Device and method for predicting service life of machine tool cutter

Abstract

The invention relates to a machine tool cutter service life prediction device and a service life prediction method, belongs to the field of numerical control machine tool cutter service life prediction, and solves the problems of inconsistent machine tool protocols, non-uniform data, inaccurate cutter service life prediction and the like in the prior art. The machine tool cutter service life prediction device comprises an edge unit and a monitoring control unit; the edge unit comprises an AI microprocessor, a 4G DTU, an ESP WIFI and an A / D input port. And the monitoring control unit is used for processing the machine tool data uploaded by the edge unit. According to the device and method for predicting the service life of the tool in real time, on the premise that normal machining of the machine tool is not affected, operation data of the machine tool during machining are obtained, a deep learning method is used, a service life prediction model is established and deployed in the device, the service life of the tool is predicted in real time on the edge side, and the utilization rate and accuracy of the tool are improved.

Description

Technical Field

[0001] The present invention belongs to the field of tool life prediction for numerical control machine tools, and relates to a device and a method for predicting the tool life of machine tools, and specifically to a device and a method for predicting the tool life of an LSTM network based on an attention mechanism on the edge side. Background Art

[0002] Failures often occur during CNC machine tool processing, among which accidents caused by tool damage account for 80% of the total factory accidents, tool failure causes more than 20% of machine tool downtime, and maintenance costs are as high as 15% to 40% of production costs. Therefore, it is particularly important to monitor the tool status in real time and predict its remaining life. Different equipment manufacturers have different degrees of openness of equipment data and communication protocols, which also makes interoperability between various types of equipment a challenge.

[0003] At present, most companies rely on experience or regular replacement strategies when replacing cutting tools, which leads to insufficient use of cutting tools, increased costs, and inability to promptly handle problems that arise during the use of cutting tools. In response to the existing problems, some researchers have also conducted relevant research.

[0004] Xidian University proposed a tool life prediction method in its application publication number CN113560955A, "A method, system and application for predicting the remaining service life of CNC machine tool tools". This method collects controller signals and sensor signals during the operation of CNC machine tools, and uses long short-term memory networks and attention mechanisms to establish a tool remaining service life prediction model to achieve the prediction of the remaining service life of CNC machine tool tools, thereby improving the generalization ability of the tool remaining service life prediction model. However, it does not solve the problems of inconsistent machine tool protocols and inconsistent data, and its actual application scope is relatively limited.

[0005] iFlytek South China Artificial Intelligence Research Institute (Guangzhou) Co., Ltd. applied for "A tool life prediction method and related device" with publication number CN114378639A. The method extracts multiple image signals and sensor signals of the tool to be tested, combines the image and sensor feature data, and uses a preset attention mechanism to achieve accurate prediction of the remaining tool life. By comprehensively utilizing the characteristics of image and sensor signals, the invention can significantly improve the prediction accuracy, thereby improving processing efficiency and quality, and reducing production costs. However, its implementation process is complex, requires high computing resources, and is highly dependent on data quality. There may be problems with equipment adaptation, resulting in limited versatility. It may also bring higher costs and potential real-time problems.

[0006] Shenzhen Institute of Advanced Technology, in its application publication number CN115600512A, "A tool life prediction method based on distributed learning", achieves accurate prediction of tool life by training the model locally on each terminal with tool feature data and sending the training loss to the central terminal for model parameter update. This method can improve the accuracy of the prediction model while protecting data privacy by using data from multiple manufacturers. However, this method also has some shortcomings, including complex implementation process, high computing resources required, and possible problems with data synchronization and communication delays. In addition, distributed learning has high system configuration and maintenance requirements, which may lead to increased deployment and operating costs.

[0007] In view of the problems of complex installation, high cost and poor adaptability of existing tool monitoring and life prediction solutions, the present invention proposes an improved method, that is, based on the LSTM architecture based on the attention mechanism, the attention mechanism is uniquely optimized. An adaptive attention mechanism is introduced in the input layer, so that it can dynamically adjust the weight and optimize the information flow processing. At the same time, the modeling ability of long-term and short-term dependencies is improved through the improved LSTM unit. In addition, intelligent data fusion technology is introduced in the data preprocessing stage to automatically correct and standardize data from different sources at the edge layer to ensure the consistency and reliability of the data. The present invention aims to provide a real-time tool life prediction device and method that is easy to install, cost-effective and widely applicable, so as to reduce the pressure on the bandwidth of the local area network and improve the utilization of tools. Summary of the invention

[0008] In response to problems such as inconsistent machine tool protocols, inconsistent data, and inaccurate tool life prediction, the present invention proposes a device and a method for predicting machine tool tool life. The device and method are applicable to real-time prediction of tool life for multiple types of machine tools. Without affecting the normal processing of the machine tool, the operating data during processing is obtained, and a life prediction model is established using a deep learning method. The model is deployed in the device to predict the tool life in real time on the edge side, thereby improving its utilization and accuracy.

