Machine tool predictive maintenance method based on multi-measuring-point vibration sensing monitoring

Abstract

The invention discloses a machine tool predictive maintenance method based on multi-measuring-point vibration sensing monitoring, and the method comprises the following steps: S1, determining a key measuring point, installing a vibration sensor, collecting a vibration signal, and carrying out the preprocessing of the vibration signal; s2, vibration signals are classified, continuous signals are subjected to spectral analysis, and discontinuous signals are extracted and spliced through sliding window extremum search; s3, performing frequency domain analysis and fast Fourier transform on the continuous signal sequence to extract frequency characteristics, and comparing the frequency spectrum characteristics of the measuring points and the time periods; s4, calculating the characteristic value of the vibration signal, fusing the characteristics by adopting an improved spatial pyramid pooling, and constructing a radar map; and S5, performing predictive maintenance based on the frequency domain features and the radar map feature values, and triggering early warning and executing maintenance inspection when abnormity occurs. The machine tool health monitoring precision is improved, precise predictive maintenance is achieved, sudden failures are reduced, the production efficiency is improved, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment health monitoring, and particularly to a predictive maintenance method for machine tools based on multi-point vibration sensing monitoring. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, higher requirements are put forward for the intelligent management and maintenance of machine tools. Any failure or downtime of a machine tool may lead to production interruption, affect delivery time, and even damage the enterprise's reputation. Therefore, the traditional regular maintenance and after-fact maintenance modes can no longer meet the needs of modern manufacturing. Against this background, predictive maintenance (PdM), as an emerging maintenance strategy, came into being. It realizes proactive maintenance, reduces unexpected downtime, and improves production efficiency by monitoring the operating state of the machine tool in real time, analyzing its performance data, and predicting potential failures.

[0003] The core of predictive maintenance technology lies in using advanced sensor technology, data analysis, and machine learning algorithms to monitor and analyze the health status of the machine tool in real time. By collecting a large amount of data generated during the operation of the machine tool, including vibration, temperature, sound, current, and voltage, etc., the predictive maintenance system can identify abnormal patterns and trends, predict possible failures, and take maintenance measures before the failures occur. This condition-based maintenance strategy can not only reduce maintenance costs, but also improve production efficiency and product quality, which is of great significance for reducing the risk of unexpected downtime and extending the service life of the machine tool.

[0004] Existing predictive maintenance technologies mainly rely on data collected by various sensors, combined with signal processing and data analysis methods, to monitor and evaluate the machine tool status. In recent years, many researchers have conducted in-depth studies in this field and proposed various methods, including physics-based methods, statistical analysis methods, and intelligent diagnosis methods based on machine learning and deep learning.

[0005] Physics-based methods usually rely on the physical characteristics of the machine tool to establish a mathematical model to simulate the operating state of the machine tool, and judge whether the machine tool is abnormal by monitoring the changes of key parameters. However, this method has high requirements for the structure and working environment of the machine tool, the modeling process is complex, and it is difficult to adapt to different types of machine tools.

[0006] Statistical analysis methods use historical data and statistical characteristics to establish a failure probability model, such as time series analysis, principal component analysis (PCA), and hidden Markov model (HMM), etc. These methods can effectively analyze the change trend of the machine tool health status, but there are still certain limitations in their analysis ability when facing high-dimensional and multi-modal data.

[0007] In recent years, with the development of artificial intelligence technology, predictive maintenance methods based on machine learning and deep learning have gradually become a research hotspot. These methods can build intelligent models through training data, automatically extract key features, and achieve fault diagnosis and prediction. For example, traditional machine learning algorithms such as Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT) are widely used in fault classification and anomaly detection. At the same time, deep learning methods such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory Network (LSTM) can process high-dimensional, time-series related data and improve the accuracy of prediction.

