TWI680370B - Time series signal analysis method - Google Patents

Abstract

The time-series signal analysis method of the present invention is used to analyze a time-series signal generated by a DUT during operation. The steps include: measuring the time-series signal; frequency-analyzing the time-series signal to generate a set of spectrum. Data; input one of the estimated speed and frequency of the object to be tested; compare the estimated speed and frequency with the spectrum data to output an actual speed. When the method of the invention is applied to a machine tool, the actual rotational frequency of the machine tool spindle can be accurately found. With this analysis method, time series signals can be presented, compared and judged under credible standards. And parts that cause abnormal sounds can be easily identified to solve many existing problems in the industry.

Description

Analysis method of time series signal

The invention relates to a method for analyzing a time series signal, in particular to an analysis method for finding out the actual rotational speed to accurately analyze the source of abnormal sound of a machine tool.

During the processing operation of the machine tool, the processing status is mainly affected by key parameters such as temperature, lubrication, workpiece status and control system. The energy contributed by each parameter also reflects the processing quality of the machine to varying degrees. Because the processing quality or status of the product line is limited by the environment or the signal response is not sensitive enough, the above parameters are not easy to detect immediately. Vibration or noise is a physical phenomenon manifested by the above-mentioned parameters. It is one of the currently recognized more efficient methods to analyze vibration or noise.

The parts inside the machine tool may not run smoothly due to improper installation, looseness, foreign objects or wear. At this time, the machine tool often produces unexpected sounds, called abnormal sounds. In other words, when the machine tool produces a strange sound, it means that the machine tool is not in an optimal state. At this time, the machine tool needs further inspection or maintenance. However, at present, there is no uniform standard for judging abnormal sounds in the use of machine tools. If the technician thinks that the machine tool has a strange sound, it is difficult to accurately predict the source part that produces the strange sound.

In the conventional technology, the industry has been trying to establish a machine tool model, simulate the signal sent by each part, and then compare the detected signals to determine the source of the abnormal sound. However, because the speed of parts is limited by the spindle speed, and the actual spindle speed is often different from the estimated speed output by the controller, The deviation of the rotation speed of the part obtained by the operation of the simulation tool machine is made. The industry believes that the practice of simulating machine tools to infer parts is not reliable, and this method is only theoretically feasible, which has become an urgent problem for the industry.

In view of this, the present invention proposes a method for analyzing time series signals, comparing the theoretical frequency of parts to analyze the source of abnormal sounds, and breaking through the difficulty that the theoretical frequency of parts often does not match the actual frequency, with a view to serving as a A standard method for judging abnormal sounds.

The time-series signal analysis method of the present invention is used to analyze a vibration signal generated by a test object including a plurality of parts. The steps include: measuring vibration signals; acquiring vibration signals regarding a plurality of time data and a plurality of frequency data corresponding to the plurality of time data; and generating a corresponding frequency-frequency according to the time data and the corresponding frequency data. Figure; Analyze the frequency-frequency diagram to generate a characteristic frequency of the part; Simulate the operation of the object under test to obtain a theoretical frequency of each part corresponding to each part; The part corresponding to the characteristic frequency.

In a specific embodiment, the object to be measured is a machine tool, and the machine tool includes a spindle. The step of simulating the operation of the object under test to obtain the theoretical frequency of each part corresponding to each part further includes the following sub-steps: establishing a machine tool model; generating one of the main shafts based on time data and corresponding frequency data Spindle speed; and input the spindle speed to simulate the operation of the object under test to obtain the theoretical frequency of each part.

Among them, in the sub-step of generating the spindle rotation speed based on the time data and the corresponding frequency data, a Fourier transform is performed on the time data and the corresponding frequency data to obtain an amplitude-frequency diagram, and then from the amplitude-frequency diagram Retrieve the spindle speed.

Among them, the step of measuring the vibration signal is a comprehensive sound signal generated when the part of the object to be measured is actually operated.

