A numerical control machine tool feed shaft health state identification method, device and equipment

Abstract

The application discloses a numerical control machine tool feed shaft health state identification method and device, equipment and a readable storage medium, and relates to the technical field of big data. The method comprises the following steps: acquiring vibration data of a target numerical control machine tool feed shaft; the vibration data comprises vibration acceleration, vibration displacement and current signals; the vibration acceleration, the vibration displacement and the current signals are preprocessed, and the vibration acceleration, the vibration displacement and the current signals are fused and spliced in the channel dimension to obtain fused data; and the fused data is input into a feed shaft health state identification model to obtain a health state identification result of the target numerical control machine tool feed shaft. The method increases the accuracy and efficiency of health state identification.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically to a method, apparatus, device, and readable storage medium for identifying the health status of CNC machine tool feed axes. Background Technology

[0002] Current research on CNC machine tools focuses on fault diagnosis, specifically the study of wear and tear on critical components that occur after long-term operation. At this stage, the critical components have already suffered irreversible damage, and the research aims to determine the specific type and nature of the damaged components. However, damaged components cannot be used and must be replaced, which increases the operating costs of the CNC machine tool.

[0003] To reduce the operating costs of CNC machine tools, it is necessary to identify the condition of components (such as feed axes) before substantial damage occurs, i.e., before substantial damage occurs but before it does. Currently, actual production workshops monitor the health status of CNC machine tool feed axes using vibration and current signal acquisition devices, and perform manual analysis based on corresponding indicators. However, the amount of operational data is enormous, manual analysis is subject to human factors, and it can only determine whether the operation is normal, but cannot accurately locate the type of abnormal component, which is not conducive to subsequent rapid maintenance. Therefore, there is an urgent need for a method for identifying the health status of CNC machine tool feed axes that can overcome the above-mentioned shortcomings. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device, equipment, and readable storage medium for identifying the health status of CNC machine tool feed axes. By collecting vibration data of the target CNC machine tool feed axes, the invention enables the determination of the health status of the target CNC machine tool feed axes based on the vibration data using a feed axis health status identification model. Compared with the traditional method of manual data analysis, this invention not only increases the accuracy of the health status determination of the target CNC machine tool feed axes but also increases the efficiency of the health status determination.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a method for identifying the health status of a CNC machine tool feed axis, the method comprising:

[0007] Acquire vibration data of the feed axes of the target CNC machine tool; the vibration data includes vibration acceleration, vibration displacement, and current signals;

[0008] The vibration acceleration, vibration displacement, and current signals are preprocessed and then fused and stitched together in the channel dimension to obtain fused data.

[0009] The fusion data is input into the feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft.

[0010] In some embodiments, the vibration acceleration, vibration displacement and current signals are preprocessed, including:

[0011] The vibration acceleration, vibration displacement and current signals are sampled by using sliding interval sampling to obtain vibration acceleration, vibration displacement and current signals with consistent time series length;

[0012] The vibration acceleration, vibration displacement and current signals are decomposed by using the ensemble empirical mode decomposition method;

[0013] The noise in the vibration acceleration, vibration displacement and current signals is removed by using the wavelet denoising method with adaptive threshold.

[0014] In some embodiments, the method further includes:

[0015] The use spectrum entropy, high-frequency energy ratio, total energy and energy reduction rate of the preprocessed vibration acceleration, vibration displacement and current signals are calculated;

[0016] The denoising effect of the vibration acceleration, vibration displacement and current signals is judged according to the use spectrum entropy, high-frequency energy ratio, total energy and energy reduction rate;

[0017] If the denoising effect does not meet the denoising expectation, the noise in the vibration acceleration, vibration displacement and current signals is removed again by using the wavelet denoising method with adaptive threshold until the denoising effect meets the denoising expectation.

[0018] In some embodiments, the fusion data is input into the feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft, including:

[0019] The fusion data is input into the convolutional neural network and attention module of the feed shaft health state recognition model to extract spatial features and key features of the fusion data;

[0020] The spatial features and key features are multiplied element by element to obtain fusion features;

[0021] The fusion features are input into the bidirectional long short-term memory network to obtain time series features;

[0022] The time series features are input into the hybrid pooling layer to obtain reduced dimension features;

[0023] The reduced dimension features are input into the multi-label multi-class classification layer for classification recognition to obtain a health state recognition result of the target numerical control machine tool feed shaft.

