TWI517934B - Method to separate signal and noise for chatter monitor and device thereof - Google Patents

Description

Signal and noise separation method and device for flutter monitoring

The present invention relates to a technique for flutter monitoring, and more particularly to a method and apparatus for noise noise separation for flutter monitoring applied during tool machining.

When the general tool machine is used for machining, the chattering is an abnormal state of machining. The flutter is caused by the unintended relative motion between the tool and the workpiece, resulting in poor surface precision of the machined workpiece and may even damage the machine tool structure or workpiece. The cheapest way to monitor flutter is to monitor the sound of vibration through a microphone. However, the machine itself or the surrounding environment may also generate noise, which may affect the judgment of the flutter monitoring. For example, because the erosion sound intensity of the cutting fluid is greater than the stable machining cutting sound, it may cause the monitoring program to misjudge the occurrence of chattering. Or because of unexpected sounds around the machine, it may also cause the program of the monitoring machine to misjudge the occurrence of chatter. In view of this, techniques for improving flutter monitoring have been proposed in the field of flutter monitoring. However, these improved technologies fail to identify the sound of the machine tool itself and the processing environment, so that filtering is performed during the monitoring process. Audio noise does not work well.

In fact, existing noise filtering or separation techniques include wavelet transform, low-pass filtering, error averaging, EMD modal decomposition, local enhancement, background noise cancellation, etc., using filtering, enhancement, Techniques such as signal subtraction to eliminate noise, but are accompanied by the disadvantage of signal distortion. Only Independent Component Analysis is considered to be a method of separating signals without loss of signal characteristics, which provides more reliable feature signal identification; however, the independent component analysis algorithm cannot actively provide any feature signal prompts. In other words, when the independent component analysis obtains the array signal components based on the original signal analysis, the independent component analysis cannot distinguish which one is the characteristic signal or which is the noise. For example, when the cutting sound is monitored by the microphone, the machining sound of the cutting edge cutting workpiece and the machining fluid are collected. The sound, the sound of the spindle bearings, and other surrounding sounds. At this time, the independent component analysis algorithm can be used to analyze the sound of the microphone measurement, and several signal components are obtained; however, these signal components cannot know which is the processing sound of the blade cutting workpiece, the sound of the machining liquid flushing, and the running sound of the spindle bearing. Or other sounds. Therefore, it is necessary to characterize these signal components in order to obtain the feature signals to be monitored.

The following is a few examples of prior art documents to understand the parts of the existing flutter monitoring equipment that need to be improved. For example, the improvement of the US Patent No. 5,170,358 focuses on the interaction technology between the algorithm and the surrounding electromechanical equipment. The case reveals the sound measured by the filter and then transmits it to the flutter monitoring system, but the on-site processing environment noise may come from personnel conversation, mechanical operation sound, cutting fluid flushing, and current noise, etc., by a single filter. The results of the flutter signal analysis are not good.

For example, the US Patent No. 4853680 is also a machine tool monitoring system and method, which discloses that a sensor is used to sense the vibration state between the tool and the workpiece, and a signal is generated, and the signal is analyzed by digitally to improve the machine tool. Vibration conditions during processing.

It can be seen from the technical contents disclosed in the above previous cases that the possible sources of flutter are not fully recognized, collected, and analyzed, so that the flutter monitoring effect is not good. Therefore, the present inventors have studied and developed the noise noise separation method and apparatus of the present invention using the independent component analysis technology and the ideal characteristics of the processing signal to provide a more reliable characteristic signal vibration for the shortcomings of the flutter monitoring device. The vibration determination algorithm is used to improve the existing technology and improve the reliability of the flutter identification.

The method and device for separating the noise and vibration of the flutter monitoring of the present invention configure the position of the microphone by using the characteristics of the sound transmission at high and low frequencies, and when the tool is machined, the flutter spectrum, the processing characteristic spectrum and the noise spectrum are obtained by independent component analysis, that is, More reliable judgments can be made to adjust the processing conditions.

