Precision cutting machine tool fault prediction method

By constructing a serial mechanism of convolutional network and self-attention mechanism, multi-dimensional operating parameters of precision cutting machine tools are extracted, which solves the problem of low accuracy in fault judgment of precision cutting machine tools and realizes accurate prediction and control of early faults.

CN123115666APending Publication Date: 2026-11-17ADVANCED TECH ACHIEVEMENTS WESTERN (MIANYANG) TRANSFORMATION CENT (MIANYANG SCI & TECH CITY ADVANCED TECH RES INST)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611298012.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the existing technology, the accuracy of fault diagnosis of precision cutting machine tools is low, it is easy to misjudge, and it is difficult to predict in the early stage, which makes it difficult to control the trend of fault deterioration.

Method used

By constructing a sequential mechanism of convolutional network, long short-term memory network and self-attention mechanism, multi-dimensional operating parameters of machine tool are obtained, local impact and long-term degradation features are extracted layer by layer, and the weight of fault segments is amplified by self-attention mechanism for feature fusion, thereby improving fault judgment accuracy and early detection capability.

Benefits of technology

It improves the accuracy of fault diagnosis, reduces the possibility of fault deterioration, and enables accurate identification and prediction of early faults.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115666A_ABST
    Figure CN123115666A_ABST
Patent Text Reader

Abstract

This invention provides a method for predicting faults in precision cutting machine tools, comprising the following steps: acquiring the operating parameters of the precision cutting machine tool as training samples; preprocessing the operating parameters; constructing a fault prediction model, the fault prediction model including a first feature extraction module, a second feature extraction module, a self-attention module, a feature fusion module, and a prediction output module; inputting the preprocessed operating parameters into the fault prediction model for training until training is complete; acquiring the real-time operating parameters of the precision cutting machine tool to be estimated, preprocessing them, and then inputting them into the trained fault prediction model to obtain the fault prediction result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a fault prediction method, and more particularly to a fault prediction method for precision cutting machine tools. Background Technology

[0002] Precision cutting machining machines are a crucial manufacturing technology in industries such as aerospace, shipbuilding, and automobiles, and are widely used in the machining of complex and high-precision components.

[0003] In precision machining, besides the setting of process parameters affecting the machining accuracy of components, machine tool fault diagnosis is a key factor related to product processing efficiency and quality. If faults are not predicted and diagnosed early, machine tool failures can easily lead to serious machining accidents. Current technology typically diagnoses machine tool faults by examining apparent characteristic values ​​such as spindle vibration and current, i.e., by checking if these characteristic values ​​exceed limits. However, this method results in low accuracy and is prone to misjudgment. More importantly, even if a real fault occurs after these characteristic values ​​exceed limits, it indicates a relatively severe fault, making early intervention difficult and failing to mitigate the tendency for the fault to worsen.

[0004] Therefore, in order to solve the above-mentioned technical problems, it is urgent to propose a new technical approach. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method for predicting faults in precision cutting machine tools. By acquiring the operating parameters of the machine tool under test in different dimensions and establishing a prediction model with a serial mechanism of convolutional network, long short-term memory network and self-attention mechanism, the method can progressively advance from local impact features to long-term degradation features. Moreover, the self-attention mechanism amplifies the weight of fault segments and then performs feature fusion, thereby enabling the extraction of early weak fault features and fault judgment. This is beneficial to improving the accuracy of fault judgment and early fault detection, reducing the trend of fault deterioration and the possibility of serious fault accidents.

[0006] The present invention provides a method for predicting faults in precision cutting machine tools, comprising the following steps:

[0007] S1. Obtain the operating parameters of the precision cutting machine tool as training samples;

[0008] S2. Preprocess the operating parameters;

[0009] S3. Construct a fault prediction model, which includes a first feature extraction module, a second feature extraction module, a self-attention module, a feature fusion module, and a prediction output module;

[0010] The first feature extraction module takes preprocessed operating parameters as input, the output features of the first feature extraction module are input to the second feature extraction module, the output features of the second feature extraction module are input to the self-attention module, the output features of the self-attention module are input to the feature fusion module, the output features of the feature fusion module are input to the prediction output module, and the prediction output module outputs the fault prediction result.