[0024] In some embodiments, the multi-label multi-class classification layer is provided with a multi-label multi-classification form label, and a construction method of the multi-label multi-classification form label comprises the following steps:

[0025] Obtain sample vibration data of a feed shaft of a numerical control machine tool in different health conditions;

[0026] According to the sample vibration data and the health conditions of the sample numerical control machine tool feed shaft, a multi-label multi-classification form label is constructed.

[0027] In some embodiments, the method further comprises:

[0028] The model parameters of the feed shaft health state recognition model are optimized in real time by using an Optuna algorithm.

[0029] In a second aspect, the present application further provides a numerical control machine tool feed shaft health state recognition device, which comprises:

[0030] A data acquisition module is configured to acquire vibration data of a target numerical control machine tool feed shaft, wherein the vibration data comprises vibration acceleration, vibration displacement and current signals;

[0031] A data processing module is configured to preprocess the vibration acceleration, vibration displacement and current signals, and fuse and splice the vibration acceleration, vibration displacement and current signals in the channel dimension to obtain fused data.

[0032] A health recognition module is configured to input the fused data into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft.

[0033] In a third aspect, the present application further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the numerical control machine tool feed shaft health state recognition method provided in the first aspect when executing the computer program.

[0034] In a fourth aspect, the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the numerical control machine tool feed shaft health state recognition method provided in the first aspect.

[0035] In a fifth aspect, the present application further provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the numerical control machine tool feed shaft health state recognition method provided in the first aspect.

[0036] The beneficial effects of the present application are that the numerical control machine tool feed shaft health state recognition method provided in the present application first acquires vibration data of a target numerical control machine tool feed shaft; the vibration data includes vibration acceleration, vibration displacement and current signals; then the vibration acceleration, vibration displacement and current signals are preprocessed, and the vibration acceleration, vibration displacement and current signals are fused and spliced in the channel dimension to obtain fused data; finally, the fused data is input into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft. By collecting the vibration data of the target numerical control machine tool feed shaft, the health state of the target numerical control machine tool feed shaft is determined according to the vibration data of the target numerical control machine tool feed shaft by using the feed shaft health state recognition model, compared with the method of manually analyzing data in the traditional technology, not only the accuracy of the health state determination of the target numerical control machine tool feed shaft is increased, but also the health state determination efficiency of the target numerical control machine tool feed shaft is increased.

[0037] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, and to implement the content of the description, the following will be described in detail with the preferred embodiments of the present application and with the help of the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 A flowchart of a numerical control machine tool feed shaft health state recognition method according to an embodiment of the present application is shown.

[0039] Figure 2 A flowchart of the ensemble empirical mode decomposition method and the wavelet denoising method according to an embodiment of the present application is shown.

[0040] Figure 3 An optimization process diagram of the Optuna algorithm according to an embodiment of the present application is shown.

[0041] Figure 4 A structural diagram of a numerical control machine tool feed shaft health state recognition device according to an embodiment of the present application is shown.

[0042] Figure 5 An electronic device structure diagram according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0043] The technical solutions of the present application will be described in detail below with the help of the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0044] It is to be noted that the reference in the description to "one embodiment", "an embodiment", "example embodiment", etc., means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment. Further, when a particular feature, structure, or characteristic is said to have been added in an embodiment, it means that unless otherwise specified, such feature, structure, or characteristic can, but is not required to, be added in other embodiments.

[0045] In addition, the technical features involved in the different embodiments of the application described below can be combined with each other as long as there is no conflict.

[0046] In some embodiments, as shown in Figure 1 A flowchart of a health state identification method of a numerical control machine tool feed shaft is provided, and the specific method includes:

[0047] S101, obtaining vibration data of a target numerical control machine tool feed shaft.

[0048] The vibration data includes vibration acceleration, vibration displacement, and current signals.

[0049] Specifically, the target numerical control machine tool feed shaft includes X-axis, Y-axis, and Z-axis. A three-axis vibration acceleration sensor can be placed at a fixed position of the target numerical control machine tool, and the three-axis vibration acceleration sensor can collect vibration acceleration, vibration displacement, and current signals of the X-axis, Y-axis, and Z-axis.