In addition, the step of comparing the theoretical frequency of the part with the characteristic frequency of the part to confirm the part corresponding to the characteristic frequency of the part is to find a theoretical frequency of the part that matches the characteristic frequency of the part from the theoretical frequency of the part, and then return from the theoretical frequency of the part. Derive the part characteristic frequency.

In a specific embodiment, the object to be measured is a vehicle, and the analysis method of the time series signal is used to detect or monitor the vibration signal generated by the vehicle.

Further, the method for analyzing time series signals of the present invention is used to analyze a time series signal generated by a DUT during operation, and the steps include: measuring the time series signal; performing frequency analysis on the time series signal to generate A set of spectrum data; input one of the estimated rotational speed frequency of the object to be tested; compare the estimated rotational frequency with the spectral data to output an actual rotational frequency.

Further, the step of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency further includes the following sub-steps: acquiring a characteristic frequency within a predetermined range in the spectral data according to the estimated rotational frequency; The characteristic frequency and the frequency spectrum data; if the characteristic frequency meets one of the frequency correspondence rules of the estimated rotational frequency, the characteristic frequency is regarded as the actual rotational frequency output.

The frequency correspondence rule means that the characteristic frequency has a corresponding primary frequency, one side frequency, or one octave in the spectrum data.

Further, the sub-step of extracting the characteristic frequency within a predetermined range in the spectrum data according to the estimated rotational frequency further includes the following steps: extracting a predetermined range corresponding to the estimated rotational frequency from the spectral data; Active frequency Rate; output active frequency as characteristic frequency.

Furthermore, in the sub-step of extracting a characteristic frequency in a predetermined range in the spectrum data according to the estimated rotational frequency, a peak frequency is extracted from the predetermined range, and the characteristic frequency is calculated from the peak frequency.

In a specific embodiment, the step of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency further includes a sub-step: if each characteristic frequency in the predetermined range does not meet the frequency correspondence rule of the estimated rotational frequency , Then the actual rotational frequency is obtained from the spectrum data calculation according to the frequency correspondence rule.

Further, the frequency spectrum data includes a plurality of frequency values and a plurality of amplitude values corresponding to the frequency values. In the sub-step of acquiring a characteristic frequency within a predetermined range in the spectrum data according to the estimated rotational frequency, a frequency value having an amplitude value greater than a noise level amplitude in the predetermined range of the spectrum data is acquired as the characteristic frequency. .

The predetermined range is between 85% of the frequency value of the estimated speed frequency and 115% of the frequency value of the estimated speed frequency. The noise value amplitude is obtained by calculating and calculating the root mean square value of the signal from the amplitude value corresponding to each frequency value in the spectrum data.

In a specific embodiment, the step of performing frequency analysis on the time-series signal to generate the set of spectrum data further includes a sub-step: using a method of fine-tuning the rotation frequency to improve one of the spectrum data resolutions. The fine-tuning speed-frequency method is an overlap-add-convolution method (zero padding) or a Chirp z-transform method.

In a specific embodiment, the object to be measured is a machine tool. The machine tool includes a main shaft and a plurality of parts. The time series signal is generated by the main shaft and the parts. The estimated speed and frequency are estimated by one of the main shafts. Spindle speed frequency, the actual speed frequency is one of the actual spindle speed The analysis method of the international spindle speed and frequency and time series signal further includes the following steps: establishing a machine tool model; inputting the actual spindle speed and frequency to simulate the operation of the machine tool; and calculating a theoretical frequency of each part.

In summary, in order to achieve accurate simulation of the operation of the object under test to obtain the theoretical frequency of each part corresponding to each part, the present invention proposes a method for calculating the actual spindle speed. In this method, the time series signals are subjected to data analysis or visual analysis to obtain spectrum data. The input estimated rotational frequency needs to be compared with the frequency spectrum data, and the physical frequency rule is used to estimate and check the calculation to output the accurate actual rotational frequency. Therefore, the actual rotational frequency found by the method of the present invention is substituted into the model of the object to be tested, and the parts emitting time series signals can be found. When the invention is applied to the analysis of abnormal sound signals, the judgment can no longer rely on manpower and experience, but a systematic and efficient establishment of an abnormal sound judgment standard can solve the quality control, delivery, detection, and monitoring in this field. , Maintenance and other multiple knowledge issues.