[0008] Although certain progress has been made in the existing technologies, there are still some defects and limitations. First, traditional predictive maintenance methods often rely on single sensor data or a limited feature set, which restricts the ability to comprehensively evaluate the state of the machine tool. Second, the existing methods still need to be improved in terms of real-time performance and accuracy, especially under complex working conditions. In addition, for the prediction of faults and the estimation of remaining useful life, the existing technologies often lack sufficient precision and reliability. These problems limit the application effect of predictive maintenance technology in actual production, and there is an urgent need for new technologies to solve these problems.

[0009] Therefore, how to achieve real-time monitoring of the full range of states of the machine tool and improve the accuracy and reliability of fault prediction through comprehensive analysis of multi-point measurement data is the key to improving the predictive maintenance technology of the machine tool. Summary of the Invention

[0010] An object of the present invention is to propose a predictive maintenance method for machine tools based on multi-point vibration sensing monitoring. By introducing a radar chart comparison method and combining multi-point vibration signal monitoring technology, the present invention can not only provide a more comprehensive view of the machine tool state, but also compare the vibration signal characteristics of different measurement points and time periods in the form of an intuitive radar chart, so as to more accurately identify and predict potential faults. This method not only improves the accuracy of fault diagnosis, but also enables maintenance personnel to more quickly understand and respond to the maintenance requirements of the machine tool through intuitive radar chart comparison, thus realizing more efficient and economical equipment management.

[0011] The predictive maintenance method for machine tools based on multi-point vibration sensing monitoring according to the embodiments of the present invention includes the following steps:

[0012] S1. Determine the key measurement point positions of the machine tool, install vibration sensors at the key parts of the rotating table, spindle, and X, Y, and Z axes of the machine tool, collect the vibration signals during operation, and perform preprocessing;

[0013] S2. Classify the preprocessed vibration signals into continuous rotational motion signals and non - continuous rotational motion signals. Conduct spectral analysis and feature extraction on the continuous rotational motion signals, and use an improved sliding window extreme value search technique to extract the effective motion process for the non - continuous rotational motion signals, and splice them to form a continuous signal sequence;

[0014] S3. Conduct frequency - domain analysis on the continuous signal sequence. Use the fast Fourier transform to convert the time - domain signal into a frequency - domain signal, analyze the frequency components and amplitudes, compare the spectral characteristics of different measuring points and different time periods, and extract the frequency - domain features related to the operating state of the machine tool;

[0015] S4. Calculate the eigenvalues of the vibration signals based on the frequency - domain features, use improved spatial pyramid pooling to achieve feature fusion and construct a radar chart. Each axis of the radar chart corresponds to different eigenvalues, and each point of the radar chart represents the signal characteristics of the measuring point and the time period. Based on the radar chart, compare and analyze the vibration characteristics of different measuring points and time periods;

[0016] S5. Based on the frequency - domain features and the vibration characteristics of the radar chart, conduct periodic predictive maintenance on the machine tool. When an anomaly occurs, trigger the warning mechanism, perform maintenance inspections, and update the maintenance strategy.

[0017] Optionally, the preprocessing includes noise removal, anomaly detection, and signal normalization processing.

[0018] Optionally, the specific content of S2 includes:

[0019] S21. Determine the continuity of the normalized vibration signals, decompose the continuous vibration components and non - continuous vibration components, and set up a signal decomposition model based on variational mode decomposition:

[0020]

[0021] where min is the minimum operator, is the time derivative, j is the imaginary unit, x(t) is the original vibration signal, u k is the k - th modal component, ω k is the central frequency of the modal component, K is the total number of modal components, λ is the balance parameter, u k (t) is the k - th modal component at time t, ||·|| 2 is the square of the Euclidean norm;

[0022] S22. Extract the effective motion process for the non - continuous rotational motion signals, use an improved sliding window extreme value search algorithm with an adaptive window, and set up a local extreme value determination function within the window:

[0023]

[0024] where P(x k ) is the local extreme value probability of the k-th data point, x k is the k-th data point in the signal sequence, N w is the sliding window size, and H(·) is the Heaviside step function;

[0025] S23. Concatenate the extracted valid signals, and use a signal alignment method based on dynamic time warping to define the calculation formula for the optimal matching path:

[0026]

[0027] where D(i, j) is the cumulative matching cost of two signals at point (i, j), and d(i, j) is the Euclidean distance between two signals at point (i, j).