[0050] S102, preprocessing the vibration acceleration, vibration displacement, and current signals, and fusing and splicing the vibration acceleration, vibration displacement, and current signals in the channel dimension to obtain fused data.

[0051] Specifically, the vibration acceleration, vibration displacement, and current signals can be first denoised and enhanced, and then fused and spliced in the channel dimension to obtain the fused data.

[0052] Optionally, preprocessing the vibration acceleration, vibration displacement, and current signals can include: sampling the vibration acceleration, vibration displacement, and current signals using sliding interval sampling to obtain vibration acceleration, vibration displacement, and current signals with consistent time series length; decomposing the vibration acceleration, vibration displacement, and current signals using an ensemble empirical mode decomposition method; and removing noise in the vibration acceleration, vibration displacement, and current signals using a wavelet denoising method with an adaptive threshold.

[0053] Specifically, in the actual collection process of vibration acceleration, vibration displacement and current signals, the same running action needs to be adopted for each collection of each numerical control machine tool, so that the meaning represented by the data collected each time is consistent. Since the reciprocating movement of the feed shaft represents the collection of state data once, the specific running action is that the workbench starts running from the bearing end of the feed shaft to the motor end. Since the fast-moving speed reaches 24 m / min, in order to ensure the stability of the running during the return stroke, the workbench needs to pause for 2 seconds after reaching the motor end to dissipate the residual vibration, and then perform signal testing in the return stroke section. Since the sampling frequency of vibration acceleration and vibration displacement is 3 kHz, the sampling frequency of the current signal is 4 kHz, and a complete cycle of the running action of the feed shaft needs about 5 seconds (including the 2-second pause), so that the vibration data collected by a single running reaches about 15000, and the current data reaches about 20000. The long time series is not conducive to the subsequent model training and recognition, and may cause gradient explosion problems; and the inconsistent time series length of different features also affects the model recognition. Therefore, the originally long and inconsistent sampling frequency data needs to be processed into data segments with appropriate and consistent time series length through sliding interval sampling method. The specific method is: 1. Take out the i(th initial i=0) window size data, then take out the window size data every k steps, and splice the taken data to form a group of short sequence segments Xi; 2. Add 1 to the value of i, and judge whether i≤k is true. If true, continue the operation in 1. If not, it means that the current long time series data has been segmented and sampled, and the original signal wave characteristics are still retained in the short sequence segments.

[0054] In addition, if the feed shaft of the numerical control machine tool runs abnormally, there will be pulses or mutations in the collected signals, which belong to non-stationary signals. The low-frequency components of these signals usually contain information corresponding to the abnormal state type, while noise signals generally exist in high-frequency signals and have small amplitude. Since the present application studies the early stage of abnormal state, the mechanical structure of the machine tool has not been substantially damaged, so the fluctuation of abnormal data is small. Therefore, the noise reduction of the original data needs to ensure that the noise is reduced while the information containing the abnormal type is retained as much as possible, so the ensemble empirical mode decomposition method (EEMD) is used to decompose the original signal, the correlation coefficient of each component in the decomposition is calculated, and the wavelet denoising method with adaptive threshold is used for noise reduction work for the components containing noise. Finally, the denoised signal with high abnormal state information retention degree is obtained through signal reconstruction, and the specific process can be referred to Figure 2 .

[0055] Optionally, in order to ensure that the noise reduction effect in the above method meets the expectation, it is also necessary to calculate the use spectral entropy, high frequency energy ratio, total energy and energy reduction rate of the pre-processed vibration acceleration, vibration displacement and current signals; judge the noise reduction effect of the vibration acceleration, vibration displacement and current signals according to the use spectral entropy, high frequency energy ratio, total energy and energy reduction rate; if the noise reduction effect does not meet the noise reduction expectation, then use the wavelet noise reduction method with adaptive threshold to remove the noise in the vibration acceleration, vibration displacement and current signals again until the noise reduction effect meets the noise reduction expectation.

[0056] Specifically, the frequency entropy is used to measure the complexity of the frequency spectrum distribution of the signal according to the concept of information entropy, and reflects the uniformity and randomness of the signal frequency spectrum. The greater the frequency entropy value is, the more uniform the signal frequency spectrum distribution is, the higher the complexity is, and the more noise components are contained; on the contrary, it is relatively pure, and the noise content is less. Assuming that the signal is divided into k windows with a length of M, the Fourier transform of each window is calculated according to formula (1), and the power spectrum calculation of formula (2) is performed on each window.