[0009] The present invention provides a device for predicting the life of a machine tool tool, comprising an edge unit and a monitoring control unit;

[0010] The edge unit includes an AI microprocessor, 4G DTU, ESP WIFI, and A / D input port;

[0011] Monitoring control unit, used to process machine tool data uploaded by edge unit.

[0012] Optionally, the AI ​​microprocessor is used to collect and process machine tool data in real time, perform processing tasks, extract various data features to determine the tool status, and run a life prediction model at the same time, inputting feature data to obtain the remaining life of the tool.

[0013] Optionally, the 4G DTU is used to communicate with the monitoring control unit, transmit the collected data to the cloud server for life prediction model training, and deploy the trained life prediction model to the edge unit.

[0014] Optionally, the ESP WIFI is used to communicate with a monitoring control unit to display machine tool data, status and prediction information in real time.

[0015] Optionally, the A / D input port is used to collect tool data during processing.

[0016] In another aspect, the present invention further provides a method for predicting the life of a machine tool tool, using the aforementioned device for predicting the life of a machine tool tool, and the specific steps are as follows:

[0017] Step 1: Collect machine tool data; divide the machine tool data into a training set and a test set after processing;

[0018] Step 2: Build a life prediction model and use the training set to train the life prediction model;

[0019] Step 3: After training the life prediction model, use the test set to set different starting points, and use the starting point and previous machine tool data for prediction;

[0020] Step 4: Predict whether the value at the next time point is greater than the maximum life threshold of the tool. If it is not greater than the maximum life threshold, add the predicted data to the end of the test set and return to step 3. If it is greater than the maximum life threshold, proceed to step 5.

[0021] Step 5: Perform multi-step prediction until the predicted maximum spindle load is greater than the threshold value, and the prediction ends to obtain the prediction result.

[0022] Optionally, in step 1, the machine tool data is collected; and the specific steps of dividing the machine tool data into a training set and a test set after processing are as follows:

[0023] Step 11, dividing the machine tool data according to the processing cycle; calculating multiple eigenvalues ​​of the machine tool data of each processing cycle, and selecting the eigenvalue positively correlated with tool wear from the multiple eigenvalues;

[0024] Step 12, reducing the dimension of the eigenvalues ​​positively correlated with tool wear obtained in step 11, and selecting the most important eigenvalues ​​related to tool wear using the minimum redundancy maximum correlation method;

[0025] Step 13: Define the training data set and test set based on the five most important eigenvalues ​​related to tool wear.

[0026] Optionally, the plurality of characteristic values ​​include mean, variance, peak factor, margin factor, skewness, kurtosis, waveform factor, maximum value, standard deviation and root mean square amplitude.

[0027] Optionally, in step 2, a life prediction model is built, and the specific steps of training the life prediction model using the training set are as follows:

[0028] Step 21, using a sequence segmentation function to segment the spindle load signals at continuous time points in the training data set into an input sequence of input time point samples and output time point samples;

[0029] Step 22: normalize the input sequence of the input time point samples and the output time point samples to the interval [0, 1];

[0030] Step 23, adjusting the input sequence of the input time point samples and the output time point samples normalized to the interval [0,1] to the input format of the shape required by the long short-term memory network LSTM layer;

[0031] Step 24: Based on the input format obtained in step 23, a four-layer long short-term memory network LSTM based on the attention mechanism is constructed, and the optimal hyperparameters are found using the grid search method to obtain a trained life prediction model.

[0032] Optionally, the four-layer attention-based long short-term memory network LSTM includes an attention layer, an LSTM layer, a Dropout layer and a fully connected layer.

[0033] Compared with the prior art, the present invention has at least the following beneficial effects:

[0034] 1. The device and method for predicting the tool life of machine tools disclosed in the present invention directly communicate with the machine tool to obtain internal data and collect spindle current data, define a node model common to the machine tool to collect data, pre-process the machine tool data, and eliminate invalid data, thereby realizing the isomorphism of heterogeneous equipment, alleviating bandwidth pressure, and improving real-time performance.

