Figure 1 History of Rod Mill Finishing Speeds
IN-LINE MATERIAL CHARACTERIZATION MEASUREMENTS IN
HIGH SPEED ROD ROLLING MILLS
B. V. Kiefer and P. L. Keyzer
Morgan Construction Company
Worcester, MA 01605 USA
Phone: (508) 755-6111
Fax: (508) 755-6140
e-mail: kieferb@morganco.com and keyzerp@morganco.com
K. L. Schafer, Jr.
B. B. Tech Enterprises / Vision Research
63 Greene Street (204)
New York, NY 10012-4310 USA
Phone: (212) 966-0185
Fax: (212) 334-4833
e-mail: ken@bbtvr.com
ABSTRACT
The ability to provide reliable and timely feedback on the performance of the thermomechanical process in a rolling mill is becoming more important as rolling speeds continue to increase. With rolling speeds now exceeding 120 m/s and production rates exceeding 100 t/h on single strand mills, rapid measurement of the finished product is required to evaluate the process before much unacceptable product is rolled. Conventional means of sample taking, specimen preparation and testing are not only labor-intensive, but also result in relatively slow feedback to the process operators.
This paper describes the application of surface volume imaging technology to on-line measurement of pertinent material properties of steel rod. The system examines the microstructural features of the product, such as grain size, crystallographic texture and phase distribution, and correlates those features with mechanical properties such as tensile strength. Furthermore, surface characteristics of the oxide layer can be evaluated and quantified.
INTRODUCTION
The last three decades have experienced dramatic improvements in the level of technology in ferrous rod rolling mills, resulting in tremendous increases in productivity, efficiency and product quality. Like many other steel product lines, rod rolling mills have made improvements in steelmaking, casting, rolling equipment technology, process monitoring and control. In rod rolling, some literally revolutionary improvements have completely changed the nature of the mill operation and level of product integrity. Demands for increased finishing speeds with better rod quality were the driving force for development of the Morgan No-Twist® Mill (NTM®), coil-forming Laying Heads and Stelmor® Controlled Cooling Systems. Recent developments such as the TekisunTM Reducing Sizing Mill, the Morgan Ring Distributor and mill temperature control have further improved rod dimensional tolerances, final coil presentation and uniformity of product quality.
An example of the results of improved rod mill technology can be seen
in Figure 1, which shows the history of mill finishing speeds and the developments
which have enabled the increases. Not represented there, but of equal
importance, are the increases in product quality and consistency that have
accompanied the higher production rates. Quality is measured in terms
of dimensional tolerance, surface characteristics, mechanical and metallurgical
properties and the uniformity of these parameters throughout a coil containing
as much as 10 km of rod.
Figure 1 History of Rod Mill Finishing Speeds
To take full advantage of the mechanical innovations and insure consistency of the product, more sophisticated monitoring, control and management systems have also dramatically improved for rod mill operations. A key element in these systems has been sensor technology, for monitoring of the entire process, including conditions in the reheat furnace, rolling mill equipment, and fluid systems, plus process section dimensions, processing temperatures and speeds.
Although the process has become very controllable, there is still a
weakness with the inability to include accurate evaluation of final material
properties in the process control loop. For all of the increases
in equipment technology making high speeds possible and systems to keep
it in control, FINAL product evaluation still relies on time-consuming,
decades-old methods of manual sample cutting, preparation and testing using
an off-line laboratory to provide critical feedback to the process.
Recent developments in sensor technology, accompanied by computational
implementation of real-time adaptive algorithms now enables direct evaluation
of mechanical properties in the process line or immediately after processing
to help close that loop.
METALLURGICAL QUALITY AND THE ROD MILL
Rod Products
Coiled product from a rod rolling mill is typically subjected to further forming operations resulting in various other finished and semi-finished products, which number literally in the thousands. For many of these products, it is very critical for the as-rolled rod to have the correct microstructure and surface scale condition, as well as dimensional tolerance and freedom from defects.
Of equal importance is the uniformity of the microstructure and scale conditions along the length of the rod. Proper microstructure and consistency are of utmost importance in critical products such as cold heading grades (e.g., for fasteners), spring steels, bearing steels and grades subjected to severe wire drawing requirements like tire cord, welding wire and fine wire applications. During the drawing process, non-uniformity of microstructure and mechanical properties can result in breakage and excess wear, resulting in costly down-time in the drawing operation, or variability in the finished product.
Microstructure control is of importance also for rod products requiring
post-rolling heat treatment, such as spheroidize annealing, in which the
degree of spheroidization throughout the material is directly proportional
to the as-rolled grain size distribution. Critical applications such
as bearing quality steels depend on the uniformity of the microstructure
for optimization of the finished product.
The Rod Mill
A rod rolling mill, depicted schematically in Figure 2, consists of a series of rolling stands which gradually reduce the cross-sectional area of a starting billet down to a finished rod as small as 5.0 mm in diameter at finishing speeds up to about 120 m/s. At various positions in the rod mill, water cooling units control the temperature of the steel, to achieve the desired thermomechanical history, maintain a laying (i.e., coil-forming) temperature, and control high temperature austenite grain size and surface scale conditions. Good control of the process has been achieved through the implementation of sophisticated equipment control strategies and extensive stock temperature monitoring through the mill line, including both feedback and feedforward control loops. The approach has been to control the process using temperature as the critical variable effecting final properties.
