A method for extracting time-frequency features of kneading machine reducer fault

Abstract

The present invention discloses a method for extracting time-frequency features of a kneading machine reducer fault, comprising: obtaining the original vibration signal of the kneading machine reducer, sliding the original vibration signal with a first sliding window, calculating the signal spectrum in the first window at each sliding position, and constructing a time-frequency distribution matrix of the original vibration signal; taking the data of each row of the time-frequency domain distribution matrix as a sub-signal, and performing sliding time-domain synchronous averaging processing on the sub-signal, comprising: first performing a mean removal processing on the sub-signal, and then performing a time-domain synchronous averaging calculation through a second sliding window; performing an autocorrelation operation on each sub-signal in the time-frequency distribution matrix after the sliding time-domain synchronous averaging processing, and determining the frequency of the fault pulse based on the time-frequency distribution matrix after the autocorrelation operation. The method of the present invention can significantly improve the accuracy and efficiency of vibration signal processing, and is particularly suitable for complex and changeable signal environments.

Description

Technical Field

[0001] The invention relates to the field of signal processing, and in particular to a method for extracting time-frequency features of a kneading machine reducer fault. Background Art

[0002] In the diagnosis of weak faults in the kneading machine reducer system, fault feature extraction is of vital importance. Effective fault feature extraction is the most critical part of the diagnosis process. Its core purpose is to find feature expressions that can accurately reflect the health status and fault type of the system from complex signals. At a constant speed, the frequency components of the fault components of the kneading machine reducer are relatively stable, showing obvious ring-stationary characteristics. These ring-stationary components show obvious differences due to different fault types, mainly reflected in characteristics such as amplitude and frequency. When the fault has not yet become prominent and precursors appear, the energy of the fault component is weak, and it is difficult to accurately extract it under a strong noise background. However, it is critical to detect the precursors of the fault in a timely manner, track the development of the fault, and take corresponding treatment measures, which can improve the operational stability of the system. Summary of the invention

[0003] The purpose of the present invention is to provide a time-frequency feature extraction method for a kneading machine reducer fault, which significantly improves the accuracy and efficiency of vibration signal processing by combining short-time Fourier transform, sliding time domain synchronous averaging and autocorrelation operations, and is particularly suitable for complex and changeable signal environments.

[0004] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0005] A method for extracting time-frequency features of a kneading machine reducer fault comprises the following steps:

[0006] Acquire the original vibration signal of the kneading machine reducer, slide the original vibration signal using the first sliding window, calculate the signal spectrum in the first window at each sliding position, and construct a time-frequency distribution matrix of the original vibration signal;

[0007] The data of each row of the time-frequency distribution matrix is ​​taken as a sub-signal, and sliding time-domain synchronous averaging processing is performed on the sub-signal, including: firstly removing the mean value of the sub-signal, and then performing time-domain synchronous averaging calculation through a second sliding window on this basis;

[0008] An autocorrelation operation is performed on each sub-signal in the time-frequency distribution matrix after the sliding time-domain synchronous averaging process, and the frequency of the fault pulse is determined based on the time-frequency distribution matrix after the autocorrelation operation.

[0009] Furthermore, the original vibration signal of the kneading machine reducer is collected by using a sensor, and the collected original vibration signal is represented in the form of a one-dimensional array x[t], where t is a time parameter;

[0010] The original vibration signal in the first sliding window is subjected to a short-time Fourier transform and the frequency of the original vibration signal is calculated, so as to convert the original vibration signal into a time-frequency domain; wherein the formula of the short-time Fourier transform is:

[0011]

[0012] In the above formula, STFT[t,f] represents the original vibration signal after short-time Fourier transform, f represents the frequency of the original vibration signal, is the time index, is the window function, e is a natural constant, j is an imaginary unit, and N is the length of the first sliding window;

[0013] The result of the short-time Fourier transform of each first sliding window is used as a row of data of the time-frequency distribution matrix, so as to construct the time-frequency distribution matrix S1[t,f].