[0057] ;

[0058] ;

[0059] wherein, represents a Hanning window function, k represents the kth window, M represents the window length, represents the result of Fourier transform, represents the power spectrum, is the signal frequency, n represents the kth signal. n

[0060] According to formula (3), the average of all windows is calculated, and the normalized processing is performed to obtain the power spectrum.

[0061] ;

[0062] wherein, is the signal power spectrum density, is the normalized power spectrum.

[0063] In the normalized power spectrum obtained, the frequency entropy is defined as formula (4).

[0064] ;

[0065] wherein, is the frequency entropy.

[0066] ​The total energy of a signal measures the overall strength of the signal. In the noise reduction process, the total energy of the signal usually decreases because the noise is removed. The total energy is defined as the sum of the square of the signal, calculated as shown in equation (5):

[0067] ;

[0068] wherein, represents the time-domain signal data, is the total energy.

[0069] The high-frequency energy ratio represents the proportion of the energy of the high-frequency component of the signal to the total energy. The high frequency is often related to the noise, and the high-frequency energy ratio should decrease significantly after noise reduction. However, excessive reduction of the high-frequency energy will smooth the signal, resulting in loss of signal details. Assuming that the total energy of the signal is as shown in equation (5), the energy of the high-frequency part is calculated using equation (6):

[0070] ;

[0071] wherein, represents the signal power spectral density, represents a high-frequency threshold value set at 50% of the Nyquist frequency of the signal, is the energy of the high-frequency part.

[0072] The high-frequency energy ratio is calculated as shown in equation (7):

[0073] ;

[0074] wherein, is the high-frequency energy ratio.

[0075] The energy reduction rate represents the change in the signal energy after noise reduction relative to the original signal energy, and measures the loss of signal energy. Excessive energy reduction may mean that the useful signal is weakened too much, while insufficient energy reduction may indicate that the noise reduction effect is insufficient. The calculation process is shown in equation (8):

[0076] ;

[0077] wherein, is the energy reduction rate, is the original signal energy, is the lost signal energy.

[0078] If it is calculated that the spectral entropy, the high-frequency energy ratio, the total energy, and the energy reduction rate are within the preset range, it is proved that the noise reduction effect meets the noise reduction expectation, and no further processing is needed. If it is calculated that the spectral entropy, the high-frequency energy ratio, the total energy, and the energy reduction rate are not within the preset range, it is proved that the noise reduction effect does not meet the noise reduction expectation, and further noise reduction processing is needed until the noise reduction effect meets the noise reduction expectation.

[0079] S103, input the fusion data into the feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft.

[0080] Optionally, the convolutional neural network and the attention module of the feed shaft health state recognition model can be inputted with the fusion data to extract spatial features and key features of the fusion data; the spatial features and the key features are multiplied element by element to obtain fusion features; the fusion features are inputted into a bidirectional long short-term memory network to obtain time sequence features; the time sequence features are inputted into a mixed pooling layer to obtain reduced dimension features; the reduced dimension features are inputted into a multi-label multi-class classification layer for classification recognition to obtain the health state recognition result of the target numerical control machine tool feed shaft.

[0081] Specifically, the original input is divided according to two different scales, each convolutional neural network (CNN) module inputs a subsequence with a length of s (original scale), and the input length of the corresponding attention module (Attention module) is l a subsequence with a length of s (cross-scale), the cross-scale subsequence is centered on the center of the original scale subsequence, and l >s, so that the attention module can more comprehensively master the context information to accurately obtain the saliency features.

[0082] The CNN module is stacked by multiple one-dimensional networks, each layer contains a convolution layer, a batch normalization layer, a nonlinear layer, and simultaneously realizes sampling aggregation through an average pooling layer, and creates a hierarchical structure of gradually extracting more abstract features through the stacking of the convolution layer. The Attention module is composed of two parts: feature aggregation and scale restoration. The feature aggregation part uses the stacking of multiple convolution and pooling layers to extract key features from the cross-scale subsequence, and uses the multi-head attention mechanism to extract the attention degree of different features.