In order to achieve the object and effect of the above invention, the present invention discloses a method and device for separating the noise and noise of the flutter monitoring. The vibration and noise separation device of the flutter monitoring includes a controller for controlling the spindle speed and two detecting sounds for cutting. The first monitoring microphone and the second monitoring microphone, and a flutter analysis module. The flutter analysis module is used for receiving A signal from the first original spectrum signal and the second original spectrum is set to identify whether chattering has occurred. The flutter analysis module further comprises a signal processing module, an independent component analysis module, a component identification module and a flutter identification module. The first monitoring microphone is disposed near the tool for measuring the first original spectrum; the second monitoring microphone is disposed at a distance from the cutting position for measuring the second original spectrum. The signal processing module generates a modified spectrum, a simulated processing spectrum and a simulated flutter spectrum spectrum group from the first original spectrum and the second original spectrum; the independent component analysis module analyzes the spectrum group to generate a component spectrum a group and a separation matrix; the component identification module identifies the processing spectrum and the flutter spectrum according to the component spectrum group and the separation matrix described above; and further utilizes a flutter identification module, according to the separation matrix, the processing spectrum and the vibration The vibration spectrum checks for flutter.

Due to the increased transmission distance during the sound transmission, the high frequency signal is more easily attenuated than the low frequency signal. Therefore, the first monitoring microphone that detects flutter should be placed close to the processing position to reduce the distance effect. When the sound is transmitted, it is difficult to measure the stable signal in the near sound field, and the microphone can only measure the clear signal in the far/free sound field. The separation point of the far/near sound field is about 1 to 2 times the wavelength λ.

The method for separating the noise and vibration of the flutter monitoring of the present invention is the signal noise separating device with the flutter monitoring, and the method for separating the noise and vibration of the flutter monitoring comprises the following steps: Step 1: capturing the sound source signal, Acquiring the first original spectrum by using the first monitoring microphone in the near processing area, and obtaining the second original spectrum by the second monitoring microphone remote from the processing place; Step 2: calculating the modified spectrum: according to the processing sound source, the first monitoring microphone, and the second monitoring The positional relationship of the microphone eliminates the spectral attenuation effect in the spectrum, and the original spectrum produces a modified spectrum. Step 3: Check the machining state. If the tool does not actually cut the workpiece, return to step 1 and continue to retrieve the processed sound source signal; Step 4: Create a simulation The spectrum group establishes a segment filter by the blade pass frequency and its multiplication frequency, and generates a simulated processing spectrum from the modified spectrum; a band filter is established with a flutter frequency range, and an abnormal spectrum is generated from the modified spectrum; Step 5: Perform independent component analysis. The input is the modified spectrum, the simulated processing spectrum and the simulated abnormal spectrum. The output is the component spectrum group and the separation matrix. Step 6: Identify the processing spectrum and the abnormal spectrum according to the component spectrum group and the separation matrix to calculate the processing feature strength. And flutter characteristic strength; Step 7: Perform flutter identification. If the ratio of the flutter characteristic intensity to the processing characteristic intensity is higher than the threshold value, the flutter avoidance processing is performed; if not, return to step 1 to continue the processing. Signal.

10‧‧‧Tool machine

11‧‧‧ Spindle

12‧‧‧ Controller

13‧‧‧Tools

14‧‧‧Workpiece

20‧‧‧First monitoring microphone

21‧‧‧Second monitoring microphone

30‧‧‧Signal Processing Module

S ₁ ‧‧‧The first monitoring microphone and tool distance

S ₂ ‧‧‧Second monitoring microphone and tool distance

40‧‧‧Independent Component Analysis Module

50‧‧‧Component Identification Module

60‧‧‧ flutter identification module

Figure 1 is a schematic view of an embodiment of the invention applied to a machine tool.

Figure 2(A) shows the signal separation point representation of the far/near sound field.

Figure 2 (B) The first and second monitoring microphone position measurement signal map.

Figure 3 Flow chart of spectrum acquisition, calculation, analysis and identification of the present invention.

The spectrum of Fig. 4(A) is corrected to the simulated and simulated abnormal spectrum.

Figure 4(B) Enter the signal to the ICA spectrum analysis.

