US20070227214A1

US20070227214A1 - Quality-control method for wire/rod rolling system

Info

Publication number: US20070227214A1
Application number: US11/731,852
Authority: US
Inventors: Martin Sauerland
Original assignee: SMS Meer GmbH
Current assignee: SMS Group GmbH
Priority date: 2006-03-30
Filing date: 2007-03-30
Publication date: 2007-10-04
Also published as: EP1839767A1; DE102007009946A1

Abstract

Quality of a wire or rod is controlled in a rolling mill having a succession of roll stands spaced apart in a workpiece-travel direction by a method having the first step of, before rolling a wire or rod, statically measuring a sample of at least one wire or rod with typical real roll flaws and storing the flaw data in standard format in a control computer. A single cross-sectional shape or, if there is noise, an average of a plurality of averaged cross-sectional shapes are stored standardized as polar coordinates. During actual rolling of a wire or rod its actual-shape is continuously measured and actual-shape data is generated. A similarity analysis is continuously effected by comparing the flaw data in standard format with the actual-shape data of the wire or rod being rolled. When the actual cross-sectional shape data corresponds to the flaw data, an alarm is generated.

Description

FIELD OF THE INVENTION

The present invention relates to the rolling or round-section stock, that is wire or rod, hereinafter referred to wire. More particularly this invention concerns a method of controlling such a computer-controlled wire-rolling system for quality control, that is to ensure that the finished product is of the desired size and shape.

BACKGROUND OF THE INVENTION

The rolling of round profiles, particularly of rods or of wire, is performed in the known manner in a corresponding rolling mill. The rolling quality is inspected at the outlet end of the rolling mill. Methods are known for determining the diameter of the rolled rod or wire, for instance at a number of different angularly offset positions, and inspect it as to whether the measured values, meaning the actual values, are within the defined tolerances.
However, during this process it is not possible, or only conditionally and/or only with the appropriate experience of experts, to use the measured diameters across the circumference of the profile to find rolling-related flaws.

OBJECTS OF THE INVENTION

It is therefore an object of the present invention to provide an improved quality-control method for wire/rod rolling system.
Another object is the provision of such an improved quality-control method for wire/rod rolling system that overcomes the above-given disadvantages, in particular that allows improved discovery of rolling-related flaws based on the data of the rolled round profile measured at the outlet of the rolling mill.
Yet another object is to provide such a method that can be carried out automatically, without resorting to the judgment of experts. Hence, the invention is intended to provide improved rolling stock quality and increased output of the rolling mill. Furthermore, rolling flaws, for example stand overfill or stand underfill, must be to be identified and classified automatically.

SUMMARY OF THE INVENTION

Quality of a wire or rod is controlled in a computer-controlled rolling mill having a succession of roll stands spaced apart in a workpiece-travel direction by a method comprising the first step of, before rolling a wire or rod, statically measuring a sample wire with typical real roll flaws and storing the flaw data in standard format in a control computer. A single cross-sectional shape or, if there is noise, an average of a plurality of averaged cross-sectional shapes are stored standardized as polar coordinates. During actual rolling of a wire or rod, its actual-shape is measured are measured and actual-shape data is generated. A similarity analysis is continuously effected by comparing the flaw data in standard format with the actual-shape data of the wire or rod being rolled. When the actual cross-sectional shape data corresponds to the flaw data, an alarm is generated.
In other words according to the invention before starting the production of the round profile at least one profile sample with typical, real rolling flaws is measured statically and the sample flaws are stored in a standardized format in the computer system. An individual profile cross-section, or in the case of noise an averaged profile cross-section, is stored in polar coordinates and standardized. During production of the round profile a similarity analysis is conducted such that the standardized sample cross-section is continuously compared with the measured cross-sections of the currently rolled round profile. Any similarity deviations determined following a comparison with the sample flaws trigger a warning that results in intervention in the rolling operation, the function preferably being displayed on a control monitor.
Thus during the similarity analyses the actual cross-sectional is compared with a sample-flaw cross-section. The correlation is easily effected without complex mathematical manipulation. In this manner as a result it is possible to find flaws that would be difficult or impossible to determine mathematically. It is only necessary to compare the sample flaws with a sufficient number of support values and to store the entire data stream. This sample-data information is then used later during the actual measuring of a piece in production, with mathematical analysis.
The measuring device used for measuring the sample profile section statically must be able to produce the profile cross-section with sufficient precision as a radius profile over different angles. Before starting the production of the round profile, the following steps are taken:
a₁) measuring the profile radius (r) at a number of angular positions (Φ) around the circumference of the round profile for at least one round profile with a defined rolling flaw (F);
b) during production of the round profile:

- b₁) measuring the profile radius (r) for a number of angular positions (Φ) around the circumference of the round profile on the finished round profile;
- b₂) forming an approximation curve for the radius (r) via the angle of circumference (Φ) and the axis direction (z) of the produced round profile according to the following formula:

r _{N IST}(Φ, z)=r _{0 IST}(z)+r ₃×cos (3(Φ+Φ₀))−r ₆×cos (6(Φ+Φ₀)),
where:

- r_{0 IST}(z)=radial path along axis direction z for rolling flaw (F),
- r₃=radii portion of the third order,
- r₆=radii portion of the sixth order, and
- Φ₀=angle constant;
- b₃) storing the approximation curve r_{N IST}(Φ, z) for the produced round profile;

c) carrying out a comparison of the approximation curve (r_{N IST}(Φ, z)) determined for the produced round profile with at least the approximation curve (r_NF(Φ, z)) of the rolling flaw.
It is further proposed according to the instant invention to also standardize and store the currently determined profile cross-section before doing the correlation representing the degree of similarity, preferably a cross-correlation.
Preferably the differences determined in step c) between the approximation curve determined for the produced round profile and at least one approximation curve of the rolling flaw are displayed in graphs.
Since the high temperature of the rolled profile due to thermal expansion causes measurement flaws, it is preferably provided that during the measurement of the profile radius according to step b₁) the temperature of the measured round profile is determined. Then the values measured at the temperature are calculated to a standard temperature, preferably 20° C., based on the known coefficient of thermal expansion for the material, normally steel, in question.
The measurement of the profile radius according to steps a₁) and b₁) according to one embodiment of the invention can be carried out in a contactless fashion by means of laser. The measurement of the profile radius according to steps a₁) and b₁) is advantageously carried out for at least 36 evenly distributed circumferential angle positions, preferably for 180 evenly distributed circumferential angle positions.
The determination of the radius portions r₃and r₆according to the above formula is preferably carried out by means of a Fourier transformation, particularly a Fast Fourier Transformation.
For evaluation and documentation purposes it has proven useful to also measure the minimum and/or maximum diameters of the round profile, relative to its overall length. Furthermore, at least one data point representing the out-of-roundness or ovalness of the round profile can be determined and stored.
The measurement according to the above step b) is preferably carried out after rolling in a calibrating stand.
The comparison according to the above step c) between the approximation curve and at least one approximation curve of the rolling flaw can be carried out by means of a cross-correlation function or a cross-covariance function.
The invention hence enables effective quality assurance in a rolling mill for round profiles, particularly in rod or wire rolling mills, where a typical and commercially available profile-measuring device is employed. According to the invention, this device is expanded as explained above into a quality assurance module, allowing autonomous assessment of the measurement results.

BRIEF DESCRIPTION OF THE DRAWING

The above and other objects, features, and advantages will become more readily apparent from the following description, reference being made to the accompanying drawing whose sole FIGURE is a diagram illustrating the method of this invention.