[0083] The scale restoration part restores the key features to be consistent with the output feature size of the CNN module, and uses the sigmoid function to constrain the numerical value between 0 and 1. The extracted context features are used as the saliency feature map of the basic features. The input of each parallel module has an adjustable overlap, and adjusting the overlap step of the input can make the model better adapt to data with different rules, and can further expand the length of the overall input of the joint model at one time to capture more accurate long-term features. The non-dense overlap between the parallel modules can avoid overfitting.

[0084] The output features of the CNN module are multiplied with the saliency features output by the corresponding attention mechanism module, and the higher the importance of the CNN output features, the closer the output of the corresponding attention mechanism module is to 1; conversely, the lower the importance of the CNN output features, the closer the output of the corresponding attention mechanism module is to 0. The importance of the features is reflected by the numerical value, thereby completing the identification of important features.

[0085] The result of the fusion is input into a node of a bidirectional long short-term memory network (BiLSTM unit) as input, and sequence modeling is performed by the BiLSTM to extract time sequence features.

[0086] The time sequence features extracted for each BiLSTM unit are respectively input into a global average pooling layer to calculate the feature mean of each time step, so as to retain the statistical information of the overall time sequence. At the same time, the output features at the end of the sequence are extracted, and the global time sequence statistical information and the local key features are combined to realize the fusion of the global trend and the local details, so as to ensure that the model can capture the overall time sequence mode and highlight the information features of the key time nodes. Finally, the fused features are input into a multi-label multi-class classification layer to complete the health state recognition.

[0087] Optionally, the multi-label multi-class classification layer is provided with a multi-label multi-classification form label, and a construction method of the multi-label multi-classification form label includes: obtaining sample vibration data of a sample CNC machine tool feed shaft under different health conditions; and constructing the multi-label multi-classification form label according to the sample vibration data and the health conditions of the sample CNC machine tool feed shaft.

[0088] Specifically, first, the types of all possible health abnormal states of the sample machine tool (such as screw loosening of the screw rod, influence of the shield, and non-optimization of the machine tool operation parameters) are counted, and the abnormal states are classified and counted; the sample machine tool is tested for vibration and current to obtain first sample vibration data, and the health condition of the machine tool is analyzed and judged according to the amplitude range and frequency characteristics of the vibration waveform; if an abnormality occurs, the machine tool is repaired and adjusted, the name of the part of the machine tool where the abnormality occurs is recorded, and the repair method is also recorded; after maintenance, the vibration and current are tested again to obtain second sample vibration data, and the second test data is analyzed according to the amplitude range and frequency characteristics of the vibration waveform; if the data is normal, it indicates that the maintenance operation can solve the problem of the first test data abnormality; if the data is still abnormal, the maintenance and test data steps are performed again until the test data is normal.

[0089] Because the VM1050 series machine tool contains more state abnormal data, and the label category of the data is relatively extensive, the VM1050 series is selected as the research object of the subsequent health state identification to build the data set. The VM1050 model machine tool has three feed axes of X-axis, Y-axis and Z-axis, each feed axis has complex transmission components, and each axis needs to be labeled separately, as shown in the following table 1:

[0090]

[0091] As can be seen from table 1, the label category names of the three axes are the same, because the running components used by the three feed axes are basically similar, but the sizes of the components, the directions of the shafts and the feed distances are different, and there are also differences in the running data, so the labels of the three feed axes need to be divided. Finally, a multi-label multi-classification form label is formed, and the current data set has 36 combined categories for the three feed axes, and some categories are shown in the following table 2.

[0092]

[0093] Optionally, in the use process of the feed axis health state identification model, the model also needs to be optimized in real time, for example: using Optuna algorithm to optimize the model parameters of the feed axis health state identification model in real time.

[0094] Specifically, referring to Figure 3 , Figure 3 is the optimization process of Optuna algorithm. Optuna is an efficient parameter optimization method that can automatically find the optimal parameter combination for deep learning models to improve model prediction ability, and has efficient sampling and pruning strategies. In the parameter optimization process, Optuna will regularly monitor the intermediate target value and terminate the trial that does not meet the predetermined conditions, and can prune the trial with poor effect to improve the efficiency and accuracy of parameter optimization. Optuna takes each optimization process as the research object, and takes each evaluation target function as the trial object to optimize parameters, and the main three core elements are: objective, trial and study. Among them, objective is responsible for defining the function to be optimized and specifying the parameter search space; trial makes one execution according to the objective function; study finds the optimal hyperparameters according to the results of multiple trials and records the parameter optimization results, which helps to understand the best hyperparameters and suggest the next parameter space to be searched.