S1 ~ S10‧‧‧step

S51 ~ S53‧‧‧‧Sub-step

S521 ~ S526‧‧‧ steps

Areas A, B, C‧‧‧

S91 ~ S93‧‧‧‧Sub-step

S911 ~ S913‧‧‧ steps

FIG. 1 is a flowchart illustrating a specific embodiment of a method for analyzing time series signals according to the present invention.

FIG. 2 is a flowchart illustrating steps for simulating the operation of the object under test to obtain the theoretical frequency of each part corresponding to each part in a specific embodiment of a method for analyzing time series signals according to the present invention.

FIG. 3 is a flowchart illustrating a method for acquiring a characteristic frequency within a predetermined range of spectrum data according to an estimated rotational frequency in a specific embodiment of a method for analyzing a time series signal according to the present invention.

FIG. 4 is an amplitude-frequency diagram of a specific embodiment of a method for analyzing a time-series signal according to the present invention.

FIG. 5 is a diagram illustrating a time-series signal analysis method according to an embodiment of the present invention based on time. The sub-step flowchart of the spindle speed and the corresponding frequency data to generate the spindle speed.

FIG. 6A is a primary spectrum diagram obtained by calculating time data and frequency data through short-time Fourier transform calculation in a specific embodiment of a method for analyzing time series signals according to the present invention.

FIG. 6B is a diagram illustrating a secondary spectrum obtained by performing a fast Fourier transform calculation on a primary spectrum diagram in a specific embodiment of a method for analyzing a time series signal according to the present invention.

FIG. 6C is an amplitude-frequency diagram obtained through integral calculation of a secondary spectrum diagram in a specific embodiment of a method for analyzing a time-series signal according to the present invention.

FIG. 7A is a schematic diagram illustrating the operation of simulating a test object in a specific embodiment of a method for analyzing time series signals according to the present invention.

7B is a schematic diagram showing a theoretical frequency of a part corresponding to each part in a specific embodiment of a method for analyzing time series signals according to the present invention.

FIG. 8 is a flowchart illustrating a specific embodiment of a method for analyzing time series signals according to the present invention.

FIG. 9 is a flow chart of comparing an estimated rotational frequency and spectrum data to output an actual rotational frequency in a specific embodiment of a method for analyzing time series signals according to the present invention.

In order to make the advantages, spirits and features of the present invention easier and clearer, it will be detailed and discussed in the following with reference to the embodiments and the accompanying drawings. It is worth noting that these embodiments are only representative embodiments of the present invention, and the specific methods, devices, conditions, materials, etc. illustrated therein are not intended to limit the present invention or corresponding embodiments.

In the description of this specification, the description with reference to the terms "a specific embodiment", "another specific embodiment" or "partial specific embodiments" and the like means that the specific features, structures, materials, or characteristics described in conjunction with this embodiment include In at least one embodiment of the present invention. In this In the description, the schematic expressions of the above terms do not necessarily refer to the same embodiment. Moreover, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments.

In the description of this specification, the sequence of each step will be described in words, and the sequence of the steps will be indicated by an arrow in the diagram. It should be noted that the numbers in the labels cannot be interpreted as the sequence of steps in the method of the present invention. For example, step S5 is not necessarily after step S4; or step S10 is before step S7.

See Figure 1. FIG. 1 is a flowchart illustrating a specific embodiment of a method for analyzing time series signals according to the present invention. The time-series signal analysis method of the present invention is used to analyze a vibration signal generated by a test object including a plurality of parts. The steps include: measuring the vibration signal S1; extracting the vibration signal with respect to the plurality of time data and the plurality of frequency data S2 corresponding to the time data; and generating a corresponding frequency-frequency diagram according to the time data and the corresponding frequency data. S3; analyze the frequency-frequency diagram to generate a characteristic frequency S4 of the part; simulate the operation of the test object to obtain a theoretical frequency S5 of each part corresponding to each part; and compare the theoretical frequency of the part according to the characteristic frequency of the part, thereby Confirm the part S6 corresponding to the characteristic frequency of the part.