[0014] Further, the performing mean removal processing on the sub-signals includes:

[0015] Extract the sub-signal from the time-frequency distribution matrix S1[t,f], and perform de-meaning processing on it to remove the DC component in the sub-signal, which is expressed as follows:

[0016]

[0017]

[0018] In the above formula, L is the length of the sub-signal in the time-frequency distribution matrix S1[t,f], x i [t] is the i-th data in the sub-signal, represents the mean of the sub-signal, x i [t] represents the result after removing the mean of the i-th data in the sub-signal.

[0019] Further, performing time domain synchronous average calculation through the second sliding window includes:

[0020] A second sliding window is set on each sub-signal, and time domain synchronous average calculation is performed on the data in the second window at each sliding position; specifically as follows:

[0021] L1=k STSA ·N d

[0022] L STSA =L-L1

[0023]

[0024] In the above formula, L1 represents the length of the second sliding window, LSTSA Indicates the sliding length, N d Indicates the number of cycle points of the fault pulse, k STSA represents the number of second sliding windows, j represents the jth second sliding window, L is the length of the sub-signal, x STSA [t] represents the result of time-domain synchronous averaging of sub-signals;

[0025] The time-frequency distribution matrix S1[t,f] is expressed as S2[t,f] after sliding time domain synchronous averaging.

[0026] Furthermore, the autocorrelation function of each sub-signal in S2[t,f] is calculated as follows:

[0027]

[0028] in, represents the autocorrelation function calculated for each sub-signal in S2[t,f], represents the length of the sub-signal in S2[t,f], represents the time delay, f represents the signal frequency, Represents a sub-signal in S2[t,f] using the representation form of time parameter t and signal frequency f; S2[t,f] is represented as S3[t,f] after autocorrelation operation.

[0029] Further, the determining the frequency of the fault pulse based on the time-frequency distribution matrix after the autocorrelation operation includes:

[0030] By comparing the signal spectrum of S3[t,f] and the signal spectrum corresponding to the time-frequency distribution matrix S1[t,f] of the original vibration signal, if a pulse signal is enhanced in the signal spectrum of S3[t,f] compared with the signal spectrum of S1[t,f], it is determined to be a fault pulse.

[0031] Furthermore, the time-frequency distribution matrix after the autocorrelation operation is input into the fault diagnosis model to diagnose the operating state of the kneading machine reducer and determine whether the kneading machine reducer is healthy or what faults exist.

[0032] Furthermore, the method further comprises:

[0033] Perform gamma transformation on the signal spectrum corresponding to the time-frequency distribution matrix after the autocorrelation operation, so that the corresponding signal spectrum has a higher image contrast:

[0034]

[0035] in, is the result of gamma transformation of the neutron signal S3[t,f], lin and h in Respectively represent the lower and upper limits of the signal spectrum input interval, l in ≤R[τ,f] / I max ≤h in ;I max Represents all autocorrelation functions The maximum absolute value of l out and h out They represent the lower and upper limits of the signal spectrum output interval respectively; γ is a preset exponential parameter.

[0036] A terminal device comprises a processor, a memory and a computer program stored in the memory; when the processor executes the computer program, a method for extracting time-frequency features of a kneading machine reducer fault is implemented.

[0037] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, a method for extracting time-frequency features of a kneading machine reducer fault is implemented.

[0038] Compared with the prior art, the present invention has the following technical features:

[0039] The method of the present invention significantly improves the ability to extract time-frequency features of faults; short-time Fourier transform can deeply analyze the signal in the time domain and frequency domain, and separate the fault signal from the noise; the sliding time domain synchronous averaging effectively suppresses noise, enhances the fault pulse characteristics in the signal, and makes fault detection more intuitive. The application of the autocorrelation function further improves the ability to recognize the periodicity and regularity of the signal, greatly reducing the possibility of false alarms. The gamma transform improves the contrast between the fault pulse and the noise in the signal spectrum diagram. In addition, the present invention has a wide range of applicability, not only suitable for kneading machine fault diagnosis, but also can play an important role in the fields of acoustic signal processing. The comprehensive use of these advanced signal processing technologies provides strong support for the efficient operation of equipment and the reduction of maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flow diagram of the extraction method of the present invention;