Figure 4(C) Figure ICA Spectrum analysis components.

Figure 4(D) Spectrum Spectrum intensity map.

Figure 4(E) A spectrogram of the vibration of the tool head mount using an acceleration gauge.

FIG. 5 is a flow chart of spectrum spectrum acquisition, calculation, analysis, identification, and storage of another embodiment of the present invention.

In order to provide a further understanding and understanding of the structure and function of the present invention, the embodiments of the present invention will be described in detail with reference to the accompanying drawings. Please refer to FIG. 1 , which is a schematic diagram of a vibration and noise monitoring device for flutter monitoring applied to a power tool 10 according to the present invention. The noise and noise separation device of the flutter monitoring includes a controller 12 for controlling the rotation speed of the spindle 11, two first monitoring microphones 20 and a second monitoring microphone 21 for detecting the cutting sound, and a flutter analysis module. The flutter analysis module further includes a signal processing module 30, an independent component analysis module 40, a component identification module 50, and a flutter identification module 60. The first monitoring microphone 20 is disposed near the tool 13 for measuring the first original spectrum; the second monitoring microphone The wind 21 is disposed away from the cutting position for measuring the second original spectrum. The signal processing module 30 generates a modified spectrum, a simulated processing spectrum, and an abnormal spectrum spectrum group from the first original spectrum and the second original spectrum; the independent component analysis module 40 is configured to analyze the spectrum group to generate components. a spectrum group and a separation matrix; the component identification module 50 identifies the processing spectrum and the flutter spectrum according to the component spectrum group and the separation matrix described above; and further uses a flutter identification module 60 to process the separation matrix according to the foregoing The spectrum and flutter spectrum check for flutter, and the flutter avoidance process is performed in response to the occurrence of chatter.

The method for separating the noise and vibration of the flutter monitoring of the present invention comprises the following steps: Step 1: extracting the processed sound source signal, and obtaining the first original spectrum by using the first monitoring microphone at the near processing point, and the second monitoring by the processing station The second original spectrum is obtained by the microphone; Step 2: Calculating the corrected spectrum: according to the positional relationship of the processing sound source, the first monitoring microphone, and the second monitoring microphone, eliminating the spectrum attenuation effect in the spectrum, and the original spectrum generating the corrected spectrum; Step 3, checking In the machining state, if the tool does not actually cut the workpiece, return to step 1 and continue to capture the processed sound source signal; step four, establish a pseudo-spectrum group, establish a segment filter with the blade pass frequency and its multiplier, and generate a simulation from the modified spectrum. Processing spectrum; band filter is established in the range of flutter frequency, and pseudo-abnormal spectrum is generated from the modified spectrum; step 5, independent component analysis is performed, and the input is the modified spectrum, the simulated processing spectrum and the simulated abnormal spectrum, and the output is the component spectrum group and separation. Matrix; step six, according to the component spectrum group and the separation matrix to identify the processing spectrum and the abnormal spectrum, Processing characteristic intensity and flutter characteristic intensity; Step 7. Perform flutter identification. If the ratio of the flutter characteristic intensity to the processing characteristic intensity is higher than the threshold value, perform flutter avoidance processing; if not, return to step one to continue Capture processing sound source signals.

Please refer to FIG. 3, which is a flowchart of a method for separating noise and noise according to the flutter monitoring of the present invention. The first monitoring microphone 20 of the noise and noise separation device of the flutter monitoring device measures the first original spectrum, and the second monitoring microphone 21 measuring the second original spectrum, because the transmission distance increases, the high frequency signal is more easily attenuated than the low frequency signal during the sound transmission process, and the positional relationship between the processing sound source, the first monitoring microphone and the second monitoring microphone is utilized. Eliminate the frequency attenuation effect in the spectrum to produce a modified spectrum. Therefore, the first monitoring microphone 20 for detecting flutter should be disposed close to the processing position to reduce the distance effect, that is, the first monitoring microphone 20 is disposed at a distance a of the cutter 13 to be less than 69 cm (cm). The distance between the second monitoring microphone 21 and the tool 13 should be as far as possible from the tool 13. When the sound is transmitted, it is difficult to measure the stable signal in the near sound field. As shown in the second (A), the microphone can only measure the clear signal in the far sound field. The separation point of the far/near sound field is about 1 to 2 times the wavelength, and the wavelength λ is as follows:

The flutter frequency is mostly the spindle-knife-tool flexible mode, which is about 1000~5000Hz. After conversion, the first monitoring microphone 20 that measures the dither frequency should be within 7 to 69 centimeters (cm) of the tool 13; if the frequency of the dither is high, the first monitoring microphone 20 needs to be closer to the tool. In order to keep the relative position of the microphone and the tool constant, the optimal position of the first monitoring microphone 20 is the spindle head of the machine tool. For example, by using the impact tamper test, the frequency range of the chatter vibration can be further confirmed, and the position of the first monitoring microphone 20 can be adjusted.

Also affected by the air damping effect, the intensity of the high frequency sound decays faster. For example, the data graph of Fig. 2(B) is the first monitoring microphone amount 20 and the second monitoring microphone 21 measured at 34 cm from the processing point, and the signal intensity attenuation of the microphone at 120 cm is observed. fast. Therefore, the signals obtained by the microphone measurement are not actual processing signals, and need to be corrected by the signal processing module 30. The following is the spectrum representation vector formula, X = { x _i } ^T = { x ₁ , x ₂ , x ₃ ,..., x _n } ^T

Where x _i is the response strength of frequency i . If the signal at the tool, ie the processing spectrum, is: Y =( y _i } ^T =( y ₁ , y ₂ , y ₃ ,..., y _n } ^T

The first original spectrum can be expressed as:

The response strength of each frequency can be expressed as

Similarly, the second original spectrum is

Where E = { ε _i } is the noise spectrum, r ( ω ) is the attenuation coefficient, and S ₁ and S ₂ are the distances from the tool to the microphone, respectively. A and b are constant coefficients, respectively, and are affected by the measurement environment and hardware specifications.

In order to restore the spectrum signal at the tool, the attenuation coefficient is eliminated first. The invention utilizes the characteristics that the low-frequency signal intensity is less affected by the propagation distance, and performs proportional correction. During the processing, the frequency of the blade passing frequency y _tp is lower and the intensity is more significant. Therefore, it is assumed that the y _tp intensity of the two original spectra is the same, so it can be used as a reference for eliminating the constant coefficients a ₁ and a ₂ . That is, the blade pass frequency in the first original spectrum is

Since the blade passes the frequency intensity significantly, the influence of other components can be ignored.

If the blade pass frequency strength is equal to one, the intensity changes of other frequencies are:

Similarly, the frequency strength of the second original spectrum can also be adjusted to

Let the above two equations be subtracted, and the approximate value of the attenuation function is obtained.

Substitute the above formula or Corrected spectrum

among them

The distances s ₁ and s ₂ can be calculated based on the controller's machine coordinates and the actual microphone position.

Then check the machining status and check if the tool has actually cut the workpiece. Conventional techniques are usually based on spindle current or load information. The spindle load during tool blank cutting is low, while the spindle load during actual cutting is quite high, especially during rough cutting that requires flutter monitoring. Very significant difference. It is also possible to check whether the blade pass frequency in the corrected spectrum is the main peak of the spectrum. If yes, it means that the obtained sound information is in the process of processing. If not, it means that the tool has not been cut. If the tool actually cuts the workpiece, continue with the next flutter monitoring; if not, continue to use the microphone to capture the signal.

Next, using the independent component analysis algorithm to analyze the spectral combination requires a reference for analysis. According to the blade pass frequency and its multiplier to create a segment filter, please refer to the flowchart of Figure 3 and the segment filter shown in Figure 4 (A). The imitation processing spectrum X ^m can be obtained by correcting the inner spectrum of the spectrum and the segment filter. On the other hand, according to the field personnel experience, the computer intelligence record or the area where the real part of the tool frequency response function on the machine is negative, a band filter is established, and the inner spectrum of the spectrum and the segment filter is corrected. Imitation of the anomalous spectrum X ^c . Further, as shown in FIG. 4(B), the signal input to the independent component analysis module 40 includes a corrected spectrum X ⁰ , a simulated spectrum X ^m and an abnormally anomalous frequency X ^c spectrum, wherein the latter two are used as components. For reference purposes.