[0095] The numerical control machine tool feed shaft health state recognition method in the above embodiment first acquires vibration data of a target numerical control machine tool feed shaft; the vibration data includes vibration acceleration, vibration displacement and current signals; then the vibration acceleration, vibration displacement and current signals are preprocessed, and the vibration acceleration, vibration displacement and current signals are fused and spliced in the channel dimension to obtain fused data; finally, the fused data is input into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft. By collecting the vibration data of the target numerical control machine tool feed shaft, the health state of the target numerical control machine tool feed shaft is determined according to the vibration data of the target numerical control machine tool feed shaft by using the feed shaft health state recognition model, compared with the method of manually analyzing data in the traditional technology, not only the accuracy of the health state determination of the target numerical control machine tool feed shaft is increased, but also the health state determination efficiency of the target numerical control machine tool feed shaft is increased.

[0096] In order to more comprehensively show the present scheme, the present embodiment gives an optional way of a numerical control machine tool feed shaft health state recognition method:

[0097] S201, acquiring vibration data of a target numerical control machine tool feed shaft.

[0098] The vibration data includes vibration acceleration, vibration displacement and current signals.

[0099] S202, sampling the vibration acceleration, vibration displacement and current signals by using sliding interval sampling to obtain vibration acceleration, vibration displacement and current signals with consistent time sequence length.

[0100] S203, decomposing the vibration acceleration, vibration displacement and current signals by using an ensemble empirical mode decomposition method.

[0101] S204, removing noise in the vibration acceleration, vibration displacement and current signals by using an adaptive threshold wavelet denoising method.

[0102] S205, calculating the use spectrum entropy, high-frequency energy ratio, total energy and energy reduction rate of the preprocessed vibration acceleration, vibration displacement and current signals.

[0103] S206, judging the denoising effect of the vibration acceleration, vibration displacement and current signals according to the use spectrum entropy, high-frequency energy ratio, total energy and energy reduction rate.

[0104] S207, if the denoising effect does not meet the denoising expectation, then the noise in the vibration acceleration, vibration displacement and current signals is removed again by using the adaptive threshold wavelet denoising method until the denoising effect meets the denoising expectation.

[0105] S208, fusing and splicing the vibration acceleration, vibration displacement and current signals in the channel dimension to obtain fused data.

[0106] S209, input the fusion data into a convolutional neural network and an attention module of the feed shaft health state recognition model, and extract spatial features and key features of the fusion data.

[0107] S210, element-wise multiplication is performed on the spatial features and the key features to obtain fusion features.

[0108] S211, input the fusion features into a bidirectional long short-term memory network to obtain time sequence features.

[0109] S212, input the time sequence features into a hybrid pooling layer to obtain reduced dimension features.

[0110] S213, input the reduced dimension features into a multi-label multi-class classification layer for classification recognition to obtain a health state recognition result of the target CNC machine tool feed shaft.

[0111] The multi-label multi-class classification layer is provided with a multi-label multi-classification form label, and a construction method of the multi-label multi-classification form label includes: obtaining sample vibration data of a sample CNC machine tool feed shaft under different health conditions; and constructing the multi-label multi-classification form label according to the sample vibration data and the health conditions of the sample CNC machine tool feed shaft.

[0112] S214, using an Optuna algorithm to optimize model parameters of the feed shaft health state recognition model in real time.

[0113] The specific process of S201-S214 can be referred to the description of the above method embodiments, which has similar implementation principles and technical effects, and will not be repeated here.

[0114] Based on the same inventive concept, the embodiments of the present application also provide a CNC machine tool feed shaft health state recognition device for implementing the CNC machine tool feed shaft health state recognition method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more CNC machine tool feed shaft health state recognition device embodiments provided below can be referred to the limitations of the CNC machine tool feed shaft health state recognition method described above, and will not be repeated here.

[0115] In one embodiment, as shown in Figure 4 a CNC machine tool feed shaft health state recognition device is provided, which comprises:

[0116] A data acquisition module 30 is configured to acquire vibration data of a target CNC machine tool feed shaft, wherein the vibration data includes vibration acceleration, vibration displacement and current signal.