[0035] 2. The device and method for predicting machine tool tool life disclosed in the present invention use a dual-core architecture chip on the edge side to separate data collection and computing tasks, complete the prediction of tool life and display and configure relevant parameters of machine tool data, deploy the model to the edge in the cloud, complete the remote update of equipment parameters and firmware, and realize the decoupling of the controller and the machine tool and the coordination of cloud and edge tasks.

[0036] 3. The device and method for predicting the life of machine tool tools disclosed in the present invention, by establishing an LSTM model based on the attention mechanism, assigns higher weights to time points with greater influence in the input time series data, thereby increasing the accuracy of tool remaining life prediction to more than 94%, and increasing the tool service life to 1.5 times the original. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is an application system diagram of an implementation case of the present invention;

[0038] 2(a) and (b) are schematic diagrams of a device for predicting the life of a machine tool tool according to the present invention;

[0039] Figure 3 This is a flow chart of LSTM model training prediction based on attention mechanism of the present invention;

[0040] Figure 4 A flow chart of a method for predicting the life of a machine tool tool of the present invention;

[0041] Figure 5 This is a comparison chart of tool remaining life prediction curves for implementation cases of the present invention. DETAILED DESCRIPTION

[0042] In order to facilitate understanding of the purpose, advantages and application methods of the present invention, the present invention is described in detail below with reference to examples. It should be pointed out that the described embodiments are only intended to facilitate understanding of the present invention and are not intended to limit the present invention.

[0043] One embodiment of the present invention, as Figure 1-5 As shown, a device for predicting the tool life of a machine tool is disclosed, including an edge unit and a monitoring control unit.

[0044] The tool life prediction device is arranged on the machine tool. The machine tool generates a large amount of machine tool data when processing, and communicates with the tool life prediction device through Ethernet or serial communication.

[0045] Furthermore, the edge unit uses a universal OPC UA node to collect spindle load data, which can be applied to machine tools with different protocols, communicate with Fanuc machine tools through the FOCAS protocol, or communicate with Brother machine tools through messages, etc., and then pre-process the machine tool data transmitted by the machine tool, place the data uniformly on the data acquisition node, and then send it to the monitoring and control unit. It is responsible for receiving the latest algorithm model sent by the monitoring and control unit, and calling it inside the device to predict the tool life in real time;

[0046] Among them, the edge unit includes AI microprocessor, 4G DTU, ESP WIFI and A / D input port;

[0047] Further, a monitoring control unit is used to process the machine tool data uploaded by the edge unit;

[0048] Among them, the monitoring and control unit includes cloud server monitoring and real-time data display on the edge side.

[0049] The invention discloses a device for predicting the life of a machine tool tool, which directly communicates with the machine tool through an Ethernet port or a serial port to obtain the internal data of the machine tool.

[0050] The AI ​​microprocessor uses a dual-core architecture chip to collect and process machine tool data in real time, perform high-computation processing tasks, extract various data features to determine the tool status, and run a deep learning algorithm model to input feature data to obtain the remaining life of the tool;

[0051] The machine tool data collected by real-time processing is mainly the spindle processing load obtained by directly communicating with the machine tool using an Ethernet port or a serial port according to a communication protocol;

[0052] The various data features retain the main information related to tool wear in the signal, and use Kalman filtering to eliminate irrelevant information, thereby reducing the complexity of subsequent learning calculations and improving prediction accuracy;

[0053] The 4G DTU is used to communicate with the monitoring control unit, transmit the collected data to the cloud server for model training, and deploy the trained model to the edge unit to achieve collaborative work between the cloud and the edge to ensure the latest and prediction accuracy of the model;

[0054] The ESP WIFI is used to communicate with the monitoring control unit to display the machine tool data, status and prediction information in real time on the machine tool side;

[0055] The A / D input port is used to collect current data of the spindle motor during processing;

[0056] Furthermore, the monitoring control unit trains the life prediction model through the LSTM model based on the attention mechanism and converts it, and deploys the trained life prediction model to the edge unit. The specific training steps are as follows;

