JP2013010110A - Method for detecting chattering of cold rolling mill - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect chattering vibrations generated only by a rolling state by distinguishing between chattering caused by a mechanical state and chattering caused by the rolling state.SOLUTION: The method for detecting the chattering includes the step of analyzing the shape of a FFT transformed frequency wave form by means of a pattern recognition among mill vibrations generated during rolling in a tandem rolling mill which performs cold rolling, when the shape of the analyzed frequency wave form is determined to be a predetermined triangle shape whose frequency component which becomes a peak value of frequency intensity is a vertex, it is detected as the chattering vibrations generated caused by the rolling state (slip, stick).

Description

  The present invention relates to a chattering detection method for a cold rolling mill.

  There is a case where a plate thickness variation called chattering occurs due to mill vibration during rolling of a steel plate. Chattering is caused by the rolling roll chock or bearing deterioration, which causes cracks in the rotating part, which causes natural vibrations of the chock or bearing (hereinafter also referred to as “mechanical state”), and friction that is the rolling condition. Two types when longitudinal vibration caused by slip (or stick) in which the coefficient or steel plate tension is out of the stable region and the neutral point moves out of the roll bite (hereinafter also referred to as “rolling state”) There is. When chattering due to the rolling state occurs, it causes a product defect in which a striped pattern is formed on the steel plate. Therefore, it is important to reliably detect chattering caused by the rolling state and prevent product defects.

  Here, in the prior art, a method for detecting chattering only when a chattering-specific waveform component exists in both the waveform of the sound generated during mill vibration and the vibration waveform detected by the vibration sensor (see Patent Document 1). Or a method of measuring the roll peripheral speed of the work roll and the plate speed of the steel plate, detecting slip to detect chattering (see Patent Document 2), or two or more plate thicknesses in the longitudinal direction of the steel plate being rolled. A method has been proposed in which a meter is installed and detected as chattering when the difference in thickness measured by each of the meters exceeds a preset value (see Patent Document 3). Further, a method has been proposed in which chattering is detected by attaching a vibration sensor to a mill housing or the like and performing processing after performing FFT conversion on the signal (see Patent Documents 4 to 9).

JP 2000-158044 A JP-A-8-24922 Japanese Patent Publication No. 5-87325 Japanese Examined Patent Publication No. 4-46650 Japanese Patent Publication No. 6-3504 JP-A-8-141612 Japanese Patent Publication No.54-38912 JP-A-8-29250 JP-A-8-108205

However, in the method disclosed in Patent Document 1 for detecting chattering based on the sound waveform and the vibration waveform from the vibration sensor, various sounds generated around the mill such as crane traveling sound and sound generated during the mill vibration are separated. If this is not possible, there is a problem that chattering cannot be detected.
Further, in the method of comparing the roll peripheral speed and the plate speed described in Patent Document 2, chattering cannot be detected when the plate speed meter cannot be used due to fumes generated during rolling. .

  Moreover, the method of installing two or more thickness gauges described in Patent Document 3 has a problem that the equipment cost increases by installing a plurality of thickness gauges. In addition, the period of the plate thickness variation that occurs due to chattering due to the rolling state is about 100 to 250 Hz, which is a high frequency compared with the response speed of the plate thickness meter. Chattering is difficult to detect because it is difficult to measure.

In addition, in the method of attaching a vibration sensor to a mill housing or the like, when attaching a filter that allows only the natural frequency of a rolling mill to pass as in the techniques described in Patent Documents 4 to 7, the frequency band used as the frequency component of chattering In addition, if a frequency component such as vibration of the rolling drive system is included, there is a problem that false detection occurs as chattering.
Also, as in the technique described in Patent Document 8, when a two-stage roll is provided in the immediate vicinity of the mill and a vibration sensor is attached to the two-stage roll and the vibration is measured, a new two-stage roll must be installed. In other words, there are problems of securing the space of the installation place and cost increase.

  In addition, as in the technique described in Patent Document 9, the fundamental frequency is calculated based on the rolling state, and the actual vibration is subjected to FFT conversion. When the frequency component that is an integral multiple of the fundamental frequency exceeds the set value, chattering occurs. In such a case, there are a problem that chattering that occurs at a frequency other than the obtained fundamental frequency is missed, and a problem that vibration that does not lead to chattering is erroneously detected.