[0011] S4. Input the preprocessed operating parameters into the fault prediction model for training until training is complete;

[0012] S5. Obtain the real-time operating parameters of the precision cutting machine tool to be estimated, perform preprocessing, and input them into the trained fault prediction model to obtain the fault prediction results.

[0013] Furthermore, the first feature extraction module includes three sequentially connected convolutional units with identical structures;

[0014] The convolutional unit includes a sequentially connected convolutional layer, an activation function layer, and a max pooling layer;

[0015] The convolutional kernel lengths in the three convolutional units are 32, 16, and 8 respectively in the input-to-output direction; and the stride of each convolutional kernel is 2.

[0016] The pooling windows of the max pooling layers are all 2×1 with a step size of 2.

[0017] Furthermore, the second extraction module includes two cascaded LSTM networks, the number of hidden units in the LSTM networks is 128, and a Dropout layer is set between the two LSTM networks with Dropout=0.2.

[0018] Furthermore, the feature fusion module includes convolution module I, convolution module II, convolution module III, convolution module IV, a concatenation module, convolution module V, and a ReLU activation function module;

[0019] The input features of convolution module I are the output features of the self-attention module. The output features of convolution module I are input into convolution modules II and III. The output features of convolution module III are input into convolution module IV. The output features of convolution module IV and the output features of convolution module II are input into the splicing module for splicing operation. The output features of the splicing module are input into convolution module V. The output features of convolution module V and the output features of convolution module I are added element-wise and then input into the ReLU activation function module. The output features of the ReLU activation function module are used as the output features of the feature fusion module.

[0020] Among them, the convolution scale of convolution module I, convolution module III and convolution module V is 1×1, and the convolution scale of convolution module II and convolution module IV is 3×3.

[0021] Furthermore, the prediction output module includes a fully connected layer I, a fully connected layer II, and a softmax module;

[0022] The output features of fully connected layer I are input into fully connected layer II, and the output features of fully connected layer II are input into the softmax module. The softmax module outputs the prediction results. The output dimensions of fully connected layer I and fully connected layer II are 64 and 32, respectively.

[0023] Furthermore, the operating parameters include dynamic parameters and static parameters, wherein: the dynamic parameters include spindle vibration data, triaxial feed vibration, spindle current, spindle temperature, and motor current;

[0024] Static parameters include spindle speed, feed rate, and machining ratio.

[0025] Furthermore, the preprocessing of the running parameters used as training samples includes:

[0026] Timestamp alignment for dynamic parameters;

[0027] EMD (Empirical Mode Decomposition) is used to separate the effective signal from the dynamic parameters;

[0028] The separated effective signal is subjected to noise filtering.

[0029] Normalize the dynamic and static parameters;

[0030] The dynamic and static parameters are divided into sliding serial port segments according to a window size of 256.

[0031] Furthermore, the precision cutting machine tool to be estimated undergoes preprocessing, including outlier removal and noise filtering, followed by normalization.

[0032] The beneficial effects of this invention are as follows: By acquiring the operating parameters of the machine tool under test in different dimensions and establishing a prediction model with a serial mechanism of convolutional network, long short-term memory network and self-attention mechanism, the invention can progressively advance from local impact features to long-term degradation features. Moreover, the self-attention mechanism amplifies the weight of fault segments and then performs feature fusion, thereby enabling the extraction of early weak fault features and fault judgment. This is beneficial to improving the accuracy of fault judgment and early fault detection, reducing the trend of fault deterioration and the possibility of serious fault accidents. Attached Figure Description

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0034] Figure 1 This is a schematic diagram of the process of the present invention.

[0035] Figure 2 A schematic diagram of the fault prediction model structure of the present invention.

[0036] Figure 3 This is a schematic diagram of the feature fusion module structure of the present invention. Detailed Implementation

[0037] The present invention will be further described in detail below:

[0038] The present invention provides a method for predicting faults in precision cutting machine tools, comprising the following steps:

[0039] S1. Obtain the operating parameters of the precision cutting machine tool as training samples; the operating parameters include dynamic parameters and static parameters, of which: dynamic parameters include spindle vibration data, three-axis feed vibration, spindle current, spindle temperature, and motor current;

[0040] Static parameters include spindle speed, feed rate, and machining ratio. Using multi-dimensional parameters can improve the accuracy of the identification results.