In the conventional technique, the vibration analysis method is applied to the detection by observing the change of the vibration caused by the machine in an abnormal state and comparing it with the vibration during normal operation to find out the characteristics of the vibration. Further, the cause of the vibration is estimated. However, it is not easy to measure both abnormal and normal DUT under the same measurement conditions and environment. In addition, the subsequent estimation steps still need human experience to achieve, it is difficult to use as a general standard, and the accuracy rate is also limited.

Please refer to FIGS. 1, 7A and 7B. FIG. 7A illustrates a time sequence according to the present invention. Schematic diagram of the simulation of the operation of the DUT in one embodiment of the method for analyzing the signal. 7B is a schematic diagram showing a theoretical frequency of a part corresponding to each part in a specific embodiment of a method for analyzing time series signals according to the present invention. In order to solve the above-mentioned pain points, the present invention establishes a model of the object to be tested in the computer system (as shown in FIG. 7A), and simulates the operation of the object to be tested to obtain the theoretical frequency of each part corresponding to each part (see FIG. 7B As shown). The DUT is usually a machine containing multiple mechanical parts. Parts can be metal or non-metallic machine elements such as gears, belts, bearings or chains.

The method of the invention can find the characteristic frequency of the part from the measured vibration signal, and then compare the characteristic frequency of the part with the theoretical frequency of the part. It is also possible to perform the analysis method to step S3 or step S4 in advance for the vibration information generated by the normal operation test object to generate a visual standard vibration message. This analysis method is then performed from the target test object and compared with the analysis results of the normal test object to find the difference between the two. This difference can be regarded as the characteristic frequency of the part. Therefore, the present invention can standardly determine whether the object under test has abnormal sounds different from the theoretical audio; it can also efficiently push back parts that cause abnormal sounds.

Among them, the step S1 of measuring the vibration signal is a comprehensive sound signal generated when measuring the parts of the object under test in actual operation. In addition, in step S6 of comparing the theoretical frequency of the part with the characteristic frequency of the part to confirm the part corresponding to the characteristic frequency of the part, a theoretical frequency of the part corresponding to the characteristic frequency of the part is found from the theoretical frequencies of the parts, and then the theoretical part Frequency recursive parts that give out the characteristic frequency of the part.

In a specific embodiment, steps S1 to S4 are performed sequentially. Step S5 is another procedure executed independently. Finally, in step S6, the characteristic frequency of the part output in step S4 is compared with the theoretical frequency of the part output in step S5. In another specific embodiment, steps S1 to S4 They are executed sequentially. Step S5 is continued after step S2 by another procedure. The data output in step S2 can be used as the calibration data for simulation in step S5, and step S5 outputs a more accurate theoretical frequency of the part, so that the comparison accuracy of step S6 is improved.

Please refer to FIG. 2, FIG. 7A and FIG. 7B. FIG. 2 is a flowchart illustrating step S5 of simulating the operation of the object under test to obtain a theoretical frequency of each part corresponding to each part in a specific embodiment of a method for analyzing time series signals according to the present invention. In a specific embodiment, the object to be measured is a machine tool, and the machine tool includes a spindle. In step S5 of simulating the operation of the object under test to obtain the theoretical frequency of each part, the method further includes the following sub-steps: establishing a machine tool model S51; generating a spindle based on time data and corresponding frequency data One of the main shaft rotation speed S52; and inputting the main shaft rotation speed to simulate the operation of the object under test to obtain the theoretical frequency S53 of each part.

In the machine tool model, the specifications of each part, the number of teeth, the installation method, the installation position and the operating relationship can be further input. After arranging each part, input the spindle speed and simulate the operation of the object under test to calculate the theoretical frequency of each part. Among them, the main shaft pulls the related running parts, so the value of the spindle speed also affects the theoretical frequency of the parts of each related running part.

In some embodiments, the user knows in advance the spindle speed preset by the controller in the test object, and the spindle speed preset by the controller is equal to the actual spindle speed. At this time, the theoretical frequency of each part can be simply calculated.

