Method for detecting abnormal vibration of rolling mill, abnormality detection device, rolling method, and method for manufacturing metal strip

Abstract

Prevent mis-detection caused by noise and the like generated from peripheral equipment of a rolling mill, and detect abnormal vibration with high accuracy. The abnormal vibration detection method of the rolling mill includes: a collection step of collecting vibration data of the rolling mill; a frequency analysis step of performing frequency analysis of the vibration data to generate first analysis data; a principal component analysis step of performing principal component analysis of the first analysis data using reference data determined in advance based on a normal state as principal components to generate evaluation data that is a projection of the first analysis data to the reference data; and an abnormal vibration detection step of extracting a deviation component from the evaluation data and the first analysis data, and detecting an abnormality of the rolling mill based on the extracted deviation component.

Description

Technical Field

[0001] This invention relates to a method for detecting vibrations generated in a rolling mill that produces steel plates to a specified thickness, and more particularly, to a method, anomaly detection device, rolling method, and method for manufacturing metal strips for detecting abnormal vibrations in a rolling mill that cause defects on the surface of steel plates. Background Technology

[0002] Generally, steel sheets used in automobiles, beverage cans, and other products undergo continuous casting, hot rolling, and cold rolling, followed by annealing and plating processes, and are then processed according to their respective applications. The cold rolling process is the final step in determining the thickness of the steel sheet used in the final product. Since the surface of the steel sheet before plating determines the surface of the final product after plating, the cold rolling process requires measures to prevent surface defects.

[0003] Drill marks are one type of surface defect generated during the cold rolling process. Drill marks are linear markings in the width direction of a metal strip that periodically appear in the length direction, and are believed to be primarily caused by mill vibration (chatter). Furthermore, as one cause of drill marks, it is known that they can be caused by polygonal deformation of the support rolls (see Non-Patent Document 1). Non-Patent Document 1 discloses the following mechanism: when the mill meets specific conditions, self-excited vibration generates stripes in the support rolls with the same width direction as the drill marks; the markings on the support rolls become a new source of vibration, resulting in large vibrations and drill marks on the steel sheet.

[0004] Very slight chatter marks are not detected during visual inspection and thickness measurement after rolling, but are only discovered after the plating process, thus becoming a major factor hindering productivity. In addition, it is known that, especially in thin materials such as can steel plates and electromagnetic steel plates, the rapid changes in plate thickness or tension caused by chatter can lead to plate breakage and other phenomena, hindering production.

[0005] In the past, various chatter detection methods have been developed from the perspective of preventing production obstacles and surface defects (see, for example, Patent Documents 1-3). Patent Document 1 describes a method that uses a vibration detector installed on a rolling mill to measure vibration, performs frequency analysis on the obtained vibration and rolling parameters, and determines chatter based on the signal strength of the frequency that may be generated for each vibration cause.

[0006] In Patent Documents 2 and 3, vibration detectors are not only installed on the main body of the rolling mill, but also between each stand and on the inlet and outlet sides of the cold rolling mill, and on the rolls (small-diameter rolls) on which the metal sheet is wound at a certain angle or higher. Furthermore, the following method is disclosed: frequency analysis of the vibration value obtained by the vibration meter is performed, and if the vibration exceeds a threshold at a frequency that is consistent with the chordal vibration frequency of the steel sheet, it is determined that abnormal vibration has occurred.

[0007] Existing technical documents

[0008] Patent documents

[0009] Patent Document 1: Japanese Patent Application Publication No. 08-108205

[0010] Patent Document 2: Japanese Patent Application Publication No. 2016-153138

[0011] Patent Document 3: Japanese Patent Application Publication No. 2016-2582

[0012] Non-patent literature

[0013] Non-patent document 1: Liu Xiaohong et al., "Polygonization Phenomenon of Steel Making Machinery" Japanese Mechanical Society [No. 01-5] Dynamics and Design Conference 2001 CD-ROM Proceedings [2001.8.6-9, Tokyo] Summary of the Invention

[0014] The problem that the invention aims to solve

[0015] If vibrations caused by the chattering of the support rolls can be detected early, as in Non-Patent Document 1, the generation of chattering in the metal strip can be suppressed. However, in the case of Patent Document 1, noise generated from the peripheral equipment of the mill and vibrations generated from vibration sources located on the main body of the mill are also detected simultaneously, resulting in false detections. In addition, in the cases of Patent Documents 2 and 3, although the generation of vibrations caused by string vibrations can be suppressed, it is difficult to detect vibrations other than those used as vibration sources. In particular, in continuous cold rolling mills, the conveying speed (rolling speed) of the metal strip is different for each stand. As a result, the rotational speed of the work rolls is different for each stand, which makes the detection of chattering due to the overlap of multiple frequencies of vibration more difficult.

[0016] The present invention was made in view of the above-mentioned problems, and aims to provide a method, device, rolling method and method for manufacturing metal strip for detecting abnormal vibrations of rolling mills that cause chatter marks with high precision.

[0017] Methods for solving problems

[0018] [1] An abnormal vibration detection method for a rolling mill is a method for detecting abnormal vibration of a rolling mill having a pair of work rolls and multiple support rolls supporting the work rolls, comprising: a collection step of collecting vibration data of the rolling mill; a frequency analysis step of performing frequency analysis on the vibration data to generate first analysis data; a principal component analysis step of performing principal component analysis on the first analysis data using reference data determined in advance based on a normal state as principal components to generate evaluation data, i.e., the projection of the first analysis data onto the reference data; and an abnormal vibration detection step of extracting deviation components from the evaluation data and the first analysis data, and detecting abnormalities of the rolling mill based on the extracted deviation components.

[0019] [2] According to the abnormal vibration detection method of the rolling mill described in [1], in the principal component analysis step, the principal components extracted as the reference data are set for each rolling speed in the rolling mill.

[0020] [3] According to the abnormal vibration detection method of the rolling mill described in [1], the frequency analysis step generates the vibration intensity of each frequency as the first analysis data, and also includes a data transformation step that transforms the first analysis data into second analysis data representing the vibration intensity of each spacing based on the rolling speed, and the principal component analysis step performs principal component analysis on the second analysis data.

[0021] [4] In the abnormal vibration detection method of the rolling mill according to any one of [1] to [3], in the principal component analysis step, the multiple principal components extracted as the reference data are set such that the cumulative value of the contribution rate of the principal components when principal component analysis is performed on the normal analytical data obtained when rolling is performed using the normal rolling mill is above the reference contribution rate.

[0022] [5] The abnormal vibration detection method of the rolling mill according to any one of [1] to [4], wherein the rolling mill performs cold rolling on the metal strip.

[0023] [6] An anomaly detection device for a rolling mill is a rolling mill having a pair of work rolls and a plurality of support rolls supporting the work rolls, comprising: a data collection unit for collecting vibration data of the rolling mill; a frequency analysis unit for performing frequency analysis of the vibration data to generate first analysis data; a principal component analysis unit for performing principal component analysis on the first analysis data using reference data pre-determined based on a normal state as principal components to generate evaluation data, i.e., a projection of the first analysis data onto the reference data; and an anomaly detection unit for extracting deviation components from the evaluation data and the first analysis data, and detecting anomalies of the rolling mill based on the extracted deviation components.

