Milling cutter abnormity monitoring method based on blade frequency energy visualization

Abstract

A milling cutter abnormity monitoring method based on blade frequency energy visualization specifically comprises the following steps that S1, all working states of a milling cutter are collected; s2, correcting the data acquired in the step S1; s3, performing fast Fourier transform on the data obtained in the step S2 to obtain amplitudes and phases of rotation frequency harmonics from the first order to the n order of the signal; s4, reconstructing the amplitude phase information obtained in the step S3 to obtain a characteristic signal; s5, acquiring a characteristic signal obtained by selecting a group of data in the step S4; and S6, extracting the amplitude of each blade from the image obtained in the step S5. According to the method, the milling cutter fault is qualitatively analyzed, and a basis is provided for quantitative detection of the milling cutter fault; most irrelevant information and noise interference in the signals are removed, the adaptive capacity to the complex working condition environment is improved, the relative state of each blade of the milling cutter can be more clearly and stably represented, the anti-interference capacity and sensitivity of the milling cutter abnormity monitoring method are improved, and beneficial technical support is provided for safe and stable work of the milling cutter.

Description

1. Technical Field

[0001] The invention belongs to the field of tool health monitoring in a mechanical machining process, and in particular relates to a milling cutter abnormality monitoring method based on edge frequency energy visualization. 2. Background technology

[0002] With the rapid development of the world economy and science and technology, products are developing in the direction of multiple varieties, batch production, refinement and complexity. As an important processing tool of machine tools, milling cutters are widely used in plane processing, curved surface processing and complex cavity processing. In actual work, due to tool wear and degradation, improper clamping, process design errors and other reasons, a series of faults such as wear, chipping and cutter breakage will occur in the milling process. If left unchecked, the product will be scrapped and production resources will be wasted at best, and the personal safety of production workers will be threatened at worst.

[0003] In order to ensure the safe and stable operation of the milling cutter, online monitoring technology based on signal processing has become the key means of milling cutter safety monitoring, which mainly includes the following modes: (1) extracting indicators through time-frequency domain analysis and using deep convolutional neural network to complete the wear monitoring of the milling cutter; (2) using convolutional transform neural network (CNN-transformer neural network) to ) Extract the time domain characteristics of the milling cutter data to complete the monitoring of tool wear; (3) After simply extracting a large number of statistical indicators, use linear discriminant analysis to reduce the dimension of the feature set, and combine the back propagation (BP) neural network and the gray wolf optimization algorithm to complete the wear monitoring of the micro milling cutter. Although the above methods have achieved certain results in the safety of machine tool milling cutters, the above research lacks effective means when characterizing fault characteristics. Since only simple statistical indicators or deep learning algorithms are used directly for feature extraction, the indicators obtained in this way lack rationality and interpretability; in addition, when the milling cutter is working, the continuous flushing of the cutting fluid and the parallel work of the auxiliary mechanism introduce a lot of noise to the collected signal, which makes the key information of the milling process obscured. Therefore, traditional indicators will inevitably introduce irrelevant variables to the subsequent model, interfering with the accuracy of the final monitoring. In summary, under complex working conditions, it is crucial to explore a reasonable, interpretable and effective milling cutter abnormality monitoring algorithm that can effectively characterize tool fault characteristics, suppress noise interference, and eliminate irrelevant variables for abnormal monitoring of machine tool milling cutters. III. Summary of the invention