[0117] The data processing module 31 is configured to preprocess the vibration acceleration, vibration displacement and current signals, and fuse and splice the vibration acceleration, vibration displacement and current signals in the channel dimension to obtain fused data.

[0118] The health recognition module 32 is configured to input the fused data into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft.

[0119] In another embodiment, the data processing module 31 in the above Figure 4 is specifically configured to: sample the vibration acceleration, vibration displacement and current signals by using sliding interval sampling to obtain vibration acceleration, vibration displacement and current signals with consistent time series length; decompose the vibration acceleration, vibration displacement and current signals by using an ensemble empirical mode decomposition method; remove noise in the vibration acceleration, vibration displacement and current signals by using an adaptive threshold wavelet denoising method; calculate the use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate of the preprocessed vibration acceleration, vibration displacement and current signals; determine the denoising effect of the vibration acceleration, vibration displacement and current signals according to the use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate; and if the denoising effect does not meet the denoising expectation, remove the noise in the vibration acceleration, vibration displacement and current signals again by using the adaptive threshold wavelet denoising method until the denoising effect meets the denoising expectation.

[0120] In another embodiment, the health recognition module 32 in the above Figure 4 is specifically configured to: input the fused data into a convolutional neural network and an attention module of the feed shaft health state recognition model to extract spatial features and key features of the fused data; multiply the spatial features and key features element by element to obtain fused features; input the fused features into a bidirectional long short-term memory network to obtain time series features; input the time series features into a hybrid pooling layer to obtain reduced dimension features; input the reduced dimension features into a multi-label multi-class classification layer for classification recognition to obtain the health state recognition result of the target numerical control machine tool feed shaft; and use an Optuna algorithm to optimize model parameters of the feed shaft health state recognition model in real time.

[0121] The multi-label multi-class classification layer is provided with a multi-label multi-classification form label, and a construction method of the multi-label multi-classification form label includes: obtaining sample vibration data of a sample numerical control machine tool feed shaft in different health states; and constructing the multi-label multi-classification form label according to the sample vibration data and the health states of the sample numerical control machine tool feed shaft.

[0122] The embodiments of the present application also provide an electronic device, in some embodiments, the electronic device is configured to implement the above Figure 5As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions executable on the processor 730, and the processor 730 invokes the program instructions to perform the method and / or technical solutions of the health state recognition of the feed shaft of the numerical control machine tool according to the foregoing embodiments. The electronic device 700 can be a mobile terminal device such as a mobile phone or a computer.

[0123] In addition, the embodiments of the present application also provide a computer readable storage medium for storing a computer program for performing the health state recognition method of the feed shaft of the numerical control machine tool. For example, computer program instructions, when executed by a computer, can invoke or provide the method and / or technical solutions according to the present application through the operation of the computer. The program instructions for invoking the method of the present application can be stored in a fixed or removable storage medium, and / or transmitted and / or stored in a storage medium according to the program instructions running.

[0124] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be realized by general computing devices, which can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Alternatively, they can be realized by program codes executable by computing devices, so that they can be stored in storage devices for execution by computing devices, or they can be made into individual integrated circuit modules, or a plurality of modules or steps among them can be made into a single integrated circuit module to realize. Therefore, the present application is not limited to any specific combination of hardware and software.

[0125] The technical features of the above embodiments can be integrated in any manner. In order to make the description simple, not all possible integrations of the technical features in the above embodiments are described, however, as long as the integration of the technical features does not exist contradictions, it should be considered as the scope of the present application.