[0024] [7] A rolling method comprising a support roll replacement step of replacing the support roll of the rolling mill when an abnormality of the rolling mill is detected using the abnormal vibration detection method of the rolling mill described in any of the above [1] to [5].

[0025] [8] A method for manufacturing a metal strip, comprising the step of manufacturing a metal strip using the rolling method described in [7] above.

[0026] Invention Effects

[0027] According to the present invention, abnormal vibrations that cause chatter marks on metal strips are evaluated based on the deviation components of evaluation data generated by principal component analysis. This prevents false detections caused by noise from peripheral equipment of the rolling mill and enables high-precision detection of abnormal vibrations causing chatter marks. As a result, rolling mill operations can be performed with abnormal vibrations prevented or suppressed, and defects on the surface of the metal strip caused by abnormal vibrations can be prevented or suppressed, resulting in the production of metal strips with excellent appearance. Attached Figure Description

[0028] Figure 1 This is a schematic diagram illustrating an example of a rolling mill with an abnormality detection device of the present invention.

[0029] Figure 2 This is a functional block diagram illustrating a preferred embodiment of the abnormality detection device for a rolling mill according to the present invention.

[0030] Figure 3 This is a graph of Invention Example 1, which divides the deviation components of the principal component for each frequency.

[0031] Figure 4 This is a graph of Invention Example 2 showing the deviation components for each standard spacing.

[0032] Figure 5 The graphs show comparative examples 1 and 2 using a threshold relative to the vibration intensity.

[0033] Figure 6 This is a graph of Invention Example 3 showing the deviation components for each standard spacing.

[0034] Figure 7 The graph shows Comparative Example 3, which uses a threshold relative to the vibration intensity.

[0035] Figure 8 This is a functional block diagram illustrating another preferred embodiment of the abnormality detection device for the rolling mill of the present invention.

[0036] Figure 9 This is a functional block diagram illustrating yet another preferred embodiment of the abnormality detection device for the rolling mill of the present invention. Detailed Implementation

[0037] Hereinafter, with reference to the accompanying drawings, we will describe the abnormal vibration detection method, abnormal detection device, rolling method, and metal strip manufacturing method of the present invention. Figure 1 This is a schematic diagram illustrating an example of a rolling mill with an abnormality detection device of the present invention. Figure 1 The rolling mill 1 is a cold rolling mill for cold rolling steel strip, for example, a metal strip S. Four mills 2A, 2B, 2C, and 2D (four stands) are arranged along the rolling direction. Each mill 2A, 2B, 2C, and 2D has a substantially identical structure, including a housing 3, a pair of work rolls 4 housed within the housing 3 for rolling the metal strip S, multiple support rolls 5 supporting the work rolls 4, and a drive device 6 for rotating the work rolls 4. Furthermore, small-diameter rolls 7 for attaching to the rolled metal strip S are respectively provided on the downstream side of each mill 2A, 2B, 2C, and 2D in the rolling direction.

[0038] Vibration meters 8A, 8B, 8C, and 8D are respectively installed in the housing 3 of each of the rolling mills 2A, 2B, 2C, and 2D. Vibration meters 8A, 8B, 8C, and 8D measure the vibrations generated in the rolling mills 2A, 2B, 2C, and 2D, and are constructed, for example, by using accelerometers. It should be noted that the vibration meters 8A, 8B, 8C, and 8D are not limited to the housing 3, as long as they are located in positions capable of detecting the vibrations of the rolling mills 2A, 2B, 2C, and 2D. For example, they can also be installed in the bearing seats of each roll or on the small-diameter rolls 7 for the rolled metal strip S to be attached.

[0039] Specifically, when vibratory meters 8A, 8B, 8C, and 8D are installed on the small-diameter roll 7, the vibration data obtained by the vibratory meters 8A, 8B, 8C, and 8D can be considered to correspond to the vibration of the rolling mills 2A, 2B, 2C, and 2D located upstream of the small-diameter roll 7 on which the vibratory meters 8A, 8B, 8C, and 8D are installed in the rolling direction of the metal strip S. In this embodiment, the rolling speed refers to the circumferential speed of the work rolls 4 in the rolling mills 2A, 2B, 2C, and 2D, or the conveying speed (exit speed) of the metal strip S on the exit side of the rolling mills 2A, 2B, 2C, and 2D. The rolling speed is determined for each of the rolling mills 2A, 2B, 2C, and 2D on which the vibratory meters 8A, 8B, 8C, and 8D are installed (in the following description, the location where the vibratory meters 8A, 8B, 8C, and 8D are installed is sometimes referred to as the stand). When vibration meters 8A, 8B, 8C, and 8D are installed on the small-diameter roll 7, the vibration data obtained by the vibration meters 8A, 8B, 8C, and 8D are correlated with the rolling speeds of the mills 2A, 2B, 2C, and 2D located upstream of them. Furthermore, the standard rolling speed in this embodiment is an arbitrary rolling speed set for each of the mills 2A, 2B, 2C, and 2D. The standard rolling speed can be selected empirically based on the rolling speed of the mills 2A, 2B, 2C, and 2D, which are prone to chattering. For example, as the standard rolling speed for the final stand 2D, 900 m / min can be selected from the speed range of 800 m / min to 1300 m / min, where chattering is prone to occur. In this case, the standard rolling speeds of the mills 2A, 2B, and 2C upstream of the final stand 2D can be set based on the standard rolling speed set for the final stand 2D, according to the standard pass planning.

[0040] Figure 2 This is a functional block diagram illustrating a preferred embodiment of the anomaly detection device for a rolling mill according to the present invention. It should be noted that... Figure 2 The anomaly detection device 10 for the rolling mill is constructed using hardware resources such as a computer. The anomaly detection device 10 detects abnormal vibrations in the rolling mill 2A, 2B, 2C, and 2D that cause chatter marks, and may include a data collection unit 11, a frequency analysis unit 12, a principal component analysis unit 13, and an anomaly detection unit 15. Additionally, the anomaly detection device 10 may also include a data conversion unit 14, which will be described later.

[0041] The data collection unit 11 collects vibration data detected by each vibration meter 8A, 8B, 8C, and 8D. When the vibration meters 8A, 8B, 8C, and 8D are acceleration sensors, they transmit vibration acceleration data to the data collection unit 11. The data collection unit 11 continuously acquires acceleration data. Then, the data collection unit 11 integrates the acceleration data measured within a preset data sampling time (e.g., 0.2 seconds) to convert it into velocity data, and collects this as vibration data for each moment, i.e., each data sampling time. As a result, the vibration data becomes vibration velocities arranged in chronological order.