[0004] In order to overcome the existing milling cutter safety monitoring technology, due to technical limitations, there are disadvantages as described in the background. The present invention provides a method for monitoring the abnormal state of the milling cutter based on blade frequency energy visualization, which can effectively monitor the relative changes in the energy of each blade of the milling cutter in an intuitive form, thereby ensuring its safe and stable operation.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] A method for monitoring abnormality of a milling cutter based on visualization of blade frequency energy is characterized in that, first, a maximization method is used to implement full-cycle sampling to obtain more accurate amplitude and phase information, and then a characteristic signal is reconstructed according to the obtained amplitude and phase, and the characteristic signal is aligned with a reference signal using a maximum correlation method and scaled to polar coordinates, thereby obtaining a visualized blade frequency energy monitoring scheme for abnormality monitoring of a machine tool milling cutter. The method specifically includes the following process: S1: A sensor installed near the spindle of a machine tool milling cutter is used to collect single-channel vibration data of the milling cutter under various working states, which is recorded as S2: Correct the data collected in step S1 and use the maximization method to achieve full cycle sampling of the signal to obtain And the obtained signal Repeat the extension until the signal reaches 3 times the sampling frequency length, and the obtained signal is recorded as x; S3: Use the fast Fourier transform to obtain the amplitude and phase of the 1st to nth order frequency conversion harmonics of the signal from the data obtained in step S2, which are recorded as <amp i alpha i >, where amp i is the amplitude of the i-th order frequency conversion, alpha i is the phase of the i-th order frequency conversion, i = 1, 2, ... n, where n is the number of tool edges; S4: reconstruct the amplitude phase information obtained in step S3 to obtain a characteristic signal, recorded as component; S5: in step S4, the characteristic signal component obtained by selecting a set of data is marked as comp0 as a reference. Specifically, when a new set of signals is collected to obtain the characteristic signal component, the maximum correlation coefficient between the new component and comp0 is calculated, and the new component is time-shifted and corrected to obtain comp k , and comp=[comp0,comp1,…,comp k] are drawn in the same polar coordinate diagram, which is called the blade energy distribution diagram. The dynamic changes in the blade energy distribution diagram can characterize the state of each blade of the milling cutter during the milling process; S6: Extract the blade amplitude, relative ratio, minimum angle between blades and total energy from the image obtained in step S5, a total of 2n+2 characteristic indicators, and use the model for training to obtain a machine tool milling cutter abnormality monitoring model for realizing the monitoring of the milling process.

[0007] Preferably, in step S1, the specific sampling frequency fs=1000 Hz, and the sampling time is about 50 s.

[0008] Preferably, in step S2, the maximization method is used to implement the full cycle sampling of the signal, the purpose is to improve the accuracy of the subsequent amplitude and phase information calculation, that is, to remove the extra k data points at the end of the signal and set the observation index index. The specific formula is

[0009] Preferably, in step S3, the formula for obtaining the amplitude and phase of the 1st to nth order frequency conversion harmonics of the signal by fast Fourier transform is: The amplitude formula involved in the i-th order frequency conversion is amp i =|F(i×fn)|, the formula for the i-th order frequency conversion is alpha i =angle(F(i×fn)).

[0010] Preferably, in step S4, the specific characteristic signal component reconstruction formula is

[0011] Preferably, in step S6, 60% of the data involved are randomly selected from the samples as a training set, and the remaining samples are used as a test set, and the weighted KNN model is used for training. The distance measure of the weighted KNN model used is the Euclidean distance, the neighbor parameter is set to 10, the square of the Euclidean distance is used as the weight, and the remaining 40% of the data is used as a test set.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention mainly utilizes the visualized characteristic signal technical solution to characterize the changing law of the milling cutter fault on the blade, realizes the qualitative analysis of the milling cutter fault, and improves the basis for the subsequent quantitative detection of the milling cutter fault; and utilizes the continuation-reconstruction method to extract key order harmonics, removes most of the irrelevant information and noise interference in the signal, greatly improves the adaptability of the method to complex working conditions, can more clearly and stably characterize the relative state of each blade of the milling cutter, improves the anti-interference ability and sensitivity of the milling cutter abnormality monitoring method, and provides favorable technical support for the safe and stable operation of the milling cutter. IV. Description of the attached drawings

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] Figure 1 It is a schematic diagram of the test bench structure.

[0015] Figure 2 This is a picture of a milling cutter in a faulty state (left: a milling cutter with slight wear; right: a milling cutter with chipped edges).

[0016] Figure 3 Flow chart of the present invention.