[0057] Step 1, collect machine tool data; divide the machine tool data according to the processing cycle; calculate the mean, variance, peak factor, margin factor, skewness, kurtosis, waveform factor, maximum value, standard deviation and root amplitude of the machine tool data of each processing cycle, and select the characteristic value that is positively correlated with tool wear from the ten characteristic values;

[0058] Step 2, reduce the dimension of the eigenvalues ​​positively correlated with tool wear obtained in step 1, and select the five most important eigenvalues ​​related to tool wear using the minimum redundancy maximum correlation method;

[0059] Specifically, the five most important characteristic values ​​related to tool wear are standard deviation, peak factor, root square amplitude, waveform factor, and maximum value;

[0060] Furthermore, the expression of the minimum redundancy maximum correlation is:

[0061] mRMR=max s [D(S,y)-R(S)]

[0062] Among them, S is the currently selected feature set; max s It means selecting the feature that can maximize the objective function in the candidate feature subset S; y is the label quantity corresponding to the currently selected feature set, D(S,y) is the maximum correlation between the feature value and the label quantity, and R(S) is the redundancy between the feature values ​​in the feature set.

[0063] Furthermore, the expression of the maximum correlation between the feature value and the label quantity and the redundancy between the feature values ​​in the feature set is:

[0064]

[0065] Where |S| is the size of the feature set; s i ,s j are the i-th and j-th features in the feature set S; I(X,Y) is the mutual information value of X and Y. i ,y) and I(s i ,s j ), s i Corresponding to X, y and s j Corresponding to Y, the expression of the mutual information value of X and Y is:

[0066]

[0067] In the formula, P(x i ) is the i-th value x of the random variable X i The probability of P(y i ) is the jth value y of the random variable Y i The probability of P(x i ,y i ) is a random variable X and Y that takes the value x at the same time i and i The joint probability of .

[0068] Step 3: Define the training data set and test set based on the five most important eigenvalues ​​related to tool wear;

[0069] Step 4: Use a sequence segmentation function to segment the spindle load signal of continuous time points in the training data set into an input sequence of input time points (X) and output time point samples (Y);

[0070] It can be understood that each sample in the input time point (X) contains the values ​​of 5 consecutive time points, and the output time point sample (Y) is the true value of the next time point as a label;

[0071] Among them, the value at a time point corresponds to the five most important characteristic values ​​of tool wear calculated from the machine tool processing data of one cycle.

[0072] Step 5: Perform Min-Max normalization on the input sequence of input time points (X) and output time point samples (Y) to the interval [0,1];

[0073] The expression of the Min-Max normalization process is:

[0074]

[0075] Among them, x * is the data after standardization, x min is the minimum value of the data before standardization, x max is the maximum value of the data before normalization.

[0076] Step 6: Adjust the input sequence of input time points (X) and output time point samples (Y) normalized to the interval [0,1] by Min-Max normalization to the input format of the shape required by the long short-term memory network LSTM layer (i.e., batch size, time step, number of input features);

[0077] Step 7: Based on the input format obtained in step 6, construct a four-layer long short-term memory network LSTM based on the attention mechanism, use the grid search method to find the optimal hyperparameters, and obtain a trained life prediction model.

[0078] Specifically, the first layer is the attention layer, and the input shape is (i.e., batch size, time step ， The feature vector of each time step is weighted, and two transpose layers are used at the beginning and end to give higher weights to the time points with greater influence, while ensuring that the shape of the weighted feature output is the same as the input (i.e., batch size, time step size). ， Input number of features);

[0079] The second layer is the LSTM layer, which receives input data from the attention layer in the shape of (batch size, time step, number of input features), performs time series modeling on the input sequence, uses 50 neurons and ReLU activation function, calculates the hidden state vector, and outputs the hidden state of the last time step in the shape of batch size 50;

[0080] The third layer is the Dropout layer. The input is the hidden state of the last time step of the LSTM layer with a batch size of 50. A certain proportion (such as 50%) of neurons are randomly discarded to prevent overfitting. The output shape is still the hidden state of the batch size 50.

[0081] The fourth layer is a fully connected layer. The input comes from the hidden state of the Dropout layer with a batch size of 50. Each input is mapped to a new output space, set to 50 neurons, and uses an activation function to output a feature representation with a batch size of 50. Used for subsequent loss calculation tasks.