[0041] Figure 2 is a waveform diagram of an original vibration signal of a kneading machine reducer in one embodiment of the present invention;

[0042] Figure 3 is a signal spectrum diagram after the signal is short-time Fourier transformed in one embodiment of the present invention;

[0043] Figure 4 is a signal spectrum diagram after sliding time domain synchronous averaging in one embodiment of the present invention;

[0044] Figure 5 is a signal spectrum diagram after autocorrelation processing in one embodiment of the present invention;

[0045] Figure 6 It is a signal spectrum diagram after gamma transformation in one embodiment of the present invention. DETAILED DESCRIPTION

[0046] The main design idea of ​​the present invention is that the short-time Fourier transform can provide the time-frequency characteristics of the signal, divide the signal into multiple short time periods, and obtain the time-frequency representation by performing Fourier transform on each time period; the sliding time domain synchronous averaging can average the signal through a sliding window to eliminate random noise. The autocorrelation operation can extract the periodic characteristics of the signal. The autocorrelation function is used to describe the repeatability and periodicity of the signal and can enhance the specific frequency components of the signal; the gamma transform can improve the contrast of the image and make the distinction between fault pulses and noise more obvious; therefore, combining these four methods can more effectively remove the interference components in the noise signal and extract the time-frequency characteristics of the fault components in the vibration signal, which is suitable for a variety of application scenarios such as mechanical vibration signals and acoustic signals.

[0047] See attached Figure 1 The present invention provides a method for extracting time-frequency features of a kneading machine reducer fault, comprising the following steps:

[0048] Step 1, obtaining the original vibration signal of the kneading machine reducer, sliding the first sliding window on the original vibration signal, calculating the signal spectrum in the first window at each sliding position, and constructing the time-frequency distribution matrix of the original vibration signal.

[0049] In this scheme, the original vibration signal of the kneading machine reducer is collected by a sensor, and the collected original vibration signal is represented in the form of a one-dimensional array x[t], where t is a time parameter; the original vibration signal contains noise signals and fault pulses, such as Figure 2 shown.

[0050] A first sliding window with a length of N is slid in the time domain of the original vibration signal, and at each sliding position, the original vibration signal in the first sliding window is short-time Fourier transformed and the frequency of the original vibration signal is calculated to convert the original vibration signal into the time-frequency domain; the time-frequency domain conversion results of all the first sliding windows are integrated into a two-dimensional matrix to obtain the time-frequency distribution matrix S1[t,f] of the original vibration signal.

[0051] Wherein, the formula of the short-time Fourier transform is:

[0052]

[0053] In the above formula, represents the original vibration signal after short-time Fourier transform, f represents the frequency of the original vibration signal, is the time index, is the window function, e is a natural constant, j is an imaginary unit, and N is the length of the first sliding window.

[0054] The result of the short-time Fourier transform of each first sliding window is used as a row of data of the time-frequency distribution matrix to construct the time-frequency distribution matrix S1[t,f]; this matrix can represent the relationship between the time and frequency of the original vibration signal, that is, the representation of the original vibration signal in the time-frequency domain, and can identify the main frequency components and instantaneous change characteristics of the original vibration signal. At this time, the fault pulse signal component is not obviously highlighted.

[0055] Step 2, taking the data of each row of the time-frequency distribution matrix as a sub-signal, and performing sliding time-domain synchronous averaging processing on the sub-signal, including: firstly performing de-averaging processing on the sub-signal, and then performing time-domain synchronous averaging calculation through a second sliding window.

[0056] After this step, non-periodic noise such as white noise will be effectively offset during the averaging process, while periodic signals are retained and enhanced to reduce the impact of random noise.