  Therefore, the present invention has been made paying attention to such problems, and is not affected by various sounds generated around the mill, and does not require the installation of a plate speedometer or a filter. There is no need to secure a special space for installation, and while the equipment cost is relatively low, among the mill vibrations that occur during rolling in tandem rolling mills that perform cold rolling, due to the rolling state It is an object of the present invention to provide a chattering detection method for a cold rolling mill that can detect only chattering vibrations that occur.

When analyzing the mill vibration signal, the inventor has a difference in characteristics in frequency waveforms obtained by FFT conversion between vibration caused by a mechanical state such as chock and bearing deterioration and vibration caused by a rolling state such as slip. Focused on that.
That is, as the difference in the characteristics of the frequency waveform, in the case of chattering caused by the machine state, as shown in FIG. 1, the peak value of the frequency intensity has a shape in which a specific frequency component protrudes. On the other hand, in the case of the rolling state, as shown in FIG. 2, it has a triangular shape with the frequency component serving as the peak value of the frequency intensity as a vertex. The reason for this is that, in the case of the mechanical state, cracks and scratches due to deterioration of the chock and the bearing generate natural vibrations due to rotation, so that the frequency does not spread. On the other hand, in the case of the rolling state, the slip or stick causes a phenomenon in which the slip and stick occur continuously, thereby causing the natural vibration of the mill, resulting in nonlinear vibration, It is thought to be a triangular waveform with a spread. That is, mill vibration that causes chattering can be distinguished from other vibrations by paying attention to the shape of the frequency waveform that has been subjected to FFT conversion.

  From these facts, if the difference between the shape of the frequency waveform obtained by FFT-transforming mill vibration and the shape of the frequency waveform obtained by FFT-transforming mill vibration can be expressed quantitatively, chattering caused by the rolling state is caused. Mill vibration can be identified. Here, a pattern recognition method is suitable as a method for quantitatively expressing the difference between each figure, and in the present invention, this pattern recognition method is used to discriminate mill vibration that causes chattering due to the rolling state. Is what you do.

  That is, in order to solve the above-mentioned problem, the present invention analyzes the shape of the frequency waveform subjected to FFT transformation among mill vibrations generated during rolling in a tandem rolling mill that performs cold rolling, and uses a pattern recognition technique. Chattering that occurs due to the rolling state when it is determined that the shape of the analyzed frequency waveform is in a range corresponding to a predetermined triangular shape having a peak at the frequency component that is the peak value of the frequency intensity. It is characterized by detecting as vibrations.

  Here, in the chattering detection method of the cold rolling mill according to the present invention, the pattern recognition method includes a step of extracting a waveform having a frequency intensity exceeding a predetermined threshold as a waveform having a possibility of chattering, and the extraction thereof. And a step of performing chattering discrimination using a pattern recognition technique based on the Mahalanobis distance with respect to the frequency waveform, and further, the step of performing chattering discrimination is divided into predetermined frequencies for the frequency at which chattering occurs. Of the maximum values for each frequency, a step of extracting a point higher than a predetermined threshold as a candidate point, and the extracted candidate points that are maximum within a predetermined frequency centered on that point , Setting the candidate point as a vertex of a waveform with a possibility of chattering, and increasing the strength of the set vertex A step of setting a triangle for determination having a constant frequency width as a base length as the predetermined triangle shape, and a pattern recognition applying a Mahalanobis distance to the set triangle and the waveform having the possibility of chattering. Including a step of quantitatively determining the difference between the two.

  According to the present invention, the shape of the frequency waveform obtained by FFT-transforming the mill vibration is analyzed by using a pattern recognition technique, and the frequency waveform thus analyzed has a frequency component having a peak value of the frequency intensity as a vertex. Since it is detected as chattering vibration that occurs due to the rolling state when it is determined that it is within the range corresponding to the predetermined triangular shape, only mill vibration that causes chattering due to the rolling state is discriminated. Can do.

  Therefore, it is not affected by various sounds generated around the mill, and there is no need to install a plate speedometer or filter, and it is not necessary to secure a special space for installation. It is possible to detect only chattering vibrations that occur due to the rolling state among mill vibrations that occur during rolling in tandem rolling mills that perform cold rolling. Since it is also possible to detect mill vibration due to such a low strength rolling state, there is an effect that it is possible to take measures to prevent chattering.