[0082] Furthermore, the steps of the grid search method include the following steps:

[0083] S1. Define parameter grid: set the hyperparameters to be optimized and their value ranges. For model training, the hyperparameters θ1, θ2, ..., θ M , list all possible combinations (similar to a "grid"), see Table 1;

[0084] Table 1 Grid search method example

[0085]

[0086] S2, traverse all combinations: for each hyperparameter combination θ in the grid (m) Perform model training and calculate performance indicators (such as accuracy, F1 score, etc.) under corresponding hyperparameter combinations.

[0087] S3. Use cross-validation to evaluate the model performance of each hyperparameter combination and obtain the performance index P(θ (m) ).

[0088] S4. Find the hyperparameter combination θ with the best performance index best :

[0089]

[0090] Among them, Θ represents all possible parameter combinations; P represents the model performance index; θ (m) represents the parameter combination θ of the mth grid search, where m is the index of the combination; θ best represents the set of parameters that can maximize the model performance index P; P(θ(m) ) is the model in the parameter combination θ (m) , i.e., accuracy, F1 score, etc.; arg max means finding the performance index that can make P(θ (m) )The parameter that reaches the maximum value; θ∈θ means that the parameter θ belongs to the parameter set Θ.

[0091] S5. Use the Adam optimizer and mean square error (MSE) as the loss function to train the model. The number of iterations is set to 1000. The early stopping method is used to obtain the optimal model. The number of early stopping steps is set to 50 to prevent overfitting.

[0092] The expression of mean absolute error (MAE) is:

[0093]

[0094] Among them, y i is the i-th true label, is the predicted value of the ith model, and n is the number of samples.

[0095] Update the parameters using the Adam optimizer:

[0096]

[0097] Among them, m t and v t They are the estimates of the first-order moment and the second-order moment of the gradient at time t, respectively; ∈ is a smoothing term to prevent division by zero; θ represents the parameter to be optimized in the model; α represents the learning rate, which is used to control the amplitude of parameter update.

[0098] S6. In order to verify the effectiveness of the model, the model is converted into a C function and deployed to the edge unit for operation. A new tool is used for processing in the CNC machine tool. The standard deviation, peak factor, square root amplitude, waveform factor and maximum value of the spindle load during each processing are calculated in the edge unit, and then input into the model for multi-step prediction to obtain the current predicted load value. It is determined whether it is greater than the replacement threshold. The two are subtracted to obtain the predicted remaining life of the tool. The predicted remaining life of the tool is compared with the actual remaining life (that is, the output time point sample (Y) obtained in step 4). If it is close to the actual remaining life, the model prediction accuracy is high and can be used to guide tool replacement and reduce waste or overuse. If it is not close to the actual remaining life, the data set is adjusted, feature extraction is optimized, or the model is retrained to improve adaptability and accuracy.

[0099] For example, the predicted life curve of the tool at different times of use is as follows: Figure 5 As shown in the figure, the accuracy of tool remaining life prediction after 1000 times can reach more than 94% on average.

[0100] Another embodiment of the present invention discloses a method for predicting the tool life of a machine tool. The aforementioned prediction device is installed on the machine tool side, and a node model common to the machine tool is defined to collect the machining operation data of the machine tool in real time to achieve the isomorphism of heterogeneous devices. The feature values are extracted, the dimension is reduced, and the data is standardized for the collected data. Then, an LSTM model based on the attention mechanism is constructed and trained using the standardized data. Finally, the trained model is deployed to an embedded device for running calculations to achieve real-time prediction of the tool life within the edge device.

[0101] A method for predicting the tool life of a machine tool disclosed by the present invention adopts indirect prediction to analyze data indicators that can represent tool wear from the collected data, establishes a model to predict the change of this data indicator until this indicator reaches a threshold that can characterize tool failure, and then calculates the time from the prediction time point to the threshold of the indicator as the remaining life of the tool. The specific steps are as follows:

[0102] Step 1: Divide the data into a training set and a test set after processing.

[0103] Step 2: Build a life prediction model and train it using the training set.

[0104] Further, based on the aforementioned monitoring and control unit, train the life prediction model through an LSTM model based on the attention mechanism and perform conversion, and train the network model by deploying the trained life prediction model to the edge unit.

[0105] Step 3: After training the life prediction model, set different starting points using the test set and perform prediction using the starting point and the data before it.

[0106] Step 4: Predict whether the value at the next moment is greater than the maximum life threshold of the tool. If it is not greater than the threshold, add the predicted data to the end of the test set.