[0017] Figure 4 The original waveform diagram of the normal state of the present invention.

[0018] Figure 5 Time domain diagram (left) and frequency domain diagram (right) of the truncated signal of the present invention.

[0019] Figure 6 The maximization method of the present invention realizes the whole cycle sampling diagram.

[0020] Figure 7 The time domain (left) and frequency domain (right) diagrams of the positive and negative signals of the present invention.

[0021] Figure 8 Characteristic signal diagram of normal state and slight wear of the present invention.

[0022] Fig. 9 Blade energy distribution diagram of the machining process under various states.

[0023] Fig.10 The training result diagram of the invented milling cutter abnormality monitoring model.

[0024] Fig.11 Result diagram of the abnormal monitoring test of the milling cutter of the present invention. V. Specific implementation methods

[0025] Figure 3 As shown in the figure, a method for monitoring abnormalities of milling cutters based on visualization of blade frequency energy is firstly used to implement full-cycle sampling using the maximization method to obtain more accurate amplitude and phase information, and then the characteristic signal is reconstructed according to the obtained amplitude and phase. The characteristic signal is aligned with the reference signal using the maximum correlation method and scaled to polar coordinates to obtain a visualized blade frequency energy monitoring scheme for abnormalities of machine tool milling cutters. The present invention is tested on a 3-DOF FANUC small machining center. The maximum speed of the machine tool electric spindle is 24,000 rpm and the maximum load is 300 kg. KISLER acceleration sensors are used and magnetically mounted on the sides of the spindle. The schematic diagram of the test bench structure is shown in the figure. Figure 1As shown, the experimental milling cutter is a 4-edge milling cutter, and a total of 5 states are tested and verified, including normal state, light wear state, severe wear state, chipping, and severe chipping. The working speed is 3000rpm and the cutting depth is 0.2mm. The actual picture of the tool under fault state is as follows Figure 2 The specific workflow is as follows.

[0026] Figure 3 As shown in , step 1, use the acceleration sensor installed near the milling cutter spindle to collect the single-channel vibration data of the milling cutter in various states, the sampling frequency fs = 1000 Hz, and the sampling time is about 50s, as shown in Figure 4 In order to expand the number of samples, the collected data is divided into segments of 1 second each to obtain several samples, which are recorded as After truncation, the signal is Figure 5 shown; from Figure 5 It is found that there is serious energy dissipation in the signal spectrum, and the key harmonics, especially the rotation frequency and its high-order harmonics, are relatively vague, which interferes with the subsequent amplitude and phase extraction, easily leading to inaccurate feature extraction and ultimately misjudgment of the tool state, so further processing is required in the next step. Step 2: Get Any sample in For example, To improve the accuracy of subsequent amplitude and phase information calculations, the maximization method is used to achieve full-cycle sampling of the signal, that is, to remove the extra k data points at the end of the signal. Set the observation index index, as shown in formula (1):

[0027]

[0028] Among them, ω i is the frequency of interest. Considering the need to calculate the amplitude and phase of the conversion frequency to the blade frequency, in this embodiment, ω is taken i =50,100,150,200Hz, i.e. 1st to 4th order frequency conversion, N is The data length, index changes with k as follows Figure 6 As shown. Figure 6 Find the k value corresponding to the maximum value in the example data In the example, when 192 points are deleted at the end, full cycle sampling can be achieved. The signal after deletion is recorded as x * ; To further refine the spectrum, the signal x obtained by deleting the correction * Repeat the extension until the signal reaches at least 3 times the sampling frequency length. The sample data is repeated 4 times and the obtained signal is recorded as x. Figure 7 As shown. Figure 7 It can be found that compared with Figure 5The signal in the image is more clear, and the amplitude of the key frequency of the corrected signal is clearer, and the energy dissipation phenomenon is effectively improved.