Figure 4(C) shows the results of the independent component analysis, including the three component spectra U ¹⁰ , U ²⁰ and U ³⁰ . In order to extract the processing spectrum and the flutter spectrum from the three component spectra, the component identification module 50 requires a separation matrix generated by the independent component analysis process, for example:

However, the composition magnitude of each component spectrum is not necessarily the first, so the component spectrum is first unitized, that is, the sum of the frequency intensities of the spectrums of the components is unitized. For example, if the component spectrum is scaled to a total of 1.0, the frequency intensity will be output. The result of Figure 4(D). At the same time, the separation matrix coefficient is corrected to

From the composition factor of the imitation processing spectrum, ie X ^m =0.15 U ¹ +4.91 U ² -1.22 U ³

It can be known that the second coefficient (4.91) in the above equation is the largest coefficient, so the second component spectrum U ² should be the processing spectrum U ^m , that is, U ² =U ^m . Similarly, it can be seen from the combination coefficient of the pseudo-abnormal spectrum that the first coefficient (30.41) is the largest coefficient, so the first component spectrum U ¹ should be the dither spectrum U ^c , that is, U ¹ =U ^c . Similarly, other feature spectra can be identified if needed. In this way, the processing spectrum and the flutter spectrum can be quickly identified. Fig. 4(E) shows the vibration of the headstock using the acceleration gauge measuring tool. Since it is not affected by the sound noise, it should indicate the vibration of the machining. The result is similar to the processing spectrum and the flutter spectrum, which proves the invention. The method does achieve the effect of separating vibration and noise.

Then, according to the processing spectrum and the flutter spectrum obtained by the component identification module 50, the flutter identification module 60 can calculate the processing feature intensity and the flutter characteristic intensity and determine whether the flutter is generated. The modified spectrum of the previous example can be expressed as X ⁰ = 30.42 U ^c +5.0 U ^m +6.88 U ³

The maximum frequency intensity is taken from the processing spectrum and the flutter spectrum, and multiplied by the corresponding modified spectral composition coefficient as the characteristic intensity. For example, the coefficient of the processing spectrum U ^m is 5.0, and the maximum value of the processing spectrum U ^m is multiplied by 5.0 as processing. Feature strength. Similarly, the coefficient of the flutter spectrum U ^c is 30.42, and the maximum value of the flutter spectrum U ^c is multiplied by 30.42 as the flutter characteristic intensity. When the ratio of the flutter characteristic intensity to the processed feature intensity is higher than the threshold value, it is determined that the flutter is generated, and generally the threshold value may be 0.5 or less. In the previous case, the resulting ratio is 0.88, indicating that flutter has occurred and flutter avoidance should be performed. If flutter generation is not detected, continue to capture the sound with the microphone for flutter identification.

Please refer to FIG. 5 again, which is another embodiment of the present invention, which is used to process large regular noises, such as processing fluid flushing during processing. When the machining state is checked, if it is judged that the tool 14 has not been cut at the time, the corrected spectrum is stored as the background spectrum. When the processing process sound is subsequently obtained, the background spectrum and the aforementioned modified spectrum, the simulated processing spectrum and the simulated abnormal spectrum are input to the independent component analysis module 40, and after analysis, four component spectra are generated. By this action, the regular noise can be separated, and the subsequent flutter identification process is as described in the previous embodiment, and will not be described again.

In summary, the noise noise separation method and device for flutter monitoring of the present invention can check whether the machine tool is in a processing state, can identify the flutter characteristics, and can surely achieve the effect of separating the vibration and noise signals.

The above embodiments are merely illustrative of the technical features and effects of the present invention, and are not intended to limit the scope of the invention. That is, the equivalent changes and modifications made by the patent application scope of the present invention should still fall within the scope of the claims covered by the present invention.