In some embodiments, the characteristic frequency of the part does not match the theoretical frequency of the part. At this time, it may mean that the spindle speed preset by the controller is not equal to the actual spindle speed, so the theoretical frequency of each related running part is different from the actual frequency. To solve this derivative question The present invention further includes the following solutions.

See Figure 8. FIG. 8 is a flowchart illustrating a specific embodiment of a method for analyzing time series signals according to the present invention. The time-series signal analysis method of the present invention is used to analyze a time-series signal generated by a DUT during operation. The steps include: measuring the time-series signal S10; frequency-analyzing the time-series signal to generate a group Spectral data S7; input one of the estimated speed frequency S8 of the object to be tested; compare the estimated speed frequency with the frequency spectrum data to output an actual speed frequency S9. The time series signal can be a vibration signal, a sound signal, a current signal or a voltage signal.

Please refer to Figure 2 again. In a specific embodiment, in the sub-step S52 of generating the spindle speed of the main shaft according to the time data and the corresponding frequency data, Fourier transform is performed on the time data and the corresponding frequency data to obtain an amplitude-frequency diagram. At this time, the amplitude-frequency diagram can show the actual rotational frequency of the spindle. And, the actual speed frequency is usually slightly lower than the spindle speed preset by the controller. Therefore, the actual spindle speed can be easily extracted from the amplitude-frequency diagram. Then, the actual spindle speed is input in step S53.

In a specific embodiment, the object to be measured is a vehicle, and the analysis method of the time series signal is used to detect or monitor the vibration signal generated by the vehicle. The vehicle includes an engine crankshaft. The function of the engine crankshaft is equivalent to that of the main shaft of the machine tool, and the calculation of the crankshaft speed is equivalent to the main shaft speed of the machine tool. In step S5, it further includes the following sub-steps: establishing a vehicle model; generating a crankshaft speed of the engine crankshaft according to the time data and corresponding frequency data; and inputting the crankshaft speed to simulate the operation of the vehicle engine to obtain The theoretical frequency of parts for each part.

See Figure 9. FIG. 9 shows a comparison of the estimated rotational frequency and spectrum data to output an actual rotational frequency S9 in a specific embodiment of a method for analyzing time series signals according to the present invention. The flowchart. In step S9 of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency, the method further includes the following sub-steps: obtaining a characteristic frequency S91 within a predetermined range in the spectral data according to the estimated rotational frequency; The characteristic frequency corresponds to the frequency spectrum data S92; if the characteristic frequency matches a frequency correspondence rule of the estimated rotational frequency, the characteristic frequency is regarded as the actual rotational frequency output S93. Under certain conditions, the user cannot know the exact value of the spindle speed preset by the controller, but can relatively change the spindle speed, such as issuing a start command, a shutdown command, an acceleration command, or a deceleration command. The active frequency can be judged by observing the amplitude-frequency graph at different time points. When the spindle speed is not the same at different time points, the frequency that changes with the command in the amplitude-frequency diagram may be the active frequency. The active frequency may be issued from the main shaft or a part that is in a linked relationship with the main shaft; in contrast, the passive frequency is a frequency that does not change with the main shaft speed and is issued by a part that is not linked to the main shaft. Therefore, the spindle speed can be further calculated or screened from these active frequencies.

The frequency correspondence rule means that the characteristic frequency has a corresponding primary frequency, one side frequency, or one octave in the spectrum data. For example, when an actual spindle rotation frequency is 40Hz, it can be observed in the spectrum data that 80Hz and 120Hz have a characteristic frequency greater than the noise level amplitude (noise level). This amplitude may be twice and three times the frequency corresponding to the rule. frequency. If it can be observed in the spectrum data that 20Hz and 10Hz also have characteristic frequencies greater than the amplitude of the noise value, this amplitude may be the 1/2 frequency and 1/4 frequency in the frequency correspondence rule.