[0029] Figure 3 As shown in , step three, the data in step two is transformed by fast Fourier transform to obtain the amplitude and phase of the 1st to nth order frequency conversion harmonics of the signal, as shown in formula (2):

[0030]

[0031] Where F(n) is the frequency domain signal, x(t) represents the t-th sampling point of the signal x, and j is the imaginary unit; they are respectively denoted as <amp i alpha i >, then the amplitude of the i-th order frequency conversion is as shown in formula (3):

[0032] amp i =|F(i×fn)| (3)

[0033] Wherein, fn=50Hz is the rotation frequency, i=1,2,…n, where n is the number of tool edges, n=4;

[0034] The phase of the i-th order frequency conversion is shown in formula (4):

[0035] alpha i =angle(F(i×fn)) (4)

[0036] Wherein, angle(·) represents the calculation of the complex phase angle, that is, the calculation of the angle between the complex vector and the positive direction of the x-axis. The amplitude and phase information obtained in this step are used in step 4 to accurately reconstruct the characteristic signal to obtain the fault information hidden in the signal.

[0037] Figure 3 As shown in , step 4, the characteristic signal is reconstructed according to the amplitude and phase information obtained in step 3. First, the sequence θ is constructed. hat In order to simplify the subsequent calculations as much as possible and ensure the signal clarity, the control sequence length is 1 cycle and the sequence length is set to 500, as shown in formula (5):

[0038] θ hat =[0 2π / 499 … 2π] T (5)

[0039] in, The sequence length is m = 500. Considering that the amplitude of the rotation frequency has little effect on characterizing the difference in the blade state and will cover up more detailed information in the characteristic signal, the characteristic signal component is reconstructed starting from the second-order harmonic of the rotation frequency. As shown in formula (6):

[0040]

[0041] The characteristic signal component is reconstructed as a superposition of simple harmonic signals. To facilitate the next step of analysis, the minimum value of the component is set to 0. The characteristic signal extracted from the sample data is as follows: Figure 8 The characteristic signal obtained in this step removes most of the environmental noise interference and other frequency components, and contains rich fault information, providing stable and reliable support for subsequent abnormal monitoring.

[0042] Figure 3 As shown in , step 5, select the characteristic signal component obtained from the first group of data of the above five states, marked as comp0, as a reference. The new component obtained later is cross-correlated with comp0, as shown in formula (7):

[0043]

[0044] Find the k corresponding to the maximum value of R(k), and use it to perform time-shift correction on the new component to obtain comp k ; Set comp = [comp0, comp1, ..., comp k ] are plotted in the same polar coordinate diagram, such as Fig. 9 As shown. Fig. 9 It can be easily found that as the fault appears and deepens, the four originally clear "blades" in the blade energy distribution diagram gradually become blurred, and the overall amplitude becomes larger; this phenomenon indicates that with the occurrence of milling cutter failure, the energy distribution of each blade cutting has undergone tremendous changes. The dynamic changes in the blade energy distribution diagram can be used to monitor the milling cutter status.

[0045] Figure 3 As shown in the figure, in step 6, to assist in monitoring the status of the milling cutter, 10 feature indicators, including the amplitude of each blade, relative ratio, minimum angle between blades, and total energy, are extracted from the image. 60% of the data are randomly selected from the sample as the training set, and the remaining samples are used as the test set. The weighted KNN model is used for training. The distance measure of the weighted KNN model used is the Euclidean distance, the neighbor parameter is set to 10, and the square of the Euclidean distance is used as the weight. The confusion matrix of the training result is as follows: Fig.10 Using the remaining 40% of the data for testing, the test results are as follows Fig.11 As shown in the figure, it can be seen that the test accuracy is as high as 99.07%. In order to eliminate contingency, the above operation is repeated 10 times, and the average test accuracy is 98.45%. This proves that the method of the present invention can effectively identify the abnormal state of the milling cutter.

