuned, and the weights and biases of the fully connected layer neurons are updated and optimized. The specific parameter settings are shown in Table 3.

[0087] Table 3 Training parameter settings of target task

[0088]

[0089] Step S8: Evaluate the overall performance of the model in the tool wear state monitoring task by calculating the model evaluation indicators.

[0090] This step S8 specifically includes:

[0091] The model evaluation indicators are root mean square error, mean absolute error and mean absolute percentage error, and their calculation formulas are:

[0092]

[0093]

[0094] Where RMSE is the root mean square error, MAE is the mean absolute error, and MAPE is the mean absolute percentage error. y i 、 respectively represent the actual remaining life and the predicted remaining life after the i th cutting stroke; N represents the total number of predictions.

[0095] The smaller the values of RMSE, MAE and MAPE, the smaller the prediction error of the model, that is, the higher the accuracy of the model.

[0096] For example Figures 6-9As shown, the present application is first based on the experiment of PHM2010 data set using high-speed numerical control machine tool platform, using multiple sensors to collect multi-source signals, including three-dimensional cutting force, three-dimensional vibration and acoustic emission signals, and the sampling rate is 50 kHz. The experimental data covers 315 cutting cycles, in which the tool appears serious wear and fracture in the later period. The model training is divided into two stages: the first stage pre-trains the mixed residual convolutional neural network through the representation learning of unmarked data (including data-driven pre-task and regularization constraint task), and optimizes the signal feature extraction capability; the second stage fine-tunes the model using limited labeled data, and realizes high-precision prediction of tool wear value through mean square error loss. The experimental results show that the method performs well in wear monitoring and residual life prediction, and has significant industrial application value.

[0097] It can be understood that the present application is described by some embodiments, and those skilled in the art know that various changes or equivalent replacements can be made to these features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, these features and embodiments can be modified to adapt to specific conditions and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application are within the scope of the present application.

Claims

1. A tool wear amount prediction method based on physical information self-supervised learning with less labels, characterized in that: The method comprises the following steps: S1. Collecting cutting force, vibration and acoustic emission monitoring signals in the machining process; S2. Collecting signals as data samples, marking part of the samples, and performing preprocessing operations such as cutting and Z-score standardization on each sample; S3. Designing a hybrid residual convolutional neural network tool wear monitoring model to perform information fusion and high-dimensional feature extraction on multi-source sensor signals; S4. Based on the self-supervised learning framework, pre-training the tool wear monitoring model using auxiliary task one, guiding the model to extract the invariance features of the signals through enhanced view generation and twin network comparison loss calculation; S5. Based on the self-supervised learning framework, pre-training the tool wear monitoring model using auxiliary task two, enhancing the physical consistency of the feature vectors by introducing physical constraints; S6. Based on the self-supervised learning framework, gradually guiding the model to have both efficient feature extraction capability and physical law consistency by alternately optimizing auxiliary task one and auxiliary task two; S7. Based on the self-supervised learning framework, fine-tuning the pre-trained tool wear monitoring model using a limited number of labeled samples, and obtaining the final wear value evaluation result; S8. Evaluating the overall performance of the model in tool wear state monitoring task by calculating the model evaluation index.

2. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The S2 comprises the following steps: S21. Randomly selecting a small number of tools for tool wear marking during the tool life cycle, and constructing a data set composed of a small number of labeled samples and a large number of unlabeled samples; S22. Cutting each sample in the data set along the time step dimension, and in the sample cutting process, first discarding the beginning and end of the sample data points; then, retaining part of the sample points at the end of the remaining data as the initial sample; S23. The sample data is z-score standardized, and the expression is: Where x represents the value of a sampling point on a sensor in the sample; μ represents the mean of all sampling points of the x sensor data; σ represents the standard deviation of all sampling points of the x sensor data.

3. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The hybrid residual convolutional neural network tool wear monitoring model in S3 comprises a hybrid residual convolutional neural network and a fully connected regression layer, which outputs the tool wear monitoring value through the fully connected regression layer; The hybrid residual convolutional neural network is composed of one-dimensional convolution, two-dimensional convolution and residual connection, which is responsible for converting the original data into effective data representation.

4. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 3, characterized in that: The specific steps for information fusion and high-dimensional feature extraction of multi-source sensor signals in S3 are as follows: First, the input data is processed by multiple one-dimensional convolution modules to independently extract the shallow features of each sensor signal and increase the number of channels; then, these shallow features are fused through a two-dimensional convolution module to extract deep features of multi-source signals; after the deep residual network, the output features are reduced through an adaptive average pooling layer, and converted into a one-dimensional vector through a flattening layer; Then, these one-dimensional vectors are input into the fully connected regression layer.

5. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The first auxiliary task in S4 is a data-driven pre-assistant task based on a twin network, which is composed of two mixed residual convolutional neural networks sharing weights, and two branches of the twin network are constructed; The specific steps of S4 are: The auxiliary task generates two different views by randomly combining different data augmentation methods and and input them into two branches of the twin network to obtain the latent tool wear representation under different views and ; Subsequently, a predictor is introduced in the upper branch of the twin network to generate a prediction vector and cosine similarity comparison is performed with the representation of the lower branch to guide the learning of the invariant representation by the hybrid residual convolutional neural network. Cosine similarity is used to measure the similarity between them, which is defined as follows: In the formula denotes norm The loss function is calculated in a symmetrical manner by using the stop gradient operation in the lower branch and exchanging the enhanced views: In the formula represents that a gradient stop operation is performed on the input variable.

6. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The second auxiliary task in S5 is a regularization constraint auxiliary task based on physical properties, which proposes two distance measurement methods by combining the idea of triple loss and fusing physical knowledge, and constitutes a regularization constraint loss function; The triplet loss is computed by a degradation characterization of three different samples; in this sample triplet, the reference input is usually defined as the anchor sample , corresponding to the cutting process monitoring data at time ; a similar input close to the reference input is defined as a positive sample , corresponding to the time , denoted by An input that differs greatly from the reference input is defined as a negative sample , the corresponding is denoted by the symbol ; for a given triple sample , the embedded representation obtained by the backbone network is denoted by , where ; The first distance measurement method of the two distance measurement methods is constrained for the Euclidean distance in the embedding space, requiring that the anchor sample in the triple sample entering the backbone network at the same time is smaller than the negative sample in the Euclidean distance in the embedding space , so as to obtain the constraint condition: wherein, is the margin of the enforced condition between pairs of samples, denotes the square of the norm; for the i-th group of triplets of samples, the loss function for the Euclidean distance constraint is obtained as: The second distance measurement method of the two distance measurement methods is constrained for the development trend of the representation, requiring the vector to be as consistent as possible with the direction of the vector ; the cosine similarity is used as a representation vector angle measurement index of the trend requirement, and for the first group of triple samples, the vector angle constraint loss function is obtained as follows: Thus, the loss function is obtained, which constrains the Euclidean distance and the loss function, which constrains the vector angle Both are combined by weighted summation to form the loss function of task 2 representation constraint : wherein and are weighting factors.

7. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The fine-tuning of S7 using a limited number of labeled samples includes the following steps: S71. Freeze the shallow convolutional network parameters of the pre-trained model, and only adjust the deep residual module; S72. The optimization process of the deep residual module and the fully connected regression layer is to first calculate the model error by calculating the loss function, then calculate the gradient of the loss function with respect to the model parameters by back propagation, and finally optimize the parameters of the deep residual module and the fully connected regression layer by gradient descent method, and finally complete the model training by minimizing the mean square error. The optimization goal can be expressed as: In the formula, is the number of samples in the labeled dataset . , actual and predicted residual life after the nth cutting pass, respectively cutting pass S73. The mixed residual convolutional neural network of the model after fine-tuning obtains the data representation, and after average pooling and flattening, the predicted wear value is output through the optimized fully connected regression layer.

8. The tool wear amount prediction method based on physical information self-supervised learning with less mark according to claim 1, characterized in that: The model evaluation indicators in S8 are root mean square error, mean absolute error and mean absolute percentage error, and their calculation formulas are: wherein, is the root mean square error, is the mean absolute error, is the mean absolute percentage error, , denote the actual residual lifetime and the predicted residual lifetime, respectively, after the cutting pass; denotes the predicted total number.

Images