See Figure 3. FIG. 3 is a flowchart illustrating a method for acquiring a characteristic frequency S91 within a predetermined range in the spectrum data according to the estimated rotational frequency in a specific embodiment of a method for analyzing time series signals according to the present invention. The sub-step S91 of extracting a characteristic frequency in a predetermined range in the spectrum data according to the estimated rotational frequency further includes the following steps: extracting a predetermined range S911 corresponding to the estimated rotational frequency from the spectrum data; Active frequency S912; output the active frequency as the characteristic frequency S913.

Under certain conditions, there may be multiple peaks with similar or fluctuating amplitudes between adjacent (about ± 15%) frequencies of the spindle speed, making it difficult to determine which is the spindle speed. However, these peak frequencies are consistent with the side frequencies in the frequency correspondence rules. At this time, the side frequencies near the spindle speed can be captured for calculation and comparison. In another specific embodiment, in the sub-step S52 of generating the spindle speed of the main shaft according to the time data and the corresponding frequency data, the main shaft speed is calculated from the peak frequency after the peak frequency is extracted from the amplitude-frequency diagram. . At this time, the spindle speed is the average speed ω _{ave of a} plurality of peak frequencies. When ω is the peak frequency value, F (ω ² ) is the corresponding amplitude value of the peak frequency in the frequency spectrum, and the average rotation speed ω _ave is calculated as follows: ω _ave = ʃF (ω ² ) * ω * d ω

In another embodiment, the step of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency further includes a sub-step: if each characteristic frequency in the predetermined range does not correspond to the frequency corresponding to the estimated rotational frequency Rule, the actual rotational frequency is obtained from the spectrum data calculation according to the frequency correspondence rule. It means to recalculate or verify the actual speed frequency according to the rules of active frequency, frequency multiplication, secondary frequency and side frequency.

Further, the frequency spectrum data includes a plurality of frequency values and a plurality of amplitude values corresponding to the frequency values. In the sub-step of acquiring a characteristic frequency within a predetermined range in the spectrum data according to the estimated rotational frequency, a frequency value whose amplitude value is greater than a noise value amplitude in the predetermined range of the spectrum data is acquired as the characteristic frequency.

The predetermined range is between 85% of the frequency value of the estimated speed frequency and 115% of the frequency value of the estimated speed frequency. The noise value amplitude is obtained by calculating and calculating the root mean square value of the signal from the amplitude value corresponding to each frequency value in the spectrum data. The calculation formula of the signal rms value X _RMS is as follows:

Then, the noise-to-noise ratio SNR _{dB is used to} find the frequency greater than the amplitude of the noise value. The noise-to-noise ratio is calculated as follows:

In a specific embodiment, the step of performing frequency analysis on the time-series signal to generate the set of spectrum data further includes a sub-step: using a method of fine-tuning the rotation frequency to improve one of the spectrum data resolutions. In a specific embodiment, due to the limitation of the measurement environment, the measurement time is too short (for example, less than ten seconds), so that there is no clear peak frequency in the relative amplitude-frequency diagram. In the present invention, in the sub-step of generating the spindle speed of the main shaft according to the time data and the corresponding frequency data, the fine-tuning speed-frequency method is used to improve one resolution (frequency resolution) of the frequency data. The fine-tuning speed-frequency method may be an overlap-add-convolution method (zero padding) or a Chirp z-transform method. In a specific embodiment, the vibration signal of less than ten seconds is supplemented to more than ten seconds, and then the Fourier transform is continued. When the original vibration signal exceeds ten seconds, the resolution can reach 0.1Hz, thus improving the frequency resolution.

In a specific embodiment, the object to be measured is a machine tool. The machine tool includes a main shaft and a plurality of parts. The time series signal is generated by the main shaft and the parts. The estimated speed and frequency are estimated by one of the main shafts. Spindle speed frequency, the actual speed frequency is one of the actual spindle speed The analysis method of the international spindle speed and frequency and time series signal further includes the following steps: establishing a machine tool model; inputting the actual spindle speed and frequency to simulate the operation of the machine tool; and calculating a theoretical frequency of each part.