[0042] Furthermore, the data collection unit 11 performs measurements, for example, 0.2 seconds as the data sampling time, and calculates vibration data at a preset data acquisition cycle (e.g., every 1 second). In continuous cold rolling mills, the data sampling time is preferably set to 0.1 seconds or more and 1 second or less, and the data acquisition cycle is preferably set to 1 second or more and 5 seconds or less. This is because: if the data sampling time is less than 0.1 seconds, it may be impossible to obtain data sufficient to determine the degree of mill vibration; if it exceeds 1 second, the computational load for frequency analysis, etc., may become high, thus avoiding these situations. Additionally, this is because: if the data acquisition cycle is less than 1 second, the computational load for frequency analysis, etc., may become high; if it exceeds 5 seconds, it may become difficult to detect abnormal vibrations early, thus avoiding these situations. It should be noted that in the example shown here, the data collection unit 11 collects vibration data from each of the vibration meters 8A, 8B, 8C, and 8D. However, it can be configured to collect vibration data from any one of the vibration meters 8D (8A, 8B, 8C). This is because, based on the vibration data collected by any one of the vibration meters 8D (8A, 8B, 8C), chatter in the rolling mill (stand) 2D (2A, 2B, 2C) where that vibration meter 8D (8A, 8B, 8C) is installed can be reliably detected. It should also be noted that, instead of acceleration sensors, position sensors or velocity sensors capable of measuring vibration can be used instead of vibration meters 8A, 8B, 8C, and 8D. This is because acceleration, velocity, and displacement data can be transformed into each other through time integration and time differentiation.

[0043] The frequency analysis unit 12 performs frequency analysis on the vibration data collected by the data collection unit 11 during the data sampling time, generating analytical data (hereinafter referred to as the first analytical data) composed of the vibration intensity of each frequency at each data acquisition period. For example, the frequency analysis unit 12 uses Fourier transform to extract the amplitude and phase of the vibration velocity at each frequency, extracting the absolute value of the amplitude of the vibration velocity at each frequency as the vibration intensity. It should be noted that the frequency after the Fourier transform of the digital data becomes a discrete value depending on the number of data points in the Fourier transform and the sampling frequency.

[0044] In this embodiment, multiple frequencies are set for the frequency analysis unit 12 to perform frequency analysis, and these are referred to as reference frequencies. The reference frequencies can be selected arbitrarily from a frequency band less than half the sampling frequency, based on the sampling frequencies of the vibration meters 8A, 8B, 8C, and 8D. The sampling frequency refers to the number of vibrations (e.g., acceleration) measured by the vibration meter in one second, and varies depending on the specifications of the vibration meter used. In this embodiment, the lowest sampling frequency among the multiple vibration meters 8A, 8B, 8C, and 8D can be used as a representative value. Preferably, the reference frequencies are selected from a frequency band less than half the sampling frequency, with 20 to 1600 frequencies. This is because: if there are fewer than 20 reference frequencies, the generation of chatter may not be detected; if there are more than 1600, the data acquisition period needs to be set long to avoid excessive computational load on the frequency analysis unit 12, which may prevent early detection of chatter. Therefore, these situations are avoided. More preferably, the reference frequency is selected from 200 to 800 frequencies within a frequency band less than half the sampling frequency. For example, the frequency analysis unit 12 sets the sampling frequency of the vibration meters 8A, 8B, 8C, and 8D to 5120Hz, and sets the reference frequency every 5Hz (400 times) within a frequency range of 5Hz to 2000Hz, analyzing the vibration intensity for each reference frequency. It should be noted that the frequency analysis unit 12 is not limited to Fourier transform as long as it can analyze the vibration data into the vibration intensity of each frequency; it can use known frequency analysis methods such as wavelet transform and windowed Fourier transform. In this case, the reference frequency can also be set using the same method as described above.

[0045] Principal component analysis unit 13 performs principal component analysis on the first analytical data generated by frequency analysis unit 12 using reference data representing a normal state, generating evaluation data. Evaluation data refers to data obtained by projecting the observed data (in this embodiment, the first analytical data) onto a space composed of principal component vectors. That is, the evaluation data is determined by scalars obtained by projecting the observed data onto the directions of each of the multiple principal component vectors, and consists of information from a number of scalars equal to the number of principal component vectors. It should be noted that when the first analytical data regarding the vibration intensity at each frequency generated by frequency analysis unit 12 is used as is, principal component analysis unit 13 performs principal component analysis on the first analytical data composed of the vibration intensity at each frequency, generating evaluation data calculated as a projection onto a predetermined first principal component to the i-th principal component (reference data). The principal component vectors (reference data) applied to the principal component analysis will be described later. It should be noted that "principal component analysis" is sometimes used to refer to both the analysis of variables called principal components, which synthesize from many correlated variables to best represent the overall deviation from a small number of uncorrelated variables, and the calculation of the projection of observed data onto a space composed of pre-defined principal component vectors. However, the principal component analysis performed by the principal component analysis unit 13 in this embodiment is used to refer to the latter meaning. That is, the principal component analysis unit 13 in this embodiment has the function of calculating the projection (evaluation data) of the first analytical data onto a space composed of pre-defined principal component vectors (reference data) representing a normal state.

[0046] The first to i-th principal components (reference data) used in the principal component analysis performed in the principal component analysis unit 13 are set based on the vibration intensity (reference vibration data) of each frequency obtained under normal conditions when no abnormal vibrations occur in mills 2A, 2B, 2C, and 2D. That is, under normal conditions, relative to the vibration data collected by the data collection unit 11, frequency analysis is performed in the frequency analysis unit 12 to generate reference vibration data representing the vibration intensity of each reference frequency every data acquisition period. For this reference vibration data, principal component analysis is performed in the principal component derivation unit 16 (described later) to generate reference data. It should be noted that the principal component analysis performed in the principal component derivation unit 16 means the analysis of principal component vectors synthesized from many correlated variables to best represent the overall deviation from a small number of uncorrelated variables. Normal conditions when no abnormal vibrations occur in mills 2A, 2B, 2C, and 2D refer to a state where no abnormal vibrations occur in mills 2A, 2B, 2C, and 2D at standard rolling speeds. It should be noted that abnormal vibrations will be described later.

[0047] The reference vibration data is, for example, based on vibration data measured during rolling within 12 hours of the replacement of the support roll 5 with a new one. This reference vibration data, obtained by analyzing normal vibration behavior that does not produce abnormal vibrations, is sometimes referred to as normal analysis data. Alternatively, the reference vibration data can be based on vibration data measured during rolling within 24 hours of the replacement of the support roll 5 with a new one. This is because it is empirically known that abnormal vibrations will not occur for approximately two days after the support roll 5 is replaced with a new one, before the support roll 5 wears into a polygonal shape. The data sampling time for obtaining the reference vibration data is preferably set to the same time as the data sampling time for abnormality detection during operation (after 24 hours since the support roll 5 was replaced with a new one). The data acquisition period can also be set differently for obtaining the reference vibration data and for obtaining the vibration data during operation.

[0048] The reference vibration data is generated by taking the vibration intensity of each frequency acquired during the data sampling time as a dataset, and generating multiple datasets every data acquisition cycle acquired under normal conditions. Therefore, the reference vibration data has multiple datasets. The number of datasets included in the reference vibration data is preferably more than 30,000 and less than 200,000. Using the reference vibration data obtained in this way, in the principal component derivation section 16 described later, the first to the i-th principal components are determined by principal component analysis, which is called the reference data. This principal component analysis determines the principal component vector that best represents the overall deviation with a small number of uncorrelated data based on the reference frequency as the variable and multiple reference vibration data with correlation.