[0041] S2. Preprocess the running parameters; specifically: the preprocessing of the running parameters used as training samples includes:

[0042] Timestamp alignment for dynamic parameters;

[0043] EMD (Empirical Mode Decomposition) is used to separate the effective signal from the dynamic parameters;

[0044] The separated effective signal is subjected to noise filtering.

[0045] Normalize the dynamic and static parameters;

[0046] The dynamic and static parameters are divided into sliding serial port segments according to a window size of 256.

[0047] S3. Construct a fault prediction model, which includes a first feature extraction module, a second feature extraction module, a self-attention module, a feature fusion module, and a prediction output module;

[0048] The first feature extraction module takes preprocessed operating parameters as input, the output features of the first feature extraction module are input to the second feature extraction module, the output features of the second feature extraction module are input to the self-attention module, the output features of the self-attention module are input to the feature fusion module, the output features of the feature fusion module are input to the prediction output module, and the prediction output module outputs the fault prediction result.

[0049] S4. Input the preprocessed operating parameters into the fault prediction model for training until training is complete;

[0050] S5. Obtain the real-time operating parameters of the precision cutting machine tool to be estimated, perform preprocessing, and then input them into the trained fault prediction model to obtain the fault prediction result. The preprocessing of the precision cutting machine tool to be estimated includes outlier removal and noise filtering, followed by normalization.

[0051] The noise filtering employs existing filtering methods, and the self-attention module utilizes existing technology. Through a self-attention mechanism, it overcomes the problems of gradient vanishing and long-range dependency decay in long short-term memory networks over extremely long sequences, while also possessing fully parallel computing capabilities, significantly improving training efficiency. This invention acquires different dimensions of the machine tool under test's operating parameters and establishes a prediction model using a sequential mechanism of convolutional networks, long short-term memory networks, and a self-attention mechanism. This model progressively advances from local impact features to long-term degradation features, and the self-attention mechanism amplifies the weights of fault segments before feature fusion. This enables the extraction and fault diagnosis of early, weak fault features, improving fault diagnosis accuracy and early fault detection, reducing the likelihood of fault deterioration and serious accidents.

[0052] In this embodiment, the first feature extraction module includes three sequentially connected convolutional units with the same structure;

[0053] The convolutional unit includes a sequentially connected convolutional layer, an activation function layer, and a max pooling layer;

[0054] The convolutional kernel lengths in the three convolutional units are 32, 16, and 8 respectively in the input-to-output direction; and the stride of each convolutional kernel is 2.

[0055] The pooling windows of the max-pooling layers are all 2×1 with a stride of 2. This effectively extracts impact features, providing support for the subsequent second extraction module.

[0056] In this embodiment, the second extraction module includes two cascaded LSTM networks, and the number of hidden units in the LSTM networks is 128. A Dropout layer is set between the two LSTM networks with Dropout=0.2.

[0057] The feature fusion module includes convolution module I, convolution module II, convolution module III, convolution module IV, a concatenation module, convolution module V, and a ReLU activation function module;

[0058] The input features of convolution module I are the output features of the self-attention module. The output features of convolution module I are input into convolution modules II and III. The output features of convolution module III are input into convolution module IV. The output features of convolution module IV and the output features of convolution module II are input into the splicing module for splicing operation. The output features of the splicing module are input into convolution module V. The output features of convolution module V and the output features of convolution module I are added element-wise and then input into the ReLU activation function module. The output features of the ReLU activation function module are used as the output features of the feature fusion module.

[0059] Specifically, convolutional modules I, III, and V have a convolutional scale of 1×1, while convolutional modules II and IV have a convolutional scale of 3×3. First, the output features of the self-attention module are adjusted in channel dimension using the 1×1 convolutional module I. Then, this feature is divided into two paths: one path extracts local short-range temporal features through the 3×3 convolutional module II, and the other path extracts deep mid-range temporal features through the 1×1 convolutional module III and the 3×3 convolutional module IV. The two outputs are then concatenated along the channel dimension by the concatenation module, and then fused and dimensionality reduced across channels by the 1×1 convolutional module V. Finally, the output of module V is element-wise added to the output of convolutional module I, and the result is output after passing through the ReLU activation function. This structure captures processing state features across different time spans through multi-scale convolutions, ensures effective gradient backpropagation and prevents deep network degradation through residual connections, and achieves efficient inter-channel information recombination through 1×1 convolutions. Overall, it significantly improves feature representation capabilities while keeping the number of parameters controllable, thus ensuring the accuracy of the final prediction results.