The above methods are all methods for accurately obtaining the spindle speed in the present invention. Each method can be used alone or reasonably combined. The purpose is to make the spindle rotation speed input in step S5 equal to the actual spindle rotation speed in order to output the accurate theoretical frequency of the part. Only after the comparison in step S6 can there be a judgment.

In a specific embodiment, please refer to FIGS. 1, 4, 6A, 6B, and 6C. FIG. 6A is a primary spectrum diagram obtained by calculating time data and frequency data through short-time Fourier transform calculation in a specific embodiment of a method for analyzing time series signals according to the present invention. FIG. 6B is a diagram illustrating a secondary spectrum obtained by performing a fast Fourier transform calculation on a primary spectrum diagram in a specific embodiment of a method for analyzing a time series signal according to the present invention. FIG. 6C is an amplitude-frequency diagram obtained through integral calculation of a secondary spectrum diagram in a specific embodiment of a method for analyzing a time-series signal according to the present invention. In step S3, the time data and the corresponding frequency data are firstly subjected to a short-time Fourier transform to make a spectrogram (FIG. 6A), and then a fast Fourier transform is used to transform the frequency spectrum once. The figure is recalculated into a quadratic spectrum (Figure 6B). At this time, the secondary spectrum diagram can be used to analyze or extract the characteristic frequency of the part; or the secondary spectrum diagram can be further integrated to generate the vibration frequency diagram (Figure 6C), and then the component characteristic frequency can be analyzed or extracted from it.

In the conventional technology, usually after the data is converted into a frequency spectrum by a fast Fourier transform, the integration is started to generate a vibration frequency diagram (Figure 4), and the characteristic frequency of the part is extracted from the analysis. Considering that most of the abnormal sound frequencies are lower than 50Hz, the method of directly performing a Fourier transform on the data is more cluttered and inconspicuous when displaying low-frequency signals (as shown in area B in FIG. 4). So this The advantage of the method of the invention is that the analysis of the low-frequency abnormal sound is more clear and distinguishable, which is conducive to judging the quality of the abnormal sound automatically and programmatically.

Please refer to FIGS. 6A, 6B, 6C and 7B. In this embodiment, compared to the area B in FIG. 4, the area C in FIG. 6C can obviously show a peak frequency at 9.84 Hz. This peak frequency is captured as the characteristic frequency of the part, and a part with a theoretical frequency of 9.84 Hz is found in the model. As shown in FIG. 7B, it was found that the position S1 of the riding elements G0 and G1 would produce the same theoretical frequency of the part. Therefore, it can be inferred that the low-frequency abnormal sound detected in this vibration signal comes from the components G0 and G1 on S1 and S1. If the method of the present invention is not used to find the correct spindle speed, but the preset 1800 rpm is entered in FIG. 7B, the frequency of the position S1 of the load components G0 and G1 will be 10.0 Hz, causing no parts and parts to be found in the model The characteristic frequency matches 9.84Hz.

In the conventional technology, because the rotation speed of a part is limited by the rotation speed of the spindle, and the actual rotation speed of the spindle is usually different from the estimated rotation speed output by the controller, the rotation speed of the part obtained by the simulation of the machine tool operation often has deviations. The industry believes that it is not reliable to simulate machine tools to infer parts, and this method is theoretically feasible but difficult to generalize.

Compared with the conventional technology, in order to achieve accurate simulation of the operation of the test object to obtain the theoretical frequency of each part corresponding to each part, the present invention proposes a method for calculating the actual spindle speed. In this method, the time series signals are subjected to data analysis or visual analysis to obtain spectrum data. The input estimated rotational frequency needs to be compared with the frequency spectrum data, and the physical frequency rule is used to estimate and check the calculation to output the accurate actual rotational frequency. Therefore, the actual rotational frequency found by the method of the present invention is substituted into the model of the object to be tested, and the parts emitting time series signals can be found. When the invention is applied to the analysis of abnormal sound signals, the judgment can no longer rely on manpower and experience, but can be systematic and Efficiently establish a standard for different sound judgments, and solve many problems in this field, such as quality control, delivery, testing, monitoring, and maintenance.