[0049] It should be noted that, hereinafter, the reference data calculated using the vibration intensity of each reference frequency under normal conditions, as determined in this way, is sometimes referred to as the first reference data. Specifically, starting from the contribution rate of the first principal component representing the characteristic quantity of the reference vibration data in the principal component space, the cumulative value is calculated in descending order of contribution rate. Based on the condition that the calculated cumulative contribution rate reaches a preset value, i principal components are selected as reference data. Here, the preset cumulative contribution rate is referred to as the reference contribution rate or the set contribution rate. In this embodiment, the reference contribution rate can be arbitrarily set from a value of 1 (100%) or less based on the actual occurrence of chatter marks. In a typical continuous rolling mill, the reference contribution rate is preferably set to 0.4 (40%) or more and 0.7 (70%) or less, more preferably 0.6 (60%) or more and 0.7 (70%) or less. The reference contribution rate is an indicator that affects the degree (reproducibility) of reproducing the vibration behavior of the reference vibration data in the principal component space. If the reference contribution rate is too large, although the vibration behavior of the reference vibration data can be reproduced with high accuracy in the principal component space, measurement noise and other factors included in the reference vibration data will also be reproduced in the principal component space. On the other hand, if the reference contribution rate is too small, although the influence of measurement noise included in the reference vibration data can be eliminated, there is a tendency for features related to the vibration behavior of the reference vibration data to be lost in the principal component space. The appropriate range of the reference contribution rate depends on the rolling mill and the rolling conditions of the steel plate, but for the purpose of detecting abnormal vibrations in continuous rolling mills, it is preferable to set it within the above range.

[0050] When exporting baseline data, such as Figure 8 As shown, a principal component derivation unit 16 can be provided to derive principal components using reference vibration data (normal analysis data) generated in the frequency analysis unit 12 of the anomaly detection device 10 of the rolling mill. The principal component derivation unit 16 performs analysis to determine the principal component vectors that best represent the overall deviation with a small number of unrelated reference vibration data based on multiple related reference vibration data. The first to i-th principal components obtained by the principal component derivation unit 16 are sent to the principal component analysis unit 13, and the projection (evaluation data) of the analysis data (specifically, the first analysis data and the second analysis data described later) obtained by the principal component analysis unit 13 during operation onto the first to i-th principal components can be calculated. Furthermore, if the frequencies at which chatter marks are easily generated in rolling mills 2A, 2B, 2C, and 2D are known in advance, multiple frequencies of similar intensity to these frequencies can be pre-selected during the derivation of principal components in the principal component derivation unit 16, reducing the number of variables used in the principal component analysis in the principal component analysis unit 13.

[0051] The anomaly detection unit 15 determines the occurrence of abnormal vibration based on the evaluation data generated by the principal component analysis unit 13. Specifically, the anomaly detection unit 15 calculates the difference between the analytical data (specifically, the first analytical data and the second analytical data described later) and the projection (evaluation data) of the analytical data generated by the principal component analysis unit 13 relative to the first principal component to the i-th principal component, and determines this as the deviation component. The reference data represents the characteristic quantity data representing the vibration data under normal conditions, so abnormal vibration occurs in the direction of deviation from the reference data. Furthermore, by monitoring the degree of deviation of the analytical data obtained by the principal component analysis unit 13 during operation from the reference data (the first principal component to the i-th principal component), i.e., the deviation component, abnormal vibration can be determined. The deviation component is sometimes referred to as the Q statistic. Moreover, the anomaly detection unit 15 has a threshold for determining abnormal vibration based on the deviation component; if the deviation component is above the threshold, it is determined that abnormal vibration has occurred. The threshold used in the anomaly detection unit 15 can be set based on past operating performance, based on the actual value of vibration intensity obtained under conditions that do not produce flutter.

[0052] The aforementioned principal component analysis unit 13 performs principal component analysis on the first analytical data representing the vibration intensity at each frequency. On the other hand, abnormal vibrations are often caused by the rotational distance of the rotating body (described later), and chatter marks are sometimes caused by abnormal vibrations due to the rotation of the support roll 5. The frequency of abnormal vibrations corresponding to the rotational motion of the equipment constituting the rolling mills 2A, 2B, 2C, and 2D varies with the rolling speed. Therefore, it is preferable that the principal component analysis unit 13 calculates the projection of the analytical data extracted as evaluation data onto the first to the i-th principal components for each rolling speed. Furthermore, it is preferable that, under normal conditions where no abnormal vibrations occur in the rolling mills 2A, 2B, 2C, and 2D, reference vibration data is obtained for each rolling speed, and reference data is generated for each rolling speed using the principal component derivation unit 16. Therefore, in the anomaly detection unit 15, it is easy to clearly identify the difference between normal and abnormal conditions for each rolling speed, improving the accuracy of anomaly detection. The evaluation data is preferably differentiated based on the highest speed of rolling mills 2A, 2B, 2C, and 2D, and the rolling speed is divided into levels 5 and above but below 20, and evaluation data is generated for each rolling speed range.

[0053] On the other hand, even when rolling speeds differ, it is preferable to evaluate the occurrence of anomalies using the same index. Therefore, the anomaly detection device 10 for the rolling mill may also include a data transformation unit 14 that performs data transformation based on the rolling speed, converting frequency into spacing and transforming first analytical data into vibration intensity (second analytical data) for each spacing. For each of the rolling mills 2A, 2B, 2C, and 2D equipped with vibration meters 8A, 8B, 8C, and 8D, the data transformation unit 14 transforms the first analytical data of vibration intensity corresponding to a reference frequency into second analytical data representing the vibration intensity of each spacing (data transformation step). Here, the spacing in this embodiment is an index corresponding to the vibration frequency and corresponding to the distance along the length of the metal strip S or the circumferential distance of the work rolls 4 of the rolling mills 2A, 2B, 2C, and 2D. In other words, the spacing means the interval between vibration peaks that become adjacent to each other along the length of the metal strip S and the circumferential direction of the work rolls 4 as a result of the data transformation described above in the data transformation unit 14. Specifically, the spacing P (mm) is related to the rolling speed V (m / min) and the vibration frequency f (Hz) by the following formula.

[0054] P=(1000×V) / (f×60)…(1)

[0055] In the data conversion unit 14, a standard spacing is stored as a spacing corresponding to the standard rolling speed. The standard spacing refers to the spacing calculated according to the above formula (1) based on the reference frequency f and the standard rolling speed V performed by the frequency analysis unit 12. The standard spacing thus set is a series of discrete numerical strings corresponding to the reference frequency. The reason for using the standard spacing in this embodiment is as follows. That is, when rolling metal strip S using rolling mills 2A, 2B, 2C, and 2D, the rolling speed is not necessarily constant, and the rolling speed varies within the metal strip S when rolling one strip. Therefore, even if the vibration is generated at the same spacing, if the rolling speed is different, it will be measured as vibration at different frequencies. In this case, if the vibrations of multiple frequency bands overlap, it is impossible to clearly determine whether the causes of the vibrations are the same when the rolling speed changes. Therefore, in order to evaluate vibration phenomena generated from the same vibration source and observed at different frequencies according to the rolling speed using a unified index, a standard spacing is set. That is, relative to a vibration source generated at a constant spacing, the vibration behavior observed at different frequencies due to different rolling speeds is converted into vibration behavior corresponding to the standard rolling speed, and this is expressed as the vibration intensity for each spacing. Thus, the vibration intensity at any rolling speed actually achieved in operation can be evaluated using this constant index of vibration intensity corresponding to the standard spacing.