[0028] Optionally, step S3 specifically includes:

[0029] S31. Perform time-frequency conversion on the continuous signal sequence, and use the fast Fourier transform to convert the time-domain signal into a frequency-domain signal:

[0030]

[0031] where X(f) is the frequency-domain signal, x(n) is the normalized vibration signal, N is the total number of sampling points, n is the serial number of the sampling point, f is the signal frequency, and j is the imaginary unit;

[0032] S32. Extract the main frequency of the frequency-domain signal and compare it with the valid motion signal extracted by local extreme value search:

[0033]

[0034] where f m is the main frequency of the signal, |X(f)| is the amplitude of the frequency-domain signal, and arg max is the value of f that maximizes |X(f)| to identify the main frequency component of the signal;

[0035] S33. Compare the spectral characteristics of different measurement points and different time periods, calculate the spectral similarity between measurement points using normalized cross-correlation, and extract the frequency-domain characteristics of the machine tool operating state:

[0036]

[0037] where R xy (τ) is the normalized cross-correlation value between signals X p (n) and X(f), X p (n) is the concatenated continuous signal, and are respectively signals Xp The mean values of (n) and X(f), and σ X (f) are the standard deviations of the signal X p (n) and X(f) respectively, and τ is the time shift amount.

[0038] Optionally, the S4 specifically includes:

[0039] S41. Calculate the eigenvalue set of the vibration signal based on the frequency-domain signal to provide a basis for fault diagnosis:

[0040]

[0041] Among them, X EE is the eigenvalue set of the vibration signal, including the effective value, peak-to-peak value, kurtosis, and impulse factor. These eigenvalues can quantitatively describe the characteristics of the signal. P(f) is the normalized energy distribution of the frequency-domain signal at frequency f, X(f) is the frequency-domain signal, F is the f value that makes |X(f)| maximum, and log2 is the logarithmic function with base 2;

[0042] S42. Use the single feature of the vibration signal as the normalization factor to calculate the multi-scale entropy of the vibration signal to make the multi-scale entropy coordinated with the spectral energy characteristics:

[0043]

[0044] Among them, X MSE is the multi-scale entropy of the vibration signal, S is the number of scales, X s (i) is the time series component of the spliced signal X p (i) at the s-th scale, N s is the number of sampling points at this scale, P s (i) is the normalized probability distribution;

[0045] S43. Combine the time-domain characteristics of the multi-scale entropy and the frequency-domain characteristics of the eigenvalues to make the information weights of different-level characteristics more balanced, and use the improved spatial pyramid pooling to fuse the vibration signal characteristics:

[0046]

[0047] Among them, X SPP is the eigenvalue fused through the spatial pyramid pooling, L is the number of pyramid layers, M l is the number of regional divisions at the l-th layer, X MSE,j is the value of the multi-scale entropy in the j-th region, X EE,j is the value of the eigenvalue in the j-th region, W l is the weight factor of this layer, and α is the adaptive adjustment parameter;

[0048] S44. Construct a radar chart based on the fused eigenvalues, and set the calculation formulas for high-dimensional mapping and adaptive normalization radar chart:

[0049]

[0050] Among them, X radar (i) is the i-th feature point on the radar chart, D is the feature dimension, and X SPP,d (i) is the d-th dimensional eigenvalue after spatial pyramid pooling fusion, and X MSE,d (i) is the eigenvalue of multi-scale entropy in the d-th dimension, and W d is the weighting factor of the feature dimension, and tanh(·) is the non-linear mapping function;

[0051] S45. Each axis of the radar chart corresponds to different eigenvalues. Each point on the radar chart represents the signal characteristics of the measurement point and time period. Combining the eigenvalues and multi-scale entropy for feature enhancement can improve the accuracy of the radar chart in machine tool fault prediction.