SPECIFIC DESCRIPTION

At the outlet end of a rolling mill, the radius and/or the diameter of the rolled profile are measured, the measurement being carried out at a number of angularly offset positions around the circumference of the workpiece.
A data stream containing statistical data about the rod is transmitted for storage for each rolled rod. The data stream may also contain characteristic data and flaw classifications. The determination of the final content of the data stream is only done when it has been defined after initial measurements of sample steels, which measurements are the most reliable and meaningful, to obtain characteristic values that allow a conclusion regarding potential rolling flaws.
In doing so, the minimum and maximum rod diameters, relative to the overall length of the rod, as well as information regarding the out-of-roundness of the round rod are included in the data stream at all times.
Furthermore, it is possible to transmit the online data stream to the central software, that is to the central control computer.
For communication with the central software, parameterizations of the profile measuring devices are conducted. Optionally, the expanded analyses are visualized on an additional computer that is connected to the existing device via a network.
The rod is measured directly downstream of the calibrating roll stands. At the location of the measurement, about 1 meter downstream of the output of the calibrating roll stands, the rods run through the measuring devices approximately with a fixed center position. The measuring device operating with contactless laser technology supplies the comprehensive cross-section information about the currently processed rod in cyclical intervals; this is done virtually in real time.
Every cyclically determined and displayed rod cross-section is formed by 180 individual radii that have been determined with high precision, meaning the required angular resolution is 2°. The cross-sectional information supplied this way allows conclusions of the actual rolling quality. Inferences on the fill situation of the stands, —such as overfill or underfill—, on the roll offset, triangulation, and other rolling flaws, be they of symmetrical or asymmetrical nature, are visualized online during rolling.
The comprehensive data transmission to the data analysis tool described here makes it possible to determine additional characteristic values for flaw classification purposes.
A temperature-measuring device is integrated in the diameter-measuring device to calculate the rod geometry with the actual temperature at the cold dimension (at 20° C. as the reference) for the known shrinkage.
The hot, glowing round rod is measured by means of a known laser scanning method. The process is carried out using a diameter measuring device provided with six laser micrometers that are distributed evenly around the circumference, to be able to measure up to 95 mm with a measuring field of 120 mm. The device is provided in a pivotable design with the operating modes “fixed position,” “scan,” and “constant pivoting”.
Special evaluations make it possible to reliably and clearly capture, display and quantify the following rolling flaws:

- triangulation or out-of-roundness or equivalently hexagon;
- roll offset (symmetrical and asymmetrical);
- underfill and overfill (symmetrical and asymmetrical; underfill has limited application possibilities due to the system design because concave areas cannot be detected);
- roll breaks (dependent on the rolling speed and the flaw extension along the rod).

Furthermore, the positions and angles of the passes can be determined and issued or displayed with precision.
A preferred embodiment with a pivot frame increases the expected accuracy of the measurement results during pivot operation, particularly with roll offset and simultaneous fill flaws.
In particular, the procedure could be as described below, resulting also in the functionalities of the proposed system:
In the rod-measuring location, the rod profile as well as the measuring product temperature are determined along the measured length. In the background, the subsequent computations are carried out continuously and produce the additional results and visualization of the results outlined below.
The longitudinally rolled rods have more or less distinctly pronounced outside structures as a result of the manufacturing method using a finish block in a three-roll technique. This means that the rods deviate from an ideal cylindrical shape on their outer surfaces. Structures of this type can be characterized with the order analysis methods known from pipe technology.
In addition to the basic forming structures referred to above, “real” rolling flaws may occurs, such as
pass overfill,
pass underfill, and
roll offset.
All flaws can occur with three-point symmetry or asymmetrically. The roll offset may occur superimposed with the pass overfill or underfill.
If “real” rolling flaws are disregarded, for an order analysis of the three-roll forming technology the following simplified model for the rod cross-section can be derived in polar coordinates:
r(Φ, z)=r ₀(z)+r ₃×cos (3(Φ+Φ₀))−r ₆×cos (6(Φ+Φ₀)),
where
r₁=radii.
According to the present invention, it is assumed—this being an essential finding according to the invention—that the second and fourth orders from the previous forming process are not present in the finished product in any relevant magnitude.
The above model can be determined basically for any cross-section that is supplied by default by the measuring device using the LMS (Least Mean Square) methods.
This can be used to describe characteristic values, such as triangularity, for example a triangular dimension
(r_3q/r_oq)×100%
where r_3qand r_oqare mean values for the radii r₃and r_oreferred to above.
In an equivalent fashion, a hexagon dimension can be defined as
(r_5q/r_oq)×100%
where r_5qand r_oqare mean values for the radii r₆and r_oreferred to above. The averaging should always be carried out for useful fillet regions.
To be able to reduce the computing complexity and hence processor utilization, it is possible to carry out the computation only for one in two or three cross-sections.
According to the invention, the “real” rolling flaws are determined by similarity analyses. For these similarity analyses, samples are stored in the system in a standardized format for typical real rolling flaws.
The system user provides rolled products with such typical rolling flaws. Each of these sample sections is measured by the measuring device statically. An individual rod cross-section or—in the case of noise—an averaged rod cross-section, is stored and standardized in polar coordinates.
The similarity analysis is aimed at cross-correlating this standardized sample cross-section continuously with cross-sections of the current rod. Prior to the correlation, the currently measured cross-section is likewise standardized. The rod cross-section to be correlated must be freed from pulses because otherwise the Fourier transformation, on which the cross-correlation is based, would be determined incorrectly.
In test series, it is determined at what level of similarity (in percent) an flaw alarm is triggered.
Furthermore, during a test one determines whether the cross-correlation function or the cross-covariance function is used.
In addition, one takes into consideration whether the result of the similarity analysis is unambiguous enough and whether perhaps additional characteristics are required for clear flaw identification. The following characteristics are conceivable as occurring at the same time:

- overfill=high similarity with the “overfill” sample; cross-sectional surface is increased compared to nominal dimension; FFT (Fast Fourier Transformation) via the cross-section supplies considerable amplitudes above the 6^thorder.
- underfill=high similarity with the “underfill” sample, cross-sectional surface is reduced.
- roll offset=high similarity with the roll offset sample; distribution of the radii has increased skew (statistical value).

Overall, the following characteristic values can be determined and displayed for the rod by means of the measuring device and a corresponding subsequent analysis:

- mean rod diameter;
- maximum rod diameter;
- minimum rod diameter;
- diameters of the six main axes;
- rod cross-section (minimum, maximum, mean);
- out-of-roundness (minimum, maximum, mean);
- triangle dimension;
- hexagon dimension;
- overfill (optionally visualized on the monitor by change in color);
- underfill (optionally visualized on the monitor by change in color);
- roll offset (optionally visualized on the monitor by change in color) and
- skew and/or curvature,
  where different, simultaneously occurring types of flaws can be captured as characteristic vectors M, enabling advantageous flaw classification.

The visualization of the additional results is done together with the rod measurement values as follows:
The curves for the maximum, minimum and mean diameters are recorded online as the rod passes through and are displayed, provided that this does not require excess computation capacity. The threshold values for the rod diameter (such as ¼ DIN) are inserted as horizontal lines in the longitudinal profile. The percentage is written directly next to the line. The temperature can additionally be inserted as a line with additional scaling.
At the same time, the rod cross-section is illustrated. The threshold values (such as ¼ DIN) are shown as circles. It is important for the cross-section illustration that the rod cross-section on the monitor is always shown in the same size and is only labeled differently if the diameter changes. Deviation from the standard radius is illustrated differently. This deviation can be spread, for example zoomed.
In the “current measuring data” field, the arriving values, such as maximum, minimum, excess over the mean diameter, and temperature, are displayed legibly, meaning not changing too quickly, as the rod passes. After passing, the corresponding mean values are displayed.
The 180 individual angularly offset radii must be used every 100 ms to continuously determine the corresponding values. Furthermore, by integration, the cross-sectional surface can be determined via all 180 values and displayed continuously.
The graphs can include mini-statistics (for example on the left) containing the essential characteristics of the last rod. In addition to the maximum, minimum and mean diameters as well as the minimum, maximum and mean out-of-roundness, also the triangularity and hexagon dimensions are displayed.
Apart from this, possible rolling flaws such as underfill, overfill, roll offset and roll break are displayed with symbols.
If following a warning (for example illustrated in yellow), the symbol on the display changes from green (condition O.K.) to red (flaw), the faulty dimension, meaning the similarity percentage relative to the flaw of the underlying sample piece, can be displayed for example by a mouse click.
The underlying flaw thresholds must be accessible by the user and modifiable. The standard thresholds are determined during setup of the system.
Unless it is already provided for in the standard visualization, it is possible to represent the measuring data in the form of trend and bar charts.
It may optionally be required to preprocess the measured radii values before further processing. For this, the methods described below may be used.
By default, interpolation can always be conducted in the case of lapses in measuring values. In principle, however, this interpolation can also be eliminated in the filter options to provide the user this way, for example, with an idea of the quality of the measurement, meaning when deselecting the function in the measuring data progression the user can see the missing areas or lapses.
In principle, in the case of lapses in measuring values the erroneous area can be substituted with
the previous value,
the target value,
the linear interpolation value,