[0056] Furthermore, the data transformation unit 14 transforms the vibration intensity of each reference frequency (first analytical data) into the vibration intensity of each standard spacing (second analytical data) using the rolling speed during operation by performing data interpolation such as interpolation or extrapolation. At this time, linear interpolation can be used, interpolating the DC component with a frequency component of "0" as "0". Additionally, all frequencies used for extrapolation are set to "0". Therefore, even if the rolling speed varies for each metal strip, the frequencies that generate abnormalities can be evaluated using the constant index of the standard spacing. It should be noted that, from the perspective of determining the abnormality of vibration corresponding to a specific spacing, when referring to "spacing", it is used to mean a "standard spacing" established by the reference frequency and the standard rolling speed. That is, unless otherwise specified, "spacing" is synonymous with "standard spacing".

[0057] Here, the vibrations measured by vibration meters 8A, 8B, 8C, and 8D installed on rolling mills 2A, 2B, 2C, and 2D will be explained. In vibration meters 8A, 8B, 8C, and 8D, the vibrations caused by the rotation of the work roll 4, etc., and the vibrations of the natural periods of rolling mills 2A, 2B, 2C, and 2D are measured overlappingly. The former vibration varies with the rolling speed, while the latter vibration is measured as vibration independent of the rolling speed. Therefore, if the rolling speed changes, the frequency of the vibrations caused by the rotation of the work roll 4, etc., measured by vibration meters 8A, 8B, 8C, and 8D changes. On the other hand, regarding the vibration intensity corresponding to the vibrations of the natural periods of rolling mills 2A, 2B, 2C, and 2D, although the frequency of the vibration does not change significantly, the magnitude (amplitude) of the vibration intensity frequently changes. Based on the vibration characteristics of such rolling mills, methods that focus on a specific frequency and detect abnormal vibrations based on the vibration intensity at that frequency, even if they can detect abnormalities corresponding to the vibrations of the natural periods of rolling mills 2A, 2B, 2C, and 2D, sometimes it is difficult to detect abnormalities related to rotating bodies such as work rolls 4, support rolls 5, and their bearings in rolling mills 2A, 2B, 2C, and 2D. In contrast, this embodiment addresses the case of different rolling speeds, converting the vibration intensity to the vibration intensity of each standard pitch, thus making it easier to detect abnormalities in the rotational vibration system generated at specific pitches.

[0058] exist Figure 9 The diagram shows an embodiment of the anomaly detection device 10 comprising both a data transformation unit 14 and a principal component derivation unit 16. The first analytical data representing the relationship between frequency and vibration intensity generated by the frequency analysis unit 12 is transformed by the data transformation unit 14 into second analytical data composed of spacing and vibration intensity using the aforementioned equation (1). On the other hand, Figure 9The principal component derivation unit 16, as shown, uses the data transformation unit 14 to generate multiple second reference vibration data consisting of spacing and vibration intensity based on the vibration intensity (reference vibration data) at each frequency under normal conditions. It then performs analysis to synthesize a principal component vector from numerous correlated variables to best represent the overall deviation without any correlation. Figure 9 In the principal component derivation unit 16 shown, the first to i-th principal components, determined as principal component vectors that best represent the overall deviation with a small number of unrelated components based on multiple related second reference vibration data with spacing as the variable, are calculated as reference data. Hereinafter, the reference data calculated based on the second reference vibration data will be referred to as the second reference data. Figure 9 The second reference data obtained by the principal component derivation unit 16 is sent to the principal component analysis unit 13, and the projection (evaluation data) of the second analytical data obtained by the principal component analysis unit 13 during operation onto the first principal component to the i-th principal component can be calculated. Hereinafter, the evaluation data calculated as the projection of the second analytical data onto the second reference data will be referred to as the second evaluation data.

[0059] Then, in the principal component analysis unit 13, based on the second reference data (first principal component to i-th principal component) obtained from the principal component derivation unit 16 and the second analytical data representing the relationship between spacing and vibration intensity obtained from the data transformation unit 14, the projection of the second analytical data during operation onto the second reference data is generated as second evaluation data. Then, in the anomaly detection unit 15, the difference (deviation component) between the second analytical data composed of the vibration intensity of each spacing and the second evaluation data generated by the principal component analysis unit 13 is calculated. If the calculated deviation component is above a preset threshold, it is determined that abnormal vibration has occurred.

[0060] Reference Figure 1 and Figure 2The operation of the abnormal vibration detection method and abnormal detection device 10 for the rolling mill of the present invention will be explained. During the cold rolling of metal strip S, i.e., during the operation of the rolling mill 1, vibration data of the rolling mills 2A, 2B, 2C, and 2D are measured by vibration meters 8A, 8B, 8C, and 8D, and collected by the data collection unit 11 (collection step). In the collection step, data of the data sampling time is collected every data acquisition cycle. When accelerometers 8A, 8B, 8C, and 8D are used as vibration meters, the time-series data of acceleration acquired by each vibration meter 8A, 8B, 8C, and 8D is converted into vibration data of vibration velocity. The collected vibration data is frequency-analyzed by the frequency analysis unit 12 to generate first analysis data (frequency analysis step). The frequency analysis unit 12 generates the first analysis data, which is composed of the relationship between frequency and vibration intensity, every data acquisition cycle. However, when a data conversion unit 14 is provided, the first analysis data is converted by the data conversion unit 14 into second analysis data representing the relationship between spacing and vibration intensity.

[0061] The first analytical data is subjected to principal component analysis by the principal component analysis unit 13, and evaluation data is generated as a projection onto the pre-exported reference data (first reference data) (principal component analysis step). The principal component analysis performed by the principal component analysis unit 13 calculates the projection of the first analytical data relative to the space composed of the pre-set principal component vectors. Then, the anomaly detection unit 15 calculates the difference (deviation component) between the first analytical data composed of the vibration intensity of each frequency and the projection of the first analytical data relative to the first principal component to the i-th principal component (first reference data), which is the evaluation data. If the deviation component is above a pre-set threshold, abnormal vibration is detected in the rolling mills 2A, 2B, 2C, and 2D (abnormal vibration detection step).

[0062] On the other hand, when the data transformation unit 14 is provided, second analytical data is generated in the data transformation unit 14 and sent to the principal component analysis unit 13. In this case, in the principal component analysis step, the second analytical data is subjected to principal component analysis by the principal component analysis unit 13, and second evaluation data is generated as a projection onto the pre-exported reference data (second reference data). The principal component analysis performed by the principal component analysis unit 13 calculates the projection of the second analytical data relative to the space composed of principal component vectors composed of the second reference data. Afterwards, in the abnormal vibration detection step, the abnormality detection unit 15 calculates the difference (deviation component) between the second analytical data composed of the vibration intensity of each spacing and the projection of the second analytical data relative to the first principal component to the i-th principal component (second reference data), which is the second evaluation data. If the deviation component is above a pre-set threshold, abnormal vibration is detected in the rolling mills 2A, 2B, 2C, and 2D.