[0057] Extract the sub-signal from the time-frequency distribution matrix S1[t,f] and perform de-meaning processing on it to remove the DC component in the sub-signal, making the dynamic changes and frequency components of the signal more prominent, thereby improving the identifiability of the signal, as shown below:

[0058]

[0059]

[0060] In the above formula, L is the length of the sub-signal in the time-frequency distribution matrix S1[t,f], x i [t] is the i-th data in the sub-signal, represents the mean of the sub-signal, x i [t] represents the result after removing the mean of the i-th data in the sub-signal.

[0061] After the sub-signals are de-meaned, a second sliding window is set on each sub-signal, and a time domain synchronous averaging (TSA) calculation is performed on the data in the second window at each sliding position; specifically, as follows:

[0062] L1=k STSA ·N d

[0063] L STSA=L-L1

[0064]

[0065] In the above formula, L1 represents the length of the second sliding window, L STSA Indicates the sliding length, N d Indicates the number of cycle points of the fault pulse, k STSA represents the number of second sliding windows, j represents the jth second sliding window, L is the length of the sub-signal, x STSA [t] represents the result of time domain synchronous averaging calculation of the sub-signals.

[0066] like Figure 4 As shown in the figure, the time-frequency distribution matrix S1[t,f] is expressed as S2[t,f] after sliding time domain synchronous averaging. It can be preliminarily observed that the fault pulse signal is enhanced at 4000Hz.

[0067] Step 3: perform an autocorrelation operation on each sub-signal in the time-frequency distribution matrix after the sliding time-domain synchronous averaging process, and determine the frequency of the fault pulse based on the time-frequency distribution matrix after the autocorrelation operation.

[0068] The autocorrelation operation can extract the periodic characteristics of the signal while suppressing the residual noise; the autocorrelation function is calculated for each sub-signal in S2[t,f] as follows:

[0069]

[0070] Among them, R[τ,f] represents the autocorrelation function calculated for each sub-signal in S2[t,f], L2 represents the length of the sub-signal in S2[t,f], τ represents the time delay, f represents the signal frequency, and STFSA[t,f] represents a sub-signal in S2[t,f] using the time parameter t and the signal frequency f. S2[t,f] is represented as S3[t,f] after the autocorrelation operation.

[0071] Select an appropriate lag time τ to extract the main signal components, usually an integer multiple of the sub-signal period can be selected. Figure 5 As shown, it is a signal spectrum diagram of the time-frequency distribution matrix S3[t,f] obtained after autocorrelation analysis in an embodiment of the present invention; the fault pulse can be enhanced through the autocorrelation operation without introducing additional noise.

[0072] By comparing the signal spectrum of S3[t,f] and the signal spectrum corresponding to the time-frequency distribution matrix S1[t,f] of the original vibration signal, the frequency of the fault pulse can be determined; see Appendix Figure 5 , it can be seen that the fault pulse has been significantly strengthened compared with before processing.

[0073] The time-frequency distribution matrix extracted by the method of the present invention can be input into a fault diagnosis model, such as a machine learning model, so as to diagnose the operating status of the kneading machine reducer and determine whether the kneading machine reducer is healthy or what faults exist.

[0074] In order to make the fault pulse signal easier to observe in the signal spectrum diagram, the following operations can be performed on the basis of the above technical solution:

[0075] The signal spectrum diagram corresponding to the time-frequency distribution matrix S3[t,f] after the autocorrelation operation is subjected to a gamma transformation, so that the corresponding signal spectrum diagram has a higher image contrast, which is conducive to the location of the fault pulse.

[0076]

[0077] Gamma transformation can effectively improve the plane contrast of the result; is the result of gamma transformation of the neutron signal S3[t,f], l in and h in Respectively represent the lower and upper limits of the signal spectrum input interval, l in ≤R[τ,f] / I max ≤h in ;I max Represents all autocorrelation functions The maximum absolute value of l out and h out They represent the lower and upper limits of the signal spectrum output range respectively; γ is an exponential parameter. The larger the γ is, the higher the nonlinearity of the gamma transform and the higher the image contrast. If you want to obtain a higher contrast, you can adjust the value to a larger value. Usually, the γ value is set to 4.