the polynomial interpolation value (optionally), or
the spline interpolation value.
In the case of linear interpolation, only two valid values per line are connected to each other. Spline interpolation may be provided by default.
Due to the continuous formation of short-term mean values over an odd number of values, smoothing or low pass filtration can be easily carried out. In the ultrasound range, this is carried out frequently and referred to as statistical interference elimination.
The computed short-term mean value should always be associated with the interval center. Oscillation processes must be taken into consideration (see also the maximum value filtration of the temperature values).
In this respect it is important, prior to forming a continuous mean value, to completely eliminate lapses in measuring values because otherwise pseudo oscillators are produced in the measuring data progression.
Exponential averaging provides a further simple algorithm for smoothing and/or low pass filtration. The computation of each filter or smoothed value obtained is:
x* _n=(1−q)·x _n +q·x* _n-1
x _n *=q′·x _n+(1−q′)·x* _n-1
resulting in two equivalent definitions. q or q′ are weighting factors between 0 and 1, the values marked with * are smoothed values. The greater q is, the stronger is the filter effect.
One example of the filtration of a step with different weighting factors is illustrated in the sole FIGURE of the drawing as a curve for exponential illustration.
In the present embodiment, Fourier analyses are only applied to complete cross-sections. Since they are always used for a complete revolution, no discontinuous steps that would falsify the result can occur when continuing the “signal” periodically due to the FFT (Fast Fourier Transformation).
For this reason, windowing across the entire circumference is not required with FFT.
Standardizations become necessary when the rod cross-sections are supposed to be compared with the stored samples for defined (rolling) flaws with respect to similarity. One example of this that should be mentioned is the cross-correlation (CCF) between the sample cross-section and the actual rod cross-section.
To obtain comparable similarity dimensions, the sample function that has been standardized to 100% is cross-correlated with the current rod cross-section that has also been standardized to 100%.
After that, the maximum of the cross-correlation is determined. In the case of complete agreement, the result would be 1, meaning 100% similarity.
Instead of cross-correlation (CCF) also the cross-covariance function can be used. The cross-covariance function is the cross-correlation of mean value-free individual variables.
With respect to the analysis of the measured radii, the following should be noted:
The least mean square (LMS) approximations are referred to in literature also as fitting methods or the Levenberg-Marquart method. The basic idea is to vary a model approach approximating the measuring values such that the sum of the distance squares is minimized.
With them, approximation to the rod structure model becomes possible.
The so-called fitting methods are included in countless software toolboxes. In the present case, they can be used for the radii profiles obtained via the angular offset.
The approach for the 3-roll finish block is mentioned above.
For details regarding the LMS fitting methods for the determination of the amplitudes in the above equations, reference is made, for example, to the publication by Sedgewick “Algorithm”.
If as here standard FFT is used for the analyses, the corresponding preprocessing must be carried out before using the method, meaning:

- extraction of suitable value ranges,
- windowing (see also the explanations above),
- zero filling values so that the expanded data set holds second-level values, and
- conversion with respect to order.

For the analysis interval, normally the values of a full cross-section, meaning a full revolution of 360°, are used.
Fourier analyses are common in communications engineering and metrology, however there they are generally applied to time signals. For the analysis of the diameter radii curves, the following analogies must be considered:

- time t
  angle of rotation φ
- frequency f
  order 360°/φ
- signal x(t)
  radius r(φ)
- period T_p
  period angle q, e.g. for third order 3 θ=120°
- frequency f_p
  integer order 360°/θ