[0052] The beneficial effects of the present invention are as follows:

[0053] First of all, the present invention proposes a multi-measurement point vibration signal monitoring technology. By installing high-precision vibration sensors at key parts of the machine tool, real-time monitoring of the overall state of the machine tool is realized. This multi-dimensional data acquisition method can provide more comprehensive machine tool operation state information, thereby improving the accuracy and reliability of fault diagnosis.

[0054] Secondly, through the real-time health status evaluation technology, the present invention constructs a feature radar chart to evaluate the health status of the machine tool in real time. This evaluation can not only detect current faults in time, but also predict potential fault risks, thereby realizing predictive maintenance.

[0055] Finally, by combining the vibration signal characteristics and the radar chart, the present invention can predict the faults of the machine tool and make scientific maintenance decisions accordingly. This predictive maintenance method can effectively reduce unexpected downtime, extend the service life of the machine tool, and reduce maintenance costs. Description of the Drawings

[0056] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0057] Figure 1 is the flow chart of the machine tool predictive maintenance method based on multi-measurement point vibration sensing monitoring proposed by the present invention. Detailed Embodiments

[0058] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0059] Reference Figure 1 , a machine tool predictive maintenance method based on multi-point vibration sensing monitoring, comprising the following steps:

[0060] S1. Determine the key measuring point positions of the machine tool, install vibration sensors at the key parts of the rotating table, spindle and X, Y, Z axes of the machine tool, collect vibration signals during operation and perform preprocessing;

[0061] S2. Classify the preprocessed vibration signals into continuous rotational motion signals and non-continuous rotational motion signals, perform spectral analysis and feature extraction on the continuous rotational motion signals, and use an improved sliding window extreme value search technique to extract the effective motion process for the non-continuous rotational motion signals and splice them into a continuous signal sequence;

[0062] S3. Perform frequency domain analysis on the continuous signal sequence, convert the time domain signal into a frequency domain signal using the fast Fourier transform, analyze the frequency components and amplitudes, compare the spectral characteristics of different measuring points and different time periods, and extract the frequency domain features related to the operating state of the machine tool;

[0063] S4. Calculate the eigenvalue of the vibration signal based on the frequency domain features, use improved spatial pyramid pooling to achieve feature fusion and construct a radar chart. Each axis of the radar chart corresponds to a different eigenvalue, and each point of the radar chart represents the signal characteristics of the measuring point and time period. Compare and analyze the vibration characteristics of different measuring points and time periods based on the radar chart;

[0064] S5. Perform periodic predictive maintenance on the machine tool based on the frequency domain features and vibration characteristics of the radar chart, trigger an early warning mechanism when abnormal, perform maintenance inspections and update the maintenance strategy.

[0065] In this embodiment, the preprocessing includes noise removal, anomaly detection and signal normalization processing.

[0066] In this embodiment, the S2 specifically includes:

[0067] S21. Determine the continuity of the normalized vibration signal, decompose the continuous vibration component and the non-continuous vibration component, and set a signal decomposition model based on variational mode decomposition:

[0068]

[0069] where min is the minimum operator, is the time derivative, j is the imaginary unit, x(t) is the original vibration signal, u kis the k-th modal component, ω k is the center frequency of the modal component, K is the total number of modal components, λ is the balance parameter, u k (t) is the k-th modal component at time t, ||·|| 2 is the square of the Euclidean norm;

[0070] S22. For effectively extracting the motion process of the discontinuous rotational motion signal, an improved sliding window extreme value search algorithm with an adaptive window is adopted, and a local extreme value determination function within the window is set:

[0071]

[0072] where P(x k ) is the local extreme value probability of the k-th data point, x k is the k-th data point in the signal sequence, N w is the sliding window size, and H(·) is the Heaviside step function;

[0073] S23. For splicing the extracted effective signals, a signal alignment method based on dynamic time warping is adopted, and a calculation formula for the optimal matching path is defined:

[0074]

[0075] where D(i,j) is the cumulative matching cost of the two signals at point (i,j), and d(i,j) is the Euclidean distance between the two signals at point (i,j).