[0126] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for recognizing the health state of a feed axis of a numerically controlled machine tool, characterized in that, The method comprises: obtaining vibration data of a target numerical control machine tool feed shaft; the vibration data comprises vibration acceleration, vibration displacement and current signals; sampling the vibration acceleration, vibration displacement and current signals using sliding interval sampling to obtain vibration acceleration, vibration displacement and current signals with consistent time sequence lengths; preprocessing the vibration acceleration, vibration displacement and current signals; the preprocessing of the vibration acceleration, vibration displacement and current signals comprises: decomposing the vibration acceleration, vibration displacement and current signals using an ensemble empirical mode decomposition method; and removing noise in the vibration acceleration, vibration displacement and current signals using a wavelet denoising method with an adaptive threshold value; calculating the use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate of the preprocessed vibration acceleration, vibration displacement and current signals; judging the denoising effect of the vibration acceleration, vibration displacement and current signals according to the use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate; if the denoising effect does not meet the denoising expectation, removing noise in the vibration acceleration, vibration displacement and current signals using the wavelet denoising method with the adaptive threshold value until the denoising effect meets the denoising expectation; fusing and splicing the vibration acceleration, vibration displacement and current signals in the channel dimension to obtain fused data; inputting the fused data into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft; inputting the fused data into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft, which comprises: inputting the fused data into a convolutional neural network and an attention module of the feed shaft health state recognition model to extract spatial features and key features of the fused data; performing element-by-element multiplication on the spatial features and the key features to obtain fused features; inputting the fused features into a bidirectional long short-term memory network to obtain time sequence features; inputting the time sequence features into a hybrid pooling layer to obtain reduced dimension features; and inputting the reduced dimension features into a multi-label multi-class classification layer for classification recognition to obtain the health state recognition result of the target numerical control machine tool feed shaft.

2. The method for identifying the health state of the feed axis of a CNC machine tool according to claim 1, characterized in that, The multi-label multi-class classification layer is provided with a multi-label multi-classification form label, and a construction method of the multi-label multi-classification form label comprises: obtaining sample vibration data of a sample numerical control machine tool feed shaft under different health conditions; constructing the multi-label multi-classification form label according to the sample vibration data and the health conditions of the sample numerical control machine tool feed shaft.

3. The method of claim 1, wherein, The method further comprises: optimizing model parameters of the feed shaft health state recognition model in real time using an Optuna algorithm.

4. A health state recognition device for a feed axis of a numerically controlled machine tool, characterized by The device comprises: a data acquisition module configured to obtain vibration data of a target numerical control machine tool feed shaft; the vibration data comprises vibration acceleration, vibration displacement and current signals; and the vibration acceleration, vibration displacement and current signals are sampled using sliding interval sampling to obtain vibration acceleration, vibration displacement and current signals with consistent time sequence lengths; The data processing module is configured to preprocess the vibration acceleration, vibration displacement and current signals. The preprocessing of the vibration acceleration, vibration displacement and current signals comprises: decomposing the vibration acceleration, vibration displacement and current signals by using an ensemble empirical mode decomposition method; and removing noise in the vibration acceleration, vibration displacement and current signals by using a wavelet denoising method with an adaptive threshold. The use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate of the preprocessed vibration acceleration, vibration displacement and current signals are calculated. The noise reduction effect of the vibration acceleration, vibration displacement and current signals is determined according to the use spectral entropy, high-frequency energy ratio, total energy and energy reduction rate. If the noise reduction effect does not meet the noise reduction expectation, noise in the vibration acceleration, vibration displacement and current signals is removed by using the wavelet denoising method with the adaptive threshold until the noise reduction effect meets the noise reduction expectation. The vibration acceleration, vibration displacement and current signals are fused and spliced in the channel dimension to obtain fused data. The health recognition module is configured to input the fused data into a feed shaft health state recognition model to obtain a health state recognition result of the target numerical control machine tool feed shaft. The health state recognition result of the target numerical control machine tool feed shaft is obtained by inputting the fused data into the feed shaft health state recognition model, which comprises: inputting the fused data into a convolutional neural network and an attention module of the feed shaft health state recognition model to extract spatial features and key features of the fused data; performing element-by-element multiplication on the spatial features and the key features to obtain fused features; inputting the fused features into a bidirectional long short-term memory network to obtain time sequence features; inputting the time sequence features into a hybrid pooling layer to obtain reduced dimension features; and inputting the reduced dimension features into a multi-label multi-class classification layer for classification recognition to obtain the health state recognition result of the target numerical control machine tool feed shaft. The numerical control machine tool feed shaft health state recognition device is configured to implement the numerical control machine tool feed shaft health state recognition method of any one of claims 1 to 3.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the numerical control machine tool feed shaft health state recognition method of any one of claims 1 to 3 when executing the computer program.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium stores the computer program, and the computer program implements the numerical control machine tool feed shaft health state recognition method of any one of claims 1 to 3 when executed by the processor.

7. A computer program product comprising a computer program, characterized in that, The computer program implements the numerical control machine tool feed shaft health state recognition method of any one of claims 1 to 3 when executed by the processor.

Images