The cross-correlation function KKF is defined as
Φ_XY =IFFT(FFT{x(k)}·FFT ^x {y(k)})
The cross-covariance function KCF is basically the same, only that it is applied to mean value-free x(k) and y(k).
The characteristics referred to above, such as similarity values, deviations of the rod surface from the nominal value, characteristics regarding the statistics of radii distribution, for example skew, etc. can be combined into multi-dimensional characteristic vectors.
When assuming initially that each rolling flaw (such as overfill, roll offset, etc.) can be associated with three characteristic values, each characteristic vector for each rod points to a certain point in space.
For rods with sample flaws recorded during the test phase also one point each exists in this three-dimensional characteristic space. If the current characteristic value is close enough to the sample vector, for example for overfill, then the current rod can be classified as defective with overfill.
To find out how great the “proximity” is, a virtual envelope can be placed around the point in space of the flaw vector. Thereafter, basically for each current characteristic vector the distance to the sample flaw vectors is determined. The smallest distance means the highest flaw probability. In most cases, one will remain outside all flaw envelopes and the rod is fine, at least as far as its shape is concerned.
The absolute dimensions are considered separately.
In principle, the entire learning process can also be automated. One method that can be used to accomplish this are neuronal networks.
In any case, quality assurance is guaranteed by the now possible similarity analysis, optionally or preferably supplemented and expanded by an order analysis as well as flaw classification based on characteristic vectors.

Claims

1. A quality-control method for wire or rod in a computer-controlled rolling mill having a succession of roll stands spaced apart in a workpiece-travel direction, the method comprising the steps of:

before rolling a wire or rod, at least one sample wire or rod with typical real roll flaws is measured statically and flaw data in standard format is stored in a control computer, a single cross-sectional shape or, if there is noise, an average of a plurality of averaged cross-sectional shapes being stored standardized as polar coordinates;

during rolling of a wire or rod continuously measuring its actual-shape and generating actual-shape data corresponding thereto;

continuously effecting a similarity analysis by comparing the flaw data in standard format with the actual-shape data of the wire or rod being rolled; and

when the actual cross-sectional shape data corresponds to the flaw data, generating an alarm.

2. The quality-control method defined in claim 1, further comprising the step of:

storing and standardizing the actual-shape data prior to comparing it with the flaw data.

3. The quality-control method defined in claim 1, further comprising the step of

displaying a difference between the actual cross-sectional data and the flaw data.

4. The quality-control method defined in claim 1, further comprising the step of

measuring the temperature of the rod or wire being rolled.

5. The quality-control method defined in claim 4 wherein the actual-shape data is corrected to correspond to a cross-sectional shape the rod or wire would have at about 20° C.

6. The quality-control method defined in claim 1 wherein the rod or wire shape is measured by lasers.

7. The quality-control method defined in claim 1 wherein at least 38 angularly generally equispaced measurements of the rod or wire radius are made to determine its cross-sectional shape.

8. The quality-control method defined in claim 7 wherein 180 such angularly generally equispaced measurements are made.

9. The quality-control method defined in claim 1 wherein the determine radius portions of the wire or rod a Fourier analysis is used.

10. The quality-control method defined in claim 9 wherein a Fast Fourier analysis is used.

11. The quality-control method defined in claim 1 wherein to determine radius portions a fitting method of the actual cross section is used.

12. The quality-control method defined in claim 1 wherein minimum and maximum diameters of the wire or rod are measured at locations along its length to develop the actual-shape data.

13. The quality-control method defined in claim 1, further comprising generating a signal representing ovalness of the rod or wire.

14. The quality-control method defined in claim 13 wherein ovalness is determined by a triangular method.

15. The quality-control method defined in claim 13 wherein ovalness is determined by a hexagonal method.

16. The quality-control method defined in claim 1 wherein the mill includes a set of calibrating roll stands and the wire or rod is measured immediately downstream therefrom.

17. The quality-control method defined in claim 1 wherein the flaw data and actual-shape data are compared by a cross correlation function.

18. The quality-control method defined in claim 1 wherein the flaw data and actual-shape data are compared by a cross covariant function.

19. The quality-control method defined in claim 1, further comprising

collecting flaw data regarding underfilling and overfilling or roller offset are collected and stored.

20. The quality-control method defined in claim 1 wherein for each real flaw at least two references are assigned that are set mathematically as vectors and compared with similar references for the wire or rod.