[0063] According to the above implementation method, abnormal vibrations of rolling mills 2A, 2B, 2C, and 2D that cause chatter marks can be detected with high precision. In other words, by determining the principal components of characteristic quantities representing the reference vibration data under normal conditions—such as vibration components naturally generated by the meshing of gears in rolling mills 2A, 2B, 2C, and 2D, and vibration characteristics of the bearings in rolling mills 2A, 2B, 2C, and 2D—as the inherent characteristics of the equipment, such as vibration components naturally generated by the meshing of gears in rolling mills 2A, 2B, 2C, and 2D, and vibration characteristics of the bearings in rolling mills 2A, 2B, 2C, and 2D—it is possible to achieve analysis that only makes abnormal vibrations significant.

[0064] Among the abnormal vibrations in rolling mills 2A, 2B, 2C, and 2D, the majority are inherent vibrations of the mills themselves, as well as vibrations caused by equipment rotation due to bearing defects, gear meshing, poor coupling, or looseness. Therefore, conventional detection of abnormal vibrations relies on whether the amplitude of a specific frequency exceeds a certain threshold. On the other hand, in the case of chatter marks, minute vibrations occur at frequencies comparable to the spacing of the chatter marks from before their formation, and these vibrations grow over time. That is, minute vibrations caused by the equipment occur first, followed by chatter marks on the surface of the metal strip S. However, in actual operation, the stable rolling speed varies for each metal strip S. Therefore, focusing solely on a pre-set specific frequency makes it difficult to detect the minute vibrations that precede abnormal vibrations. In contrast, in this embodiment, since the frequency or spacing at which vibration intensity increases can be recognized during operation, abnormal vibrations can be detected early, regardless of the frequency band at which they occur. As a result, the operation of the rolling mill can prevent or suppress abnormal vibrations, and defects on the surface of the metal strip caused by abnormal vibrations can be prevented or suppressed, resulting in the production of metal strips with excellent appearance.

[0065] Furthermore, the investigation into the vibration source of the abnormal vibration that caused the chatter marks revealed that, at the time of chatter mark formation, there were also micro-marks at the same spacing as the chatter marks on the surface of one of the upper or lower support rolls 5 (for example, the support roll 5 was worn into a polygonal shape). Moreover, as the micro-marks on the support roll 5 gradually became clearer due to resonance with the rolling mills 2A, 2B, 2C, and 2D at a specified rolling speed, the abnormal vibrations of the rolling mills 2A, 2B, 2C, and 2D gradually increased.

[0066] The spacing of the fine imprints on support roll 5 does not change with the rolling speed. These fine imprints on support roll 5 are not visible before the mills 2A, 2B, 2C, and 2D are loaded, making it impossible to predict the wavelength (spacing) and frequency of the chatter marks. Furthermore, the vibration data also includes vibrations from many other factors, such as the meshing frequencies of bearings and gears, which contribute to the constant wavelength vibrations, preventing a clear peak value for the chatter marks from the outset.

[0067] Therefore, frequency analysis is performed on the vibration data to determine the relationship between the frequency or standard spacing and the vibration intensity at that moment. Principal component analysis is then used to distinguish the vibration peak of the chatter mark from other vibration factors, enabling the detection of abnormal vibrations. This allows for the high-precision detection of abnormal vibrations in mills 2A, 2B, 2C, and 2D caused by chatter marks.

[0068] Furthermore, the configuration can be such that, when abnormal vibration is detected using the abnormal vibration detection methods of the aforementioned mills 2A, 2B, 2C, and 2D, the rotating components causing the abnormal vibration, such as the support rolls 5 of mills 2A, 2B, 2C, and 2D, are replaced (support roll replacement step). Thus, even when rolling multiple metal strips S over a long period, rolling operations that prevent abnormal vibrations at any frequency or spacing can be achieved. Additionally, through such rolling, chatter marks are not generated on the surface of the metal strip S, or their generation can be suppressed, resulting in metal strips S with excellent appearance. On the other hand, when it is empirically known that abnormal vibrations occur at a specific frequency or spacing, actual data on the deviation components of each specific frequency or spacing can be obtained in advance, and a threshold for the deviation components that meet the product shipment criteria for the metal strip S can be set based on this actual data. Furthermore, if abnormal vibration is detected in the aforementioned abnormality detection unit 15 using a set threshold for the deviation component, the operation of the rolling mill 1 can be temporarily stopped, and the rotating parts such as the support rolls 5 of the rolling mills 2A, 2B, 2C, and 2D that caused the abnormal vibration can be replaced. Thus, even when rolling multiple metal strips S over a long period, rolling operations that prevent abnormal vibrations occurring at specific frequencies or spacings can be achieved.

[0069] Example 1

[0070] The following describes Example 1, conducted to confirm the operation and effects of the present invention. A rolling mill consisting of four stands (continuous rolling mill) was used, and vibration meters were installed in the housing of each stand. A sampling frequency of 4000 Hz was used, and the reference frequency was set every 5 Hz between 0 Hz and 2000 Hz. The data collection unit 11 obtained the intensity of the vibration velocity every data acquisition cycle with a data sampling time of 0.2 sec and a data acquisition period of 1 sec, and output it to the frequency analysis unit 12. The test material was a metal strip (hereinafter referred to as a steel plate) made of low-carbon steel, and this steel plate was prepared for Examples 1 and 2 of Embodiment 1, and Comparative Examples 1 and 2, respectively. In the rolling direction (travel direction) of the above-mentioned steel plate, the entry thickness of the rolling mill was set to 2.0 mm or more and 5.5 mm or less, the exit thickness was set to 0.5 mm or more and 2.4 mm or less, and the width of the steel plate was set to 700 mm or more and 1700 mm or less.

[0071] Vibration data during the rolling of the steel plate was collected, and the vibration data at the time when chatter marks occurred were extracted from the storage device storing the collected vibration data. Chatter mark determination was performed using a method that assesses the vibration intensity of the rolling mill, including the present invention, and determines abnormal vibrations based on the output of a vibration meter installed on the rolling mill. A method involving abrasive stone inspection of the steel plate after alloyed galvanizing was also performed, and chatter marks were visually determined by visually assessing the spacing of vibration frequencies. In the abrasive stone inspection, a light pressing force of approximately 10N was applied against the alloyed hot-dip galvanized steel plate, and manual grinding was performed in the rolling direction. It was evaluated whether the striations in the steel plate's travel direction could be visually identified as chatter marks. It should be noted that the patterns (fine marks) remaining on the support roll 5 of the fourth stand confirmed that the rolling mills that produced chatter marks were all the final stand (fourth stand). The evaluation results are shown in Table 1.