[0076] In this embodiment, S3 specifically includes:

[0077] S31. For performing time-frequency conversion on the continuous signal sequence, the fast Fourier transform is used to convert the time-domain signal into a frequency-domain signal:

[0078]

[0079] where X(f) is the frequency-domain signal, x(n) is the normalized vibration signal, N is the total number of sampling points, n is the serial number of the sampling point, f is the signal frequency, and j is the imaginary unit;

[0080] S32. For extracting the main frequency of the frequency-domain signal and comparing it with the effective motion signal extracted by local extreme value search:

[0081]

[0082] where f m is the main frequency of the signal, |X(f)| is the amplitude of the frequency-domain signal, arg max is the value of f that maximizes |X(f)|, to identify the main frequency components of the signal;

[0083] S33. Compare the spectral characteristics at different measurement points and different time periods, calculate the spectral similarity between measurement points using normalized cross-correlation, and extract the frequency-domain characteristics of the machine tool operating state:

[0084]

[0085] Among them, R xy (τ) is the normalized cross-correlation value between the signal X p (n) and X(f), X p (n) is the spliced continuous signal, and are the means of the signals X p (n) and X(f) respectively, and σ X (f) are the standard deviations of the signals X p (n) and X(f) respectively, and τ is the time shift amount.

[0086] In this embodiment, the S4 specifically includes:

[0087] S41. Calculate the eigenvalue set of the vibration signal based on the frequency-domain signal to provide a basis for fault diagnosis:

[0088]

[0089] Among them, X EE is the eigenvalue set of the vibration signal, including the effective value, peak-to-peak value, kurtosis, and impulse factor. These eigenvalues can quantitatively describe the characteristics of the signal. P(f) is the normalized energy distribution of the frequency-domain signal at frequency f, X(f) is the frequency-domain signal, F is the f value that maximizes |X(f)|, and log2 is the logarithmic function with base 2;

[0090] S42. Use the single feature of the vibration signal as the normalization factor to calculate the multi-scale entropy of the vibration signal to coordinate the multi-scale entropy with the spectral energy characteristics:

[0091]

[0092] Among them, X MSE is the multi-scale entropy of the vibration signal, S is the number of scales, X s (i) is the time series component of the spliced signal X p (i) at the s-th scale, N s is the number of sampling points at this scale, and P s (i) is the normalized probability distribution;

[0093] S43. Combine the time-domain features of multi-scale entropy with the frequency-domain features of eigenvalues to make the information weights of features at different levels more balanced, and use improved spatial pyramid pooling to fuse the vibration signal features:

[0094]

[0095] Among them, X SPP is the eigenvalue after fusion by spatial pyramid pooling, L is the number of pyramid layers, M l is the number of regional divisions in the l-th layer, X MSE,j is the value of multi-scale entropy in the j-th region, X EE,j is the value of the eigenvalue in the j-th region, W l is the weight factor of this layer, and α is the adaptive adjustment parameter;

[0096] S44. Construct a radar chart based on the fused eigenvalues, and set the calculation formulas for high-dimensional mapping and adaptive normalization radar chart:

[0097]

[0098] Among them, X radar (i) is the i-th feature point on the radar chart, D is the feature dimension, X SPP,d (i) is the d-th dimensional eigenvalue after fusion by spatial pyramid pooling, X MSE,d (i) is the eigenvalue of multi-scale entropy in the d-th dimension, W d is the weighting factor of the feature dimension, and tanh(·) is the non-linear mapping function;

[0099] S45. Each axis of the radar chart corresponds to different eigenvalues. Each point of the radar chart represents the signal features of the measuring point and time period, and combines the eigenvalues and multi-scale entropy for feature enhancement to improve the accuracy of the radar chart in machine tool fault prediction.