With the above detailed description of the preferred embodiments, it is hoped that the features and spirit of the present invention can be more clearly described, and the scope of the present invention is not limited by the preferred embodiments disclosed above. On the contrary, the intention is to cover various changes and equivalent arrangements within the scope of the patents to be applied for in the present invention. Therefore, the scope of the patent scope of the present invention should be interpreted in the broadest sense according to the above description, so that it covers all possible changes and equal arrangements.

Claims (10)

A time series signal analysis method is used to analyze a time series signal generated by a DUT during operation. The steps include: measuring the time series signal; performing frequency analysis on the time series signal to generate a group Spectrum data; input one of the estimated rotational frequency of the object to be tested; and compare the estimated rotational frequency with the spectral data to output an actual rotational frequency. According to the method of analyzing the time series signal as described in item 1 of the scope of patent application, the step of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency further includes the following sub-steps: The estimated rotational frequency is used to extract a characteristic frequency within a predetermined range in the spectrum data; compare the characteristic frequency with the spectral data; and if the characteristic frequency matches a frequency correspondence rule of the estimated rotational frequency, the characteristic is The frequency is regarded as the actual rotational frequency output. The analysis method of the time-series signal according to item 2 of the scope of patent application, wherein the frequency correspondence rule means that the characteristic frequency has a corresponding primary frequency, one side frequency, or one octave in the spectrum data. The method for analyzing a time series signal as described in item 2 of the scope of patent application, wherein the sub-step of acquiring the characteristic frequency in the predetermined range in the spectrum data according to the estimated rotational frequency further includes the following steps : Extracting the predetermined range before and after the estimated rotational frequency from the spectrum data; separating an active frequency from the predetermined range of the spectrum data; and outputting the active frequency as the characteristic frequency. The method for analyzing a time-series signal according to item 2 of the scope of patent application, wherein in the sub-step of extracting a characteristic frequency within a predetermined range in the spectrum data according to the estimated rotational frequency, it is from the predetermined range. A spike frequency is captured, and the characteristic frequency is calculated from the spike frequency. According to the method of analyzing the time-series signal described in item 2 of the scope of patent application, the step of comparing the estimated rotational frequency with the spectrum data to obtain the actual rotational frequency further includes a sub-step: if the predetermined range Each of the characteristic frequencies does not conform to the frequency corresponding rule of the estimated rotational frequency, and then the actual rotational frequency is obtained by calculating from the frequency spectrum data according to the frequency corresponding rule. The analysis method of the time-series signal as described in the second item of the patent application scope, wherein the frequency spectrum data includes a plurality of frequency values and a plurality of amplitude values corresponding to the frequency values, respectively. In the sub-step of acquiring the characteristic frequency in the predetermined range in the spectrum data, the frequency value in which the amplitude value in the predetermined range of the spectrum data is greater than a noise level amplitude is used as the characteristic frequency. . The method for analyzing a time-series signal as described in item 7 of the patent application range, wherein the predetermined range is between 85% of the frequency value of the estimated speed frequency and 115% of the frequency value of the estimated speed frequency And the noise value amplitude is calculated from the amplitude value corresponding to each of the frequency values in the spectrum data. According to the method for analyzing time series signals as described in item 1 of the scope of patent application, the step of performing frequency analysis on the time series signals to generate the set of spectrum data further includes a sub-step: fine-tuning the speed-frequency method to improve One of the resolutions of the spectral data; wherein the fine-tuning speed-frequency method is an overlap-and-add method (zero padding) or a Chirp z-transform method. According to the method of analyzing the time-series signals described in item 1 of the scope of patent application, wherein the object to be measured is a machine tool, the machine tool includes a spindle and a plurality of parts, and the time-series signals are the spindle and the parts As a result, the estimated rotation speed frequency is an estimated rotation speed frequency of one of the spindles, the actual rotation speed frequency is an actual rotation speed frequency of one of the spindles, and the analysis method of the time series signal further includes the following steps: Machine tool model; input the actual spindle speed frequency to simulate the operation of the machine tool; and calculate the theoretical frequency of one part for each part.