10‧‧‧Tool machine

11‧‧‧ Spindle

12‧‧‧ Controller

13‧‧‧Tools

14‧‧‧Workpiece

20‧‧‧First monitoring microphone

21‧‧‧Second monitoring microphone

30‧‧‧Signal Processing Module

S ₁ ‧‧‧The first monitoring microphone and tool distance

S ₂ ‧‧‧Second monitoring microphone and tool distance

40‧‧‧Independent Component Analysis Module

50‧‧‧Component Identification Module

60‧‧‧ flutter suppression module

Claims (9)

A signal noise separation device for flutter monitoring includes: a first monitoring microphone disposed near the tool for acquiring a first original spectrum; and a second monitoring microphone disposed at a distance away from the tool for obtaining a second original spectrum a flutter analysis module for identifying flutter from the signals of the first original spectrum signal and the second original spectrum; wherein the flutter analysis module further comprises: a signal processing module, by the first original spectrum And the second original spectrum generates a modified spectrum, a simulated processing spectrum and a simulated flutter spectrum spectrum group; an independent component analysis module is configured to analyze the spectrum group to generate a component spectrum group and a separation matrix; a component identification module, according to The component spectrum group and the separation matrix are used to identify the processing spectrum and the flutter spectrum; and a flutter identification module checks whether the flutter occurs according to the separation matrix, the processing spectrum and the flutter spectrum. The signal noise separation device for flutter monitoring according to claim 1, wherein the first monitoring microphone is disposed within 69 cm from the processing. The noise and noise separation device for flutter monitoring according to claim 1, wherein the first monitoring microphone is disposed on the machine tool spindle head to maintain a constant relative distance between the microphone and the tool. The noise and noise separation device for flutter monitoring according to claim 1, wherein the first and second monitoring microphones can be replaced by an acceleration gauge to obtain an original spectrum. A method for separating noise and noise with a noise and noise separation device for flutter monitoring according to claim 1 of the patent application, comprising: Step 1: capturing a processed sound source signal, obtained by a first monitoring microphone at a near processing station The first original spectrum is obtained by the second monitoring microphone remote from the processing station; Step 2: calculating the modified spectrum: eliminating the spectral attenuation in the spectrum according to the positional relationship between the processing sound source, the first monitoring microphone and the second monitoring microphone Effect The starting spectrum generates the corrected spectrum; Step 3: Check the machining state. If the tool does not actually cut the workpiece, return to step 1 and continue to capture the processed sound source signal; Step 4: Establish the pseudo-spectrum group, establish the cutting edge frequency and its multiplier A segment filter that generates a simulated spectrum from the modified spectrum; a band filter is established with a range of flutter frequencies, and an abnormal spectrum is generated from the corrected spectrum; and step 5: independent component analysis is performed, and the input is a modified spectrum, a simulated spectrum, and a simulated Abnormal spectrum, the output is the three component spectrum and the separation matrix; Step 6: Identify the processing spectrum and the anomalous spectrum according to the component spectrum group and the separation matrix, and calculate the processing feature intensity and the flutter characteristic intensity; Step 7: Perform flutter identification, if the vibration If the ratio of the intensity of the vibration characteristic to the intensity of the processing characteristic is higher than the threshold value, the flutter avoidance processing is performed; if not, the processing returns to the step 1 to continue to capture the processed sound source signal. The method for separating noise according to claim 5, wherein, in the process of checking the machining state in step three, if the tool does not actually cut the workpiece, the corrected spectrum is stored as a background spectrum. The method for separating noise and noise according to claim 5, wherein, in the process of performing independent component analysis in step 5, the input is a modified spectrum, a background spectrum, a simulated processing spectrum, and a simulated abnormal spectrum; and the output is a four-component spectrum group. With a separation matrix. For example, in the noise separation method described in claim 5 or 7, wherein the step 2 is used to calculate the corrected spectrum, the blade pass frequency intensity is used as a reference. For example, the noise noise separation method described in claim 5, wherein the simulated abnormal spectrum is based on the field personnel experience, the computer intelligence record, or the negative value of the real part of the tool frequency response function on the machine. Filter, corrected spectrum and segment filter inner product.