[0100] Example 1:

[0101] For the Makino V series five-axis linkage machining center of an aviation parts processing enterprise, to achieve predictive maintenance of the machine tool. The aviation engine blades produced by this enterprise have extremely high requirements for machining accuracy and surface quality. Once the machine tool fails, it will affect the product quality and cause production plan delays. Therefore, the enterprise hopes to improve the operating stability of the machine tool, reduce unexpected shutdowns, and improve production efficiency through the predictive maintenance method based on multi-measurement point vibration sensing monitoring provided by the present invention.

[0102] In this embodiment, the enterprise selected three Makino V80S machine tools as the pilot objects and installed vibration sensors at key positions to monitor the health status of the machine tools. After analysis, the enterprise selected five key measurement points, namely the rotary table, the spindle, the X-axis, the Y-axis, and the Z-axis of the machine tool, and installed three-axis vibration sensors with a sampling frequency of 4000 Hz at these positions to capture the minute vibration changes of the machine tool during operation in real time.

[0103] In the actual production process, this enterprise needs to perform five-axis precision machining on blade blanks every day, and the average daily operating duration of each machine tool is 18 hours. Before implementing the present invention, the enterprise mainly adopted a regular maintenance method, arranging a machine tool inspection and maintenance every 30 days. However, due to the long-term operation of the machine tools, some key components (such as spindle bearings, ball screws, etc.) may experience abnormal wear during the maintenance cycle, resulting in a decline in the machining accuracy of the machine tools and even unexpected shutdowns. Statistical data for the past six months showed that the cumulative unexpected shutdown time caused by machine tool failures in this enterprise reached 174 hours, directly affecting the delivery of 137 blades and causing an economic loss of approximately 3.12 million yuan.

[0104] After introducing the predictive maintenance method of the present invention, the enterprise uses vibration sensors to collect vibration signals during the operation of the machine tools, and extracts the effective motion process through an improved sliding window extreme value search technology, removing the fast movement and pause parts in the discontinuous motion. Subsequently, the enterprise performs frequency domain analysis on the vibration signals, extracts the main frequency components using the fast Fourier transform, and calculates characteristic values such as the effective value, peak-to-peak value, kurtosis, and impulse factor of the signals. On this basis, the enterprise constructs a radar chart to visually display the health status of the machine tools.

[0105] During the implementation process, the enterprise found that on the 26th day of the machine tool operation, the characteristic values of the radar chart at the spindle position showed obvious abnormalities. Among them, the kurtosis value increased from 3.12 during normal operation to 4.87, and the effective value increased from 0.18 m / s 2 to 0.32 m / s 2 , indicating that there may be abnormal vibration of the spindle. Further analysis of the spectrum data found that there was a sudden increase in energy at the characteristic frequency of the spindle at 7.8 kHz, which was consistent with the early characteristics of spindle bearing damage. The enterprise immediately arranged for engineers to inspect the spindle and found that there was indeed slight wear on the spindle bearing. After timely replacement of the bearing, the machine tool continued to operate normally, avoiding potential sudden failures.