[0078] like Figure 6 As shown, by observing the brightness of the color in the spectrum diagram, it can be clearly seen that the fault pulse signal has been enhanced; this enhancement makes the characteristics of the fault pulse more prominent, so that the fault frequency in the signal can be more clearly identified.

[0079] Example:

[0080] In one embodiment of the present invention, the sampling frequency f of the original vibration signal is recorded as s Set it to 30000Hz, the signal-to-noise ratio is -20db, the number of periodic points of the fault pulse is 300, divide the signal into multiple windows, the length of each second sliding window is 4500, and the number of second sliding windows is set to 15.

[0081] Figure 2The figure shows the performance of the combined signal of the noise signal and the fault pulse in the time domain; in this figure, the fault pulse signal is relatively weak and therefore difficult to identify intuitively, which makes us face certain challenges when analyzing the signal.

[0082] exist Figure 3 In the figure, the signal spectrum diagram generated by short-time Fourier transform of the vibration signal is displayed; although the time-frequency domain contains rich information of the signal, the fault pulse component is still not clear enough and difficult to extract accurately.

[0083] exist Figure 4 In the figure, the vibration signal after sliding time domain synchronous averaging shows some improvement. After this processing, the frequency component of the fault pulse signal can be preliminarily observed. Although it is still relatively weak, it has been significantly enhanced compared with the previous signal spectrum.

[0084] Figure 5 In the figure, the autocorrelation function of the signal processed by sliding time domain synchronous averaging is calculated; this step further enhances the component of the fault pulse signal, making it clearer in the signal spectrum diagram.

[0085] Figure 6 By performing gamma transformation on the signal spectrum after sliding time domain synchronous averaging processing, the contrast of the image is improved, and the frequency of the fault pulse signal can be observed more clearly. This series of processing steps effectively improves the identifiability of the signal and provides an important basis for subsequent fault diagnosis.

[0086] After the above processing, the signal spectrum shows that there is an obvious fault frequency at 4000Hz. The obvious enhancement of this frequency shows that the method effectively extracts the pulse signal related to the equipment fault and significantly improves the ability to identify weak fault characteristics. By comparing and analyzing the spectrum diagrams before and after processing, we can clearly see that the noise in the original signal masks the fault pulse, and after short-time Fourier transform, sliding time domain synchronous averaging, autocorrelation function and gamma transform processing, the ring-stationary pulse signal at 4000Hz on the time-frequency plane is significantly enhanced, and the time-frequency characteristics of the fault are clearly revealed. This result proves the effectiveness of the present invention and provides a reliable basis for subsequent fault diagnosis.

[0087] In summary, the time-frequency feature extraction method of a kneading machine reducer fault provided by the present invention can more accurately separate valuable fault information by optimizing the signal processing process; inputting the extracted features into a fault diagnosis model, such as a neural network, can improve the sensitivity and accuracy of fault detection, which will provide important support for equipment maintenance and management, help identify potential problems in advance, reduce the risk of equipment downtime, and ensure the stability of production efficiency.