It is a figure which shows the image of the frequency waveform of chattering resulting from a machine state. It is a figure which shows the image of the frequency waveform of chattering resulting from a rolling state. It is the schematic of the chattering detection system used by this invention. It is a flowchart explaining the chattering detection process which concerns on this invention. It is a figure explaining an example of pattern recognition concerning the present invention. It is a figure explaining an example of pattern recognition concerning the present invention. It is a figure which shows an example of the frequency waveform of chattering resulting from a machine state. It is a figure which shows an example of the frequency waveform of chattering resulting from a rolling state.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings as appropriate.
FIG. 3 shows a chattering detection system. Reference numeral 10 shown in the figure is a rolling mill, and as shown in the figure, the rolling mill 10 includes a mill housing 3 having a rolling mill 1 for rolling a steel plate P. The chattering detection system 20 measures the vibration acceleration by the vibration sensor 4 attached to the mill housing 3 of the rolling mill 10. The measurement cycle of the output of the vibration sensor 4 is about 1500 to 3000 Hz. The vibration acceleration signal obtained by the vibration sensor 4 is input to the FFT converter 6, and the frequency waveform subjected to the FFT conversion by the FFT converter 6 is sent to the arithmetic processor 7. In this example, the FFT converter 6 performs FFT conversion on a signal obtained by vibration acceleration at a cycle of 1 second. Then, the arithmetic processor 7 analyzes the FFT-converted frequency waveform using a pattern recognition method. Various methods such as Mahalanobis distance, neural network, and principal component analysis are known as pattern recognition methods, but the method itself is not particularly limited. In the example of this embodiment, pattern recognition based on Mahalanobis distance is performed. It is an example which performs the chattering detection process using the method. Hereinafter, the chattering detection process will be described in detail.

  When the chattering detection process is executed by the arithmetic processor 7, as shown in FIG. 4, first, the process proceeds to step S1, and a waveform having a frequency intensity exceeding a predetermined threshold is extracted as a waveform with a possibility of chattering. . In subsequent step S2, chattering determination processing is performed on the frequency waveform extracted in step S1 using a pattern recognition method based on the Mahalanobis distance.

Next, details of the chattering determination processing will be described. FIG. 5 shows an example of a frequency waveform (shown by a solid line) K extracted as a possibility of chattering and a triangle (shown by a wavy line) D used in a pattern recognition technique based on the Mahalanobis distance.
When the above-described chattering determination process is executed by the arithmetic processor 7, as shown in FIG. 4B, first, the process proceeds to step S11, and the frequency at which chattering occurs is 50 to 300 Hz. Thus, a point higher than the threshold value T is extracted as a candidate point from the maximum value every 10 Hz. As a result, candidate points A, B, and C of the vertices of the waveform having the possibility of chattering are extracted. In the subsequent step S12, if the candidate points A, B, and C extracted in step S11 are the maximum within ± 20 Hz with the point as the center, they are set as the vertices of a waveform that may be chattered. Therefore, in this example, candidate point A is set as the vertex. In subsequent step S13, a discrimination triangle D (having hypotenuses of D1 and D2) having the height of the set vertex A and a base length of 50 Hz is set as a discrimination waveform. Finally, the process proceeds to step S14 to execute pattern recognition processing. In this pattern recognition process, the difference in shape is quantitatively obtained by the pattern recognition process in which the Mahalanobis distance is applied to the triangle D set in step S13 and the waveform K having the possibility of chattering.

  A pattern recognition process using the Mahalanobis distance will be described with reference to FIG. When this pattern recognition process is executed, as shown in FIG. 4 (c), the process first proceeds to step S21, where the frequency of the vertex A of the waveform K having the possibility of chattering shown in FIG. A predetermined range (number of points) apart from the apex frequency α is determined. Here, for example, a set A1 of 8 ranges of the waveform K from (α−15) Hz to (α−8) Hz (FIG. 6B). Reference) is extracted as the processing target set A1. In the subsequent step S22, the average, variance, and covariance of the x component (frequency) and y component (frequency intensity) of the processing target set A1 extracted in step S21 are obtained. In step S23, using the average, variance, and covariance obtained in step S22, each point (triangle H1) of the triangle D (hypotenuse D1) corresponding to the frequency (α-15) Hz to (α-8) Hz ( The Mahalanobis distance is calculated for a1i (reference symbol Gm in FIG. 6), 1 ≦ i ≦ 8, and the total is calculated. This is shown by (Formula 1) below.