[0072] [Table 1]

[0073]

[0074] In Table 1, in the column for "Scattered Mark Judgment Based on Vibration Intensity," the judgment result based on the vibration intensity of the rolling mill is marked as "×" if abnormal vibration of the rolling mill is determined to have occurred, and marked as "○" if no abnormal vibration of the rolling mill is determined to have occurred. On the other hand, in the column for "Scattered Mark Judgment Based on Visual Inspection," the case where scribbles are visually detected on the surface of the alloyed hot-dip galvanized steel sheet after it has been pressed against a grinding stone is marked as "×," and the case where no scribbles are visually determined to have occurred is marked as "○." In this case, the scribbles are a quality abnormality of the steel sheet, therefore, under the condition that scribbles are detected visually, abnormal vibration of the rolling mill has occurred. In addition, in the column for "Scattered Mark Judgment Based on Vibration Intensity," the vibration frequency or spacing is recorded in parentheses when scribbles are detected. The vibration frequency when scribbles are detected is the frequency when the vibration intensity exceeds a preset threshold. The spacing when scribbles are detected is the spacing when the deviation component in the abnormality detection unit 15 exceeds a preset threshold. Similarly, in the "Visual Judgment of Drill Marks" section, when drill marks are found, the vibration frequency or spacing is recorded in parentheses. The spacing, when drill marks are found, is the spacing of the stripe pattern observed visually. The frequency, when drill marks are found, is the result of converting the visually measured stripe pattern spacing into the frequency of abnormal vibration using the actual value of the rolling speed of that portion of the steel plate (i.e., the drill mark) when it passed through the final stand (fourth stand) where the drill mark occurred.

[0075] In Example 1 of the invention described in Table 1, a vibration meter is installed on the housing of the final frame (fourth frame). After generating first analytical data using the frequency analysis unit 12, evaluation data with the first reference data as the variable is calculated using principal component analysis in the principal component analysis unit 13. Furthermore, after calculating the difference between the first analytical data and the evaluation data, i.e., the deviation component, in the anomaly detection unit 15, the presence or absence of flutter is determined based on a preset threshold for the deviation component at each frequency. On the other hand, regarding the principal components (first reference data), the first reference data is calculated using the principal component derivation unit 16. This is done on the reference vibration data calculated from the vibration data measured for one day during normal operation approximately two days after the replacement of the support roller 5. Principal component analysis is performed to synthesize a principal component vector that best represents the overall deviation from a small number of correlated variables. It should be noted that 10 principal components were selected for the principal component vector in descending order of contribution rate.

[0076] In contrast, Invention Example 2 in Table 1 uses... Figure 9 The example shown is an anomaly detection device 10 comprising both a data transformation unit 14 and a principal component derivation unit 16. First, based on reference vibration data obtained under normal conditions, second reference vibration data is generated, transformed by the data transformation unit 14 into a relationship between spacing and vibration intensity. In the principal component derivation unit 16, principal component vectors, i.e., second reference data, are derived from the second reference vibration data, representing the overall deviation best represented by a small number of uncorrelated components. Then, through principal component analysis by the principal component analysis unit 13, second evaluation data regarding the second reference vibration data, with the second reference data as a variable, is calculated. In the anomaly detection unit 15, the difference between the second analytical data and the second evaluation data is calculated as a deviation component, and for each spacing deviation component, the presence or absence of a flutter is determined based on a pre-set threshold. It should be noted that, similarly with Example 2, in the principal component derivation unit 16, 10 principal components are selected in descending order of contribution rate and used as the second reference data.

[0077] Furthermore, the principal component derivation unit 16 derives principal components (first reference data and second reference data) based on reference vibration data obtained from various rolling speeds, dividing the rolling speed into 50 m / min intervals for each rolling speed. In Invention Example 1, principal component analysis was performed on the first analytical data composed of vibration intensity at each frequency using the first reference data, and in Invention Example 2, principal component analysis was performed on the second analytical data using the second reference data.

[0078] On the other hand, in Comparative Examples 1 and 2, a vibration meter was installed on the housing of the fourth stand. Principal component analysis was not performed, and the determination was made by changing the threshold value of the first analytical data, composed of the vibration intensity at each frequency, output by the frequency analysis unit 12. In this case, Comparative Examples 1 and 2 are examples of attempting to determine abnormal vibration using different threshold values ​​for the first analytical data obtained when chatter marks were generated and when chatter marks were not generated, at the same rolling speed. That is, Comparative Example 1 set the threshold value related to vibration intensity for determining abnormal vibration as Fref1, and Comparative Example 2 set the threshold value related to vibration intensity for determining abnormal vibration as Fref2.

[0079] Figure 3 This is a graph of the deviation components calculated in the anomaly detection unit 15 for each frequency, representing Example 1 of the invention. Figure 4 This is a graph of Invention Example 2 showing the deviation components for each standard spacing. It should be noted that... Figure 3 Since the peak position of the deviation composition changes with the rolling speed, the deviation composition at a rolling speed of 14 m / sec is shown. Figure 4 This also shows the deviation composition at a rolling speed of 14 m / sec. According to... Figure 3 and Figure 4 The projection of the analytical data (first analytical data and second analytical data) onto the principal components (first reference data and second reference data) is calculated by principal component analysis. The deviation components corresponding to the first analytical data and second analytical data are calculated in the anomaly detection unit 15, so that the abnormal vibrations caused by the flutter and the normal vibrations that do not cause the flutter can be clearly distinguished. By setting an appropriate threshold Fref, the presence or absence of abnormal vibrations that cause the flutter can be determined with high precision.

[0080] Figure 5 This is a graph showing comparative examples 1 and 2, where anomaly determination was performed using the first analytical data representing the vibration intensity at each frequency output by the frequency analysis unit 12. It should be noted that in... Figure 5 Since the peak position of vibration intensity changes with the rolling speed, the frequency analysis results at a rolling speed of 14 m / sec are shown. In Comparative Example 1, even when chatter marks were generated, the vibration intensity of the abnormal vibration did not reach the threshold Fref1, and the abnormal vibration was not detected. On the other hand, in Comparative Example 2, when the threshold Fref2 was used, although the vibration intensity of the abnormal vibration could be detected, even when chatter marks were not generated, that is, when vibrations caused by frequencies other than chatter marks occurred, there were vibration intensities exceeding the threshold Fref2, resulting in false detections of abnormal vibrations.

[0081] This is because the vibration intensity at each frequency overlaps with the various vibrations generated in each stand of the continuous rolling mill, making it difficult to distinguish between situations where chatter marks have occurred and those where they haven't. Indeed, around 530Hz, a difference in vibration intensity can be observed between situations where chatter marks have occurred and those where they haven't. However, to detect chatter marks based on this difference, a frequency around 530Hz needs to be preset, along with a threshold corresponding to that frequency band. Furthermore, such preset settings are difficult to perform. Therefore, in practice, it is difficult to detect the occurrence of chatter marks with high precision. As a result, conventionally, as in Comparative Examples 1 and 2, it was necessary to pre-test and erroneously change the threshold related to vibration intensity during strip rolling, resulting in a missed rate of 10 pieces / month for chatter marks. In contrast, by performing the anomaly determination in the aforementioned Invention Example 1, the missed rate for chatter marks was reduced to 1 piece / month.