[0046] Figure 3As shown in, through the above, the present invention uses visualized characteristic signals to characterize the changing rules of milling cutter faults on the cutting edge, realizes the qualitative analysis of milling cutter faults, and improves the basis for subsequent quantitative detection of milling cutter faults; uses the extension-reconstruction method to extract key order harmonics, removes most of the irrelevant information and noise interference in the signal, greatly improves the method's adaptability to complex working conditions, and can more clearly and stably characterize the relative states of each milling cutter blade, improving the anti-interference ability and sensitivity of the milling cutter abnormality monitoring method.

[0047] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A milling cutter abnormality monitoring method based on edge frequency energy visualization, characterized in that: Firstly, the maximization method is used to realize full cycle sampling to obtain more accurate amplitude and phase information, and then the characteristic signal is reconstructed according to the obtained amplitude and phase. The characteristic signal is aligned with the reference signal using the maximum correlation method and scaled to polar coordinates. A visual blade frequency energy monitoring scheme is obtained for abnormal monitoring of machine tool milling cutters, which specifically includes the following process: S1: The single-channel vibration data of the milling cutter under various working states is collected using a sensor installed near the spindle of the machine tool milling cutter, which is recorded as S2: Correct the data collected in step S1 and use the maximization method to achieve full cycle sampling of the signal to obtain And the obtained signal Repeat the extension until the signal reaches 3 times the sampling frequency length, and the obtained signal is recorded as x; S3: Use the fast Fourier transform to obtain the amplitude and phase of the 1st to nth order frequency conversion harmonics of the signal from the data obtained in step S2, which are recorded as <amp i alpha i >, where amp i is the amplitude of the i-th order frequency conversion, alpha i is the phase of the i-th order frequency rotation, i=1,2,…n, where n is the number of tool edges; S4: reconstruct the amplitude phase information obtained in step S3 to obtain a characteristic signal, recorded as component; S5: Step S4 obtains the characteristic signal component obtained by selecting a set of data, marked as comp0, as a benchmark. Specifically, when a new set of signals is collected to obtain the characteristic signal component, the maximum correlation coefficient between the new component and comp0 is calculated, and the new component is time-shifted and corrected to obtain comp0. k , and comp=[comp0,comp1,…,comp k ] are drawn in the same polar coordinate diagram, which is called the blade energy distribution diagram. The dynamic changes in the blade energy distribution diagram can characterize the state of each blade of the milling cutter during the milling process; S6: Extract 2n+2 characteristic indicators including the blade amplitude, relative ratio, minimum angle between blades and total energy from the image obtained in step S5, and use the model for training to obtain a machine tool milling cutter abnormality monitoring model for monitoring the milling process.

2. The method for monitoring abnormality of a milling cutter based on edge frequency energy visualization according to claim 1, characterized in that: In step S1, the specific sampling frequency fs=1000 Hz, and the sampling time is about 50 s.

3. The method for monitoring abnormality of a milling cutter based on edge frequency energy visualization according to claim 1, characterized in that: In step S2, the maximization method is used to achieve full-cycle sampling of the signal, in order to improve the accuracy of subsequent amplitude and phase information calculations, that is, to remove the extra k data points at the end of the signal and set the observation index index. The specific formula is 4. The method for monitoring abnormality of a milling cutter based on edge frequency energy visualization according to claim 1, characterized in that: In step S3, the formula for obtaining the amplitude and phase of the 1st to nth order harmonics of the signal by fast Fourier transform is The amplitude formula involved in the i-th order frequency conversion is amp i =|F(i×fn)|, the formula for the i-th order frequency conversion is alpha i =angle(F(i×fn)).

5. The method for monitoring abnormality of a milling cutter based on edge frequency energy visualization according to claim 1, characterized in that: In step S4, the specific characteristic signal component reconstruction formula is 6. The method for monitoring abnormality of a milling cutter based on edge frequency energy visualization according to claim 1, characterized in that: In step S6, 60% of the data involved are randomly selected from the samples as the training set, and the remaining samples are used as the test set. The weighted KNN model is used for training. The distance measure of the weighted KNN model used is the Euclidean distance, the neighbor parameter is set to 10, the square of the Euclidean distance is used as the weight, and the remaining 40% of the data is used as the test set.

Images