[0088] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for extracting time-frequency features of a kneading machine reducer fault, characterized in that: The following steps are involved: Acquire the original vibration signal of the kneading machine reducer, slide the original vibration signal using the first sliding window, calculate the signal spectrum in the first window at each sliding position, and construct a time-frequency distribution matrix of the original vibration signal; The data of each row of the time-frequency distribution matrix is ​​taken as a sub-signal, and sliding time-domain synchronous averaging processing is performed on the sub-signal, including: firstly removing the mean value of the sub-signal, and then performing time-domain synchronous averaging calculation through a second sliding window on this basis; Performing an autocorrelation operation on each sub-signal in the time-frequency distribution matrix after the sliding time-domain synchronous averaging process, and determining the frequency of the fault pulse based on the time-frequency distribution matrix after the autocorrelation operation; The original vibration signal of the kneading machine reducer is collected by using a sensor, and the collected original vibration signal is represented in the form of a one-dimensional array x[t], where t is a time parameter; Performing a short-time Fourier transform on the original vibration signal in the first sliding window and calculating the frequency of the original vibration signal, converting the original vibration signal into the time-frequency domain; wherein the formula of the short-time Fourier transform is: In the above formula, represents the original vibration signal after short-time Fourier transform, f represents the frequency of the original vibration signal, is the time index, is the window function, e is a natural constant, j is an imaginary unit, and N is the length of the first sliding window; The result of short-time Fourier transform of each first sliding window is used as a row of data of the time-frequency distribution matrix, so as to construct the time-frequency distribution matrix S1[t,f]; The performing the de-averaging process on the sub-signals comprises: Extract the sub-signal from the time-frequency distribution matrix S1[t,f], and perform de-meaning processing on it to remove the DC component in the sub-signal, which is expressed as follows: In the above formula, L is the length of the sub-signal in the time-frequency distribution matrix S1[t,f], x i [t] is the i-th data in the sub-signal, represents the mean of the sub-signal, x i [t] represents the result after removing the mean of the i-th data in the sub-signal; The performing time domain synchronous average calculation through the second sliding window includes: A second sliding window is set on each sub-signal, and time domain synchronous average calculation is performed on the data in the second window at each sliding position; specifically as follows: L1=k STSA ·N d L STSA =L-L1 In the above formula, L1 represents the length of the second sliding window, Indicates the sliding length, N d Indicates the number of cycle points of the fault pulse, represents the number of second sliding windows, j represents the jth second sliding window, L is the length of the sub-signal, It represents the result of time domain synchronous averaging calculation of sub-signals; The time-frequency distribution matrix S1[t,f] is expressed as S2[t,f] after sliding time domain synchronous averaging.

2. The time-frequency feature extraction method for kneading machine reducer fault according to claim 1 is characterized in that: The autocorrelation function of each sub-signal in S2[t,f] is calculated as follows: in, represents the autocorrelation function calculated for each sub-signal in S2[t,f], represents the length of the sub-signal in S2[t,f], represents the time delay, f represents the signal frequency, Represents a sub-signal in S2[t,f] using the representation form of time parameter t and signal frequency f; S2[t,f] is represented as S3[t,f] after autocorrelation operation.

3. The time-frequency feature extraction method for kneading machine reducer fault according to claim 1 is characterized in that: The method of determining the frequency of the fault pulse based on the time-frequency distribution matrix after the autocorrelation operation includes: By comparing the signal spectrum of S3[t,f] and the signal spectrum corresponding to the time-frequency distribution matrix S1[t,f] of the original vibration signal, if a pulse signal is enhanced in the signal spectrum of S3[t,f] compared with the signal spectrum of S1[t,f], it is determined to be a fault pulse.

4. The method for extracting time-frequency features of a kneading machine reducer fault according to claim 1, characterized in that: The time-frequency distribution matrix after the autocorrelation operation is input into the fault diagnosis model to diagnose the operating state of the kneading machine reducer and determine whether the kneading machine reducer is healthy or what faults exist.

5. The method for extracting time-frequency features of a kneading machine reducer fault according to claim 1, characterized in that: The method further comprises: Perform gamma transformation on the signal spectrum corresponding to the time-frequency distribution matrix after the autocorrelation operation, so that the corresponding signal spectrum has a higher image contrast: in, is the result of gamma transformation of the neutron signal S3[t,f], l in and h in Respectively represent the lower and upper limits of the signal spectrum input interval, l in ≤R[τ,f] / I max ≤h in ;I max Represents all autocorrelation functions The maximum absolute value of l out and h out They represent the lower and upper limits of the signal spectrum output interval respectively; γ is a preset exponential parameter.

6. A terminal device comprising a processor, a memory and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the method for extracting time-frequency features of a kneading machine reducer fault according to any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium, wherein a computer program is stored in the medium; characterized in that: When the computer program is executed by a processor, the method for extracting time-frequency features of a kneading machine reducer fault according to any one of claims 1 to 5 is implemented.

Images