[0106] With the assistance of the present invention, the enterprise achieved a transformation from regular maintenance to predictive maintenance, significantly reducing the unexpected shutdown time caused by failures. The following table shows the statistical data on the application effect of the present invention in this enterprise:

[0107] Table 1 Comparison data of Makino V80S machine tools before and after the implementation of predictive maintenance

[0108] Indicator Before predictive maintenance implementation After predictive maintenance implementation Change situation Average monthly failure times 3.6 times 1.1 times ↓69.44％ Unexpected machine tool downtime 174 hours 34 hours ↓80.46％ Maintenance cost (half year) 2.48 million yuan 1.91 million yuan ↓23.0％ Qualified rate of blade processing 96.8％ 98.5％ ↑1.7％ Economic loss (half year) 3.12 million yuan 0.57 million yuan ↓81.73％

[0109] Table 1 shows that in the past six months, the unexpected downtime of the enterprise caused by machine tool failures has decreased to 34 hours, a reduction of 80.46% compared with that before the introduction of the present invention; the average time between machine tool failures has decreased from 3.6 times per month to 1.1 times; the processing qualification rate has increased from 96.8% to 98.5%. In addition, due to the reduction of sudden failures and emergency repairs, the maintenance cost of the machine tool has been reduced by about 23%, and it is estimated that the annual maintenance cost can be saved by about 570,000 yuan.

[0110] It can be seen from this embodiment that the predictive maintenance method proposed by the present invention can effectively improve the operation stability of the machine tool, reduce sudden failures, improve production efficiency, and significantly reduce the maintenance cost, providing an efficient and reliable machine tool health management solution for manufacturing enterprises.

[0111] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A machine tool predictive maintenance method based on multi-point vibration sensor monitoring, characterized in that: The steps include: S1. Determine the key measuring point positions of the machine tool, install vibration sensors at the key positions of the rotary table, spindle and X, Y and Z axes of the machine tool, collect vibration signals during operation and perform pre-processing; S2. Classify the preprocessed vibration signal into continuous rotation motion signal and discontinuous rotation motion signal, perform spectrum analysis and feature extraction on the continuous rotation motion signal, use improved sliding window extreme value search technology to extract effective motion process on the discontinuous rotation motion signal, and splice them to form a continuous signal sequence; S3. Perform frequency domain analysis on the continuous signal sequence, convert the time domain signal into the frequency domain signal using fast Fourier transform, analyze the frequency components and amplitudes, compare the spectrum characteristics of different measuring points and different time periods, and extract the frequency domain characteristics related to the operation status of the machine tool; S4. Calculate the characteristic value of the vibration signal based on the frequency domain characteristics, use the improved spatial pyramid pooling to achieve feature fusion and construct a radar chart. Each axis of the radar chart corresponds to a different characteristic value. Each point of the radar chart represents the signal characteristics of the measuring point and time period. Compare and analyze the vibration characteristics of different measuring points and time periods based on the radar chart. S5. Based on the frequency domain characteristics and vibration characteristics of the radar chart, periodic predictive maintenance is performed on the machine tool. When an abnormality occurs, an early warning mechanism is triggered, maintenance inspections are performed, and maintenance strategies are updated.

2. The machine tool predictive maintenance method based on multi-point vibration sensor monitoring according to claim 1 is characterized in that: The preprocessing includes noise removal, anomaly detection and signal normalization processing.

3. The machine tool predictive maintenance method based on multi-point vibration sensor monitoring according to claim 1 is characterized in that: The S2 specifically includes: S21. Perform continuity judgment on the normalized vibration signal, decompose the continuous vibration component and the discontinuous vibration component, and set a signal decomposition model based on variational modal decomposition: Among them, min is the minimum operator, is the time derivative, j is the imaginary unit, x(t) is the original vibration signal, u k is the kth modal component, ω k is the center frequency of the modal component, K is the total number of modal components, λ is the balance parameter, u k (t) is the kth modal component at time t, ||·|| 2 is the square of the Euclidean norm; S22, extracting the effective motion process of the discontinuous rotation motion signal, using an improved sliding window extreme value search algorithm of an adaptive window, and setting a local extreme value judgment function within the window: Among them, P(x k ) is the local extreme value probability of the kth data point, x k is the kth data point in the signal sequence, N w is the sliding window size, H(·) is the Heaviside step function; S23, splicing the extracted valid signals, using a signal alignment method based on dynamic time warping, and defining an optimal matching path calculation formula: Among them, D(i,j) is the cumulative matching cost of the two signals at point (i,j), and d(i,j) is the Euclidean distance between the two signals at point (i,j).