[0060] In this embodiment, the prediction output module includes a fully connected layer I, a fully connected layer II, and a softmax module;

[0061] The output features of fully connected layer I are input into fully connected layer II, and the output features of fully connected layer II are input into the softmax module. The softmax module outputs the prediction results. The output dimensions of fully connected layer I and fully connected layer II are 64 and 32, respectively.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting faults in precision cutting machine tools, characterized in that: Includes the following steps: S1. Obtain the operating parameters of the precision cutting machine tool as training samples; S2. Preprocess the operating parameters; S3. Construct a fault prediction model, which includes a first feature extraction module, a second feature extraction module, a self-attention module, a feature fusion module, and a prediction output module; The first feature extraction module takes preprocessed operating parameters as input, the output features of the first feature extraction module are input to the second feature extraction module, the output features of the second feature extraction module are input to the self-attention module, the output features of the self-attention module are input to the feature fusion module, the output features of the feature fusion module are input to the prediction output module, and the prediction output module outputs the fault prediction result. S4. Input the preprocessed operating parameters into the fault prediction model for training until training is complete; S5. Obtain the real-time operating parameters of the precision cutting machine tool to be estimated, perform preprocessing, and input them into the trained fault prediction model to obtain the fault prediction results.

2. The method for predicting faults in precision cutting machine tools according to claim 1, characterized in that: The first feature extraction module includes three sequentially connected convolutional units with the same structure; The convolutional unit includes a sequentially connected convolutional layer, an activation function layer, and a max pooling layer; The convolutional kernel lengths in the three convolutional units are 32, 16, and 8 respectively in the input-to-output direction; and the stride of each convolutional kernel is 2. The pooling windows of the max pooling layers are all 2×1 with a step size of 2.

3. The method for predicting faults in precision cutting machine tools according to claim 1, characterized in that: The second extraction module includes two cascaded LSTM networks, with 128 hidden units in each LSTM network, and a Dropout layer with Dropout=0.2 is set between the two LSTM networks.

4. The method for predicting faults in precision cutting machine tools according to claim 1, characterized in that: The feature fusion module includes convolution module I, convolution module II, convolution module III, convolution module IV, a concatenation module, convolution module V, and a ReLU activation function module; The input features of convolution module I are the output features of the self-attention module. The output features of convolution module I are input into convolution modules II and III. The output features of convolution module III are input into convolution module IV. The output features of convolution module IV and the output features of convolution module II are input into the splicing module for splicing operation. The output features of the splicing module are input into convolution module V. The output features of convolution module V and the output features of convolution module I are added element-wise and then input into the ReLU activation function module. The output features of the ReLU activation function module are used as the output features of the feature fusion module. Among them, the convolution scale of convolution module I, convolution module III and convolution module V is 1×1, and the convolution scale of convolution module II and convolution module IV is 3×3.

5. The method for predicting faults in precision cutting machine tools according to claim 1, characterized in that: The prediction output module includes a fully connected layer I, a fully connected layer II, and a softmax module; The output features of fully connected layer I are input into fully connected layer II, and the output features of fully connected layer II are input into the softmax module. The softmax module outputs the prediction results. The output dimensions of fully connected layer I and fully connected layer II are 64 and 32, respectively.

6. The method for predicting faults in precision cutting machine tools according to claim 1, characterized in that: The operating parameters include dynamic parameters and static parameters, wherein: the dynamic parameters include spindle vibration data, three-axis feed vibration, spindle current, spindle temperature, and motor current; Static parameters include spindle speed, feed rate, and machining ratio.

7. The method for predicting faults in precision cutting machine tools according to claim 6, characterized in that: Preprocessing of the runtime parameters used as training samples includes: Timestamp alignment for dynamic parameters; EMD (Empirical Mode Decomposition) is used to separate the effective signal from the dynamic parameters; The separated effective signal is subjected to noise filtering. Normalize the dynamic and static parameters; The dynamic and static parameters are divided into sliding serial port segments according to a window size of 256.

8. The method for predicting faults in precision cutting machine tools according to claim 6, characterized in that: The preprocessing of the precision cutting machine tool to be estimated includes outlier removal and noise filtering, followed by normalization.