Where μ1x: average of x components of the processing target set A1;
μ1y: average y component of the processing target set A1,
σ1x: variance of the x component of the processing target set A1,
σ1y: variance of the y component of the processing target set A1,
σ1xy: covariance of x component and y component of the processing target set A1;
a1ix: x component of a1i (symbol Ga in FIG. 6),
a1iy: y component of a1i (symbol R in FIG. 6),

In subsequent step S24, the presence or absence of another processing target set is confirmed. If there is a next processing target set, the process returns to step S22, and if not, the process proceeds to step S25. That is, the above calculation is performed from the processing target set A2 in the range from (α-14) Hz to (α-7) Hz to the processing target set A5 in the range from (α-11) Hz to (α-4) Hz. Then, processing is sequentially performed from the processing target set A6 in the range from (α + 4) Hz to (α + 11) Hz to the processing target set A10 in the range from (α + 8) Hz to (α + 15) Hz.
In step S25, the average of these values obtained as described above is calculated as the correlation value D. This is expressed by the following (Expression 2).

  In the subsequent step S26, the correlation value D of the waveform K having the possibility of chattering calculated by (Equation 2) is compared with a preset discrimination value, and the correlation value D is less than the predetermined discrimination value. If it moves to step S27, it will determine with a rolling state origin and will return a process. That is, when it is determined that the shape of the frequency waveform K to be analyzed is in a range corresponding to a predetermined triangle D having a peak at the frequency component that is the peak value of the frequency intensity, It is detected as chattering vibrations caused by this. On the other hand, if the correlation value D is equal to or greater than the predetermined determination value, the process proceeds to step S28, where it is determined that the machine state is caused and the process is returned.

  Here, as shown in Table 1, the predetermined discriminant value to be compared with the correlation value D is obtained by collecting data according to a pattern recognition method to which the Mahalanobis distance is applied in advance, and the correlation value obtained from this data is actually It was determined based on the result of judging whether the chattering content was due to the rolling state or the mechanical state.

  That is, in the data shown in Table 1, according to the actual confirmation result, No. Nos. 1 to 10 were chattering due to the rolling state. 11 to 22 were chattering caused by the machine state. From this result, since the correlation value D is 31.0 (No. 18) in chattering caused by the machine state, the correlation value D is compared with the correlation value D by setting the predetermined discrimination value to 30 in this embodiment. It was.

  In this embodiment, pattern recognition processing using the Mahalanobis distance is performed on the waveforms of FIGS. 7 and 8 that can be extracted as a frequency waveform K with the possibility of chattering. As a result, the correlation value D = 11.7 (<30) in the chattering waveform of FIG. 8, whereas the correlation value D = 84.7 (> 30) in the overdetection waveform of FIG. It was possible to detect only chattering vibration caused by the rolling state by distinguishing between chattering and chattering caused by the rolling state.

DESCRIPTION OF SYMBOLS 1 Rolling mill 2 Rolling material 3 Mill housing 4 Vibration sensor 5 Vibration amplifier 6 FFT converter 7 Arithmetic processor P Steel plate

Claims (2)

  Of the mill vibrations that occur during rolling in a tandem rolling mill that performs cold rolling, the shape of the frequency waveform that has been subjected to FFT transformation is analyzed using a pattern recognition technique, and the shape of the analyzed frequency waveform is the peak value of the frequency intensity. A cold rolling mill characterized in that it is detected as chattering vibration generated due to a rolling state when it is determined that the frequency component is in a range corresponding to a predetermined triangular shape having a frequency component as a vertex. Chattering detection method. The pattern recognition method uses a step of extracting a waveform whose frequency intensity exceeds a predetermined threshold as a waveform having a possibility of chattering, and a pattern recognition method based on the Mahalanobis distance for the extracted frequency waveform. And a step of performing chattering discrimination,
Furthermore, the step of performing the chattering determination includes a step of extracting a point higher than a predetermined threshold as a candidate point out of the maximum values for each frequency that are predetermined for the frequency at which chattering occurs, and the extraction thereof. If the candidate point is the maximum within a frequency of a predetermined width centered on that point, the step of setting the candidate point as a vertex of a waveform that may be chattered, and the strength of the set vertex as height And a step of setting a discrimination triangle having a predetermined frequency width as a base length as the predetermined triangle shape, and pattern recognition applying a Mahalanobis distance to the set triangle and the waveform having the possibility of chattering The method for detecting chattering of a cold rolling mill according to claim 1, further comprising the step of quantitatively determining the difference in shape.

Images