[0107] Specifically, let P(t + 1) be the predicted tool life value at the next moment t + 1, and L be the maximum life threshold of the tool. If P(t + 1) < L, then:

[0108] Test set ← Test set ∪ P(t + 1).

[0109] Step 5: Perform multi-step prediction, and repeat this process until the maximum predicted spindle load is greater than the threshold, at which point the prediction ends and the prediction result is obtained.

[0110] Further, obtain the tool replacement decision according to the predicted load trend of the prediction result; understand the tool wear degree and machining performance through load characteristic analysis; compare the prediction result with the actual situation to provide a basis for optimizing the model.

[0111] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A device for predicting the life of a machine tool tool, characterized in that: including an edge unit and a monitoring control unit; The edge unit includes an AI microprocessor, 4G DTU, ESP WIFI, and A / D input port; Monitoring control unit, used to process machine tool data uploaded by edge unit.

2. The device for predicting the life of a machine tool tool according to claim 1, characterized in that: The AI ​​microprocessor is used to collect and process machine tool data in real time, execute processing tasks, extract various data features to determine the tool status, and run the life prediction model at the same time, input the feature data to obtain the remaining life of the tool.

3. The device for predicting the life of a machine tool according to claim 1, characterized in that: The 4GDTU is used to communicate with the monitoring control unit, transmit the collected data to the cloud server for life prediction model training, and deploy the trained life prediction model to the edge unit.

4. The device for predicting the life of a machine tool tool according to claim 1, characterized in that: The ESP WIFI is used to communicate with the monitoring control unit to display machine tool data, status and prediction information in real time.

5. The device for predicting the life of a machine tool tool according to claim 1, characterized in that: The A / D input port is used to collect machine tool data during processing.

6. A method for predicting the life of a machine tool tool, using the device for predicting the life of a machine tool tool according to any one of claims 1 to 5, characterized in that: The specific steps are as follows: Step 1: Collect machine tool data; divide the machine tool data into a training set and a test set after processing; Step 2: Build a life prediction model and use the training set to train the life prediction model; Step 3: After the life prediction model is trained, multiple prediction starting points are set using the test set, and prediction is performed using the prediction starting point and previous machine tool data; Step 4: Predict whether the value at the next time point is greater than the maximum life threshold of the tool. If it is not greater than the maximum life threshold, add the predicted data to the end of the test set and return to step 3. If it is greater than the maximum life threshold, proceed to step 5. Step 5: Perform multi-step prediction until the predicted maximum spindle load is greater than the threshold value, and the prediction ends to obtain the prediction result.

7. The method for predicting the life of a machine tool tool according to claim 6, characterized in that: In step 1, the machine tool data is collected; the specific steps of dividing the machine tool data into a training set and a test set after processing are as follows: Step 11, dividing the machine tool data according to the processing cycle; calculating multiple eigenvalues ​​of the machine tool data of each processing cycle, and selecting the eigenvalue positively correlated with tool wear from the multiple eigenvalues; Step 12, reducing the dimension of the eigenvalues ​​positively correlated with tool wear obtained in step 11, and selecting the most important eigenvalues ​​related to tool wear using the minimum redundancy maximum correlation method; Step 13: Define the training data set and test set based on the most important feature values ​​related to tool wear.

8. The method for predicting the life of a machine tool tool according to claim 7, characterized in that: Multiple characteristic values ​​include mean, variance, crest factor, margin factor, skewness, kurtosis, crest factor, maximum value, standard deviation, and RMS magnitude.

9. The method for predicting the life of a machine tool tool according to claim 7, characterized in that: In step 2, a life prediction model is built. The specific steps for training the life prediction model using the training set are as follows: Step 21, using a sequence segmentation function to segment the spindle load signals at continuous time points in the training data set into an input sequence of input time point samples and output time point samples; Step 22: normalize the input sequence of the input time point samples and the output time point samples to the interval [0, 1]; Step 23, adjusting the input sequence of the input time point samples and the output time point samples normalized to the interval [0,1] to the input format of the shape required by the long short-term memory network LSTM layer; Step 24: Based on the input format obtained in step 23, a four-layer long short-term memory network LSTM based on the attention mechanism is constructed, and the optimal hyperparameters are found using the grid search method to obtain a trained life prediction model.

10. The method for predicting the life of a machine tool tool according to claim 9, characterized in that: The four-layer long short-term memory network LSTM based on the attention mechanism includes an attention layer, an LSTM layer, a Dropout layer, and a fully connected layer.

Images