[0082] Furthermore, according to the results shown in Table 1, in Invention Examples 1 and 2, which include principal component analysis and abnormal vibration detection steps, the evaluation results for "chatter detection using vibration intensity" and "chatter detection under visual inspection" are consistent. Therefore, it is unnecessary to perform the determination of abnormal vibrations related to the rolling mill by visually inspecting for chatter after rolling the metal strip, as was previously done; abnormal vibrations of the rolling mill can be determined online during the operation of the rolling equipment.

[0083] Example 2

[0084] Hereinafter, Example 3 and Comparative Example 3, which are embodiments of the present invention, are shown. In Example 2, the equipment subject to this invention is a continuous rolling mill consisting of five stands, with vibration meters installed in the housing of each stand. The data acquisition method, including the setting of the reference frequency, is the same as in Example 1. In Example 2, the stand where chatter marks occurred is also the final stand (in this case, the fifth stand). Therefore, mill anomaly detection was performed using data from the vibration meters installed in the working-side housing of the fifth stand. In Comparative Example 3, chatter mark determination was attempted while varying the threshold values ​​for the vibration intensity at each frequency calculated based on the vibration data from the vibration meters installed in the working-side housing of the fifth stand. The results are shown in Table 2. It should be noted that the meanings of “○”, “×”, and parentheses in Table 2 are the same as in Table 1.

[0085] [Table 2]

[0086]

[0087] Figure 6This is a graph of Invention Example 3 showing the deviation component for each standard spacing. The second reference data in Invention Example 3 is derived using the following second reference vibration data: vibration data measured one day after one week following the replacement of the support roller 5. The vibration intensity of each of the multiple frequencies collected across the entire velocity domain is converted into the vibration intensity for each standard spacing using the data transformation unit 14. That is, the principal component vector for the second reference vibration data is calculated using the principal component derivation unit 16 and obtained as the second reference data. Furthermore, through principal component analysis by the principal component analysis unit 13, second evaluation data for the second analytical data is calculated using the second reference data as a variable. In the anomaly detection unit 15, the difference between the second analytical data and the second evaluation data is calculated as the deviation component. For example... Figure 6 As shown, through principal component analysis, abnormal vibrations that produce chatter marks and normal vibrations that do not produce chatter marks are clearly distinguished. The presence or absence of abnormal vibrations in the rolling mill that cause chatter marks is determined with high precision using a set threshold Fref. That is, it means that even without pre-determining the frequency of the vibrations that produce abnormal chatter marks, a suitable threshold can be set by calculating the deviation component relative to the standard spacing.

[0088] Figure 7 This is a graph showing Comparative Example 3, which uses a threshold value for vibration intensity relative to each frequency. That is, it's an example of anomaly detection using the first analytical data representing the vibration intensity at each frequency output by the frequency analysis unit. However, since the peak position of the vibration intensity changes with the rolling speed, the frequency analysis results at a rolling speed of 14 m / sec are shown. Comparative Example 3 determines the generation of chatter marks by setting a threshold value Fref3 relative to the vibration intensity. Figure 7 As shown, if a specific frequency is preset and a threshold is set only for that frequency band, it may be possible to determine the generation of jitter. However, it is difficult to make such a setting in advance. Therefore, it is actually difficult to detect the generation of jitter with high precision.

[0089] The embodiments of the present invention are not limited to the above-described embodiments, and various modifications can be made. For example, in the above embodiments, the case where the metal strip S is cold-rolled steel sheet has been illustrated, but the metal strip S may also be stainless steel instead of cold-rolled steel sheet, or it may be hot-rolled steel sheet. The rolling mills 2A, 2B, 2C, and 2D may not have the same structure; for example, as rolling mills, a mixture of 4-segment rolling mills and 6-segment rolling mills may also exist.

[0090] Explanation of reference numerals in the attached figures

[0091] 1. Rolling equipment

[0092] Rolling mills 2A, 2B, 2C, and 2D

[0093] 3. Shell

[0094] 4 working rolls

[0095] 5 Support Rollers

[0096] 6. Drive unit

[0097] 7 Small Diameter Rollers

[0098] Vibration meters 8A, 8B, 8C, and 8D

[0099] 10. Anomaly detection device for rolling mill

[0100] 11 Data Collection Department

[0101] 12 Frequency Analysis Section

[0102] 13 Principal Component Analysis Department

[0103] 14 Data Transformation Unit

[0104] 15. Anomaly Detection Department

[0105] S-shaped metal strip.

Claims

1. An abnormal vibration detection method of a rolling mill, which is an abnormal vibration detection method of a rolling mill having a pair of work rolls and a plurality of backup rolls that support the work rolls, wherein comprising: a collecting step of collecting vibration data of the rolling mill; a frequency resolution step of performing frequency resolution of the vibration data to generate first resolution data; a principal component analysis step of performing principal component analysis of the first resolution data using reference data determined in advance based on a normal state as principal components to generate evaluation data that is a projection of the first resolution data to the reference data; and an abnormal vibration detection step of extracting a deviation component from the evaluation data and the first resolution data and detecting an abnormality of the rolling mill based on the extracted deviation component.

2. The abnormal vibration detection method of a rolling mill according to claim 1, in the principal component analysis step, the principal components extracted as the reference data are set for each rolling speed in the rolling mill.

3. The abnormal vibration detection method of a rolling mill according to claim 1, the frequency resolution step generates vibration intensity of each frequency as the first resolution data, the abnormal vibration detection method further includes a data transformation step of transforming the first resolution data into second resolution data representing vibration intensity of each pitch based on a rolling speed, the principal component analysis step performs principal component analysis of the second resolution data.

4. The abnormal vibration detection method of a rolling mill according to any one of claims 1 to 3, in the principal component analysis step, a plurality of principal components extracted as the reference data are set in a manner that a cumulative value of contribution rates of principal components when principal component analysis is performed on normal resolution data obtained when rolling is performed using the normal rolling mill becomes a reference contribution rate or more.

5. The abnormal vibration detection method of a rolling mill according to any one of claims 1 to 3, the rolling mill cold-rolls a metal strip.

6. The abnormal vibration detection method of a rolling mill according to claim 4, the rolling mill cold-rolls a metal strip. comprising:

7. An abnormality detection device of a rolling mill, which is an abnormality detection device of a rolling mill having a pair of work rolls and a plurality of backup rolls that support the work rolls, wherein a data collecting unit that collects vibration data of the rolling mill; a frequency resolution unit that performs frequency resolution of the vibration data to generate first resolution data; a principal component analysis unit that performs principal component analysis of the first resolution data using reference data determined in advance based on a normal state as principal components to generate evaluation data that is a projection of the first resolution data to the reference data; and an abnormality detection unit that extracts a deviation component from the evaluation data and the first resolution data and detects an abnormality of the rolling mill based on the extracted deviation component.

8. A rolling method including a backup roll replacement step of replacing a backup roll of a rolling mill in a case where an abnormality of the rolling mill is detected using the abnormal vibration detection method of a rolling mill according to any one of claims 1 to 6.

9. A metal strip manufacturing method including a step of manufacturing a metal strip using the rolling method according to claim 8. ​ ​

Images