4. The machine tool predictive maintenance method based on multi-point vibration sensor monitoring according to claim 1 is characterized in that: The S3 specifically includes: S31, perform time-frequency conversion on the continuous signal sequence, and use fast Fourier transform to convert the time domain signal into a frequency domain signal: Where X(f) is the frequency domain signal, x(n) is the normalized vibration signal, N is the total number of sampling points, n is the sequence number of the sampling point, f is the signal frequency, and j is the imaginary unit; S32, extract the main frequency of the frequency domain signal, and compare it with the effective motion signal extracted by local extreme value search: Among them, f m is the main frequency of the signal, |X(f)| is the amplitude of the frequency domain signal, and arg max is the value of f that maximizes |X(f)| to identify the main frequency components of the signal; S33. Compare the spectrum characteristics of different measuring points and different time periods, use normalized cross-correlation to calculate the spectrum similarity between measuring points, and extract the frequency domain characteristics of the machine tool operation status: Among them, R xy (τ) is the signal X p The normalized cross-correlation value between (n) and X(f), X p (n) is the continuous signal after splicing, and The signal X p The mean of (n) and X(f), and σ X (f) are signal X p is the standard deviation of X(n) and X(f), and τ is the time shift.

5. The machine tool predictive maintenance method based on multi-point vibration sensor monitoring according to claim 1 is characterized in that: The S4 specifically includes: S41. Calculate the characteristic value set of the vibration signal based on the frequency domain signal to provide a basis for fault diagnosis: Among them, X EE is the set of eigenvalues ​​of the vibration signal, including effective value, peak-to-peak value, kurtosis, and impulse factor. These eigenvalues ​​can quantitatively describe the characteristics of the signal. P(f) is the normalized energy distribution of the frequency domain signal at frequency f. X(f) is the frequency domain signal. F is the f value that maximizes |X(f)|. Log2 is the logarithmic function with base 2. S42. Using a single feature of the vibration signal as a normalization factor, calculating the multi-scale entropy of the vibration signal, and coordinating the multi-scale entropy with the spectrum energy characteristics: Among them, X MSE is the multi-scale entropy of the vibration signal, S is the number of scales, X s (i) is the splicing signal X p (i) Time series component at the sth scale, N s is the number of sampling points on this scale, P s (i) is the normalized probability distribution; S43. Combining the time domain features of multi-scale entropy and the frequency domain features of eigenvalues, the information weights of features at different levels are more balanced, and the vibration signal features are fused using improved spatial pyramid pooling: Among them, X SPP is the feature value after fusion through spatial pyramid pooling, L is the number of pyramid layers, M l is the number of area divisions at the lth layer, X MSE,j is the value of multi-scale entropy in the jth region, X EE,j is the value of the eigenvalue in the jth region, W l is the weight factor of this layer, and α is the adaptive adjustment parameter; S44. Construct a radar chart based on the fused eigenvalues, and set the high-dimensional mapping and adaptive normalized radar chart calculation formula: Among them, X radar (i) is the i-th feature point on the radar chart, D is the feature dimension, X SPP,d (i) is the d-th dimension feature value after spatial pyramid pooling fusion, X MSE,d (i) is the eigenvalue of the multiscale entropy in the dth dimension, W d is the weighting factor of the feature dimension, tanh(·) is the nonlinear mapping function; S45. Each axis of the radar chart corresponds to a different eigenvalue. Each point of the radar chart represents the signal characteristics of the measuring point and time period. The eigenvalue and multi-scale entropy are combined to perform feature enhancement to improve the accuracy of the radar chart in machine tool fault prediction.

Images