Figure 2 Example of Equipment Layout in a High Speed Rod Mill
Figure 3 Finishing End of a Rod Mill
Control of the process and therefore of the metallurgical properties becomes more difficult after the laying head, where the rod is decelerated and spread onto a roller conveyor in a spencerian ring pattern (see Figure 3). Conditions on the conveyor control cooling of the rod through its transformation from high temperature austenite to ambient temperature products. Difficulty achieving uniformity of mechanical properties and metallurgical structure in rod production comes about as a result of the physical configuration of the cooling conveyor. Figure 4 is a representation of the rings on the conveyor, showing the overlapping pattern. Cross-over points where the rings touch one another, plus increased density of material at the edge of the conveyor result in the difficulty achieving uniform cooling and thus uniformity of metallurgical and mechanical properties. To compensate for these geometry factors, the distribution of cooling air is varied across the width of the conveyor, concentrating more air at the edges than in the center. Good air distribution, coupled with slight shifting of the rings as they proceed along the conveyor enable good uniformity to be achieved.
Figure 4 Plan View of Stelmor® conveyor with Rod Rings
Product Testing
As with other continuous and semi-continuous forming operations, the mechanical and metallurgical properties must be determined off-line, by conventional sample testing. Typically, a small sample is cut from the finished coil after trimming of discard rings and then subjected to a destructive tensile test for determination of ultimate tensile strength and reduction of area. Depending on the product and customer requirements, metallographic and scale weight tests may also be conducted.
These conventional testing procedures have several significant disadvantages. First, they require costly manpower for sampling, testing and evaluation. Second, consistency in testing is dependent on the skill of the operator and will vary with different workers. Third, and most significant is the time lag in the process, since the mechanical test is the feedback in the manual control loop of the process. Therefore, if some variable in the controlled cooling process has been inadvertently changed, the possible detrimental effect on properties would be not be immediately detected, resulting in perhaps many tons of out-of-spec product.
Testing for property uniformity is usually based on sampling coil-to-coil
and therefore is not representative of the around-the-ring uniformity which
is indicative of the correct setup of the controlled cooling line.
Around-the-ring testing is done in cases of customer complaints about uniformity
or an unusual case of cooling system evaluation.
ON-LINE PROPERTY MEASUREMENT
In order to ensure that the rod properties are well-controlled and uniform, it is necessary to implement a closed-loop control system using actual product and not just temperature feedback. An essential step in achieving this advancement is the development of reliable sensors to non-destructively monitor the mechanical properties, metallurgical structure and surface scale condition of the rolled product.
Various sensor technologies have been under development in recent years for this purpose, including laser ultrasonic and electromagnetic sensors. Laser ultrasonic sensor technology is capable of making both microstructural and final mechanical properties predictions from the propagation and/or attenuation of ultrasonic pressure waves through the entire cross section of the steel microstructure (Kavanagh, 1996). Ultrasonic sensor technology is not sensitive to chemical differences in steel alloys. Electromagnetic technologies utilizing measurement of quality factor (Q) in a radio frequency (rf) resonator inductively coupled to a ferritic steel product are sensitive to the complex interaction of chemical and microstructural factors in the alloy. Electromagnetic technology is limited to evaluation of a surface volume in the ferritic microstructure, the depth of which is determined by the resonant frequency of the inductive-capacitance (LC) circuit. This feature of the technology can be used to selectively measure properties of microstructural development such as grain size at different depths in the material or focus on characteristics of the surface scale.
MEASUREMENT METHOD
The Electromagnetic Sensor
An electronic resonator was employed for non-contacting, nondestructive
measurement of microstruture in ferritic steel alloys. The quality
factor of an inductive-capacitance tank circuit was determined to obtain
a measure of energy dissipated in a steel sample inductively coupled to
it. Dissipative energy in the sample results from two separate sources
- hysteresis and eddy currents. Of the two, the eddy current loss
is relatively microstructure insensitive and is mainly related to chemical
composition and temperature, while the hysteresis loss is predominately
determined by microstructural factors. As an example, if microstructural
variables other than grain size and the factors determining eddy current
loss are held constant, grain size becomes the dominant factor in differential
energy loss with grain size being inversely related to hysteresis loss.
Under these conditions, grain size is inversely related to total energy
dissipation and thus the measured Q of the resonant circuit (Schafer, 1995).
The Surface Volume Image Rendering Engine
Honed by millions of years of iterating the primordial genetic algorithm, biological life forms with advanced sensory systems have managed to develop mechanisms that take the highly selected and filtered information from their sensory organs and transform it into a useful model of the real world. This model, which is constructed entirely from within the organism is none the less experienced as being outside the skin that defines the physical bounds of the organism, (von Békésy, 1967). In advanced sensory systems constructed from electronic components the goal is the same as for biological organisms: take a carefully selected amount of sensory data from the total pool of momentarily available information and transform it into a useful model of the universe within which the system is supposed to function.
 |
 |
 |
 |
The objects seen above are two-dimensional visual stimuli that get focused on a two dimensional retina in your eyes, but they are perceived as having volume and/or being located somewhere in a three dimensional space. For organisms that are capable of communicating the qualities of their perceptions, this kind of treatment of raw sensory information has proved to be both a very personally convincing and very useful construction for operating effectively in the real world.
An advanced electronic sensory system was constructed that can measure non-destructively, and at a distance, the mechanical properties of ferritic steels within the confines of a rod mill production line. The design of this system exploits one of the oldest and most pervasive sensory system constructions of the biological genetic algorithm, lateral inhibition. A neural network or plexus that neurophysiologically implements the process of lateral inhibition was first discovered by Hartline (1949) in the compound eye of the 400 million year old horseshoe crab, Limulus polyphemus. In their most basic form lateral inhibitory mechanisms take spatio-temporal signals from one element in a two dimensional array and inhibit or suppress the strength of responses from other spatially contiguous elements in the array. The result of this process is an enhanced signal-to-noise ratio or contrast function for stronger signals in the array at the expense of weaker ones.
In its present form the lateral inhibitory process is generated by a microprocessor implemented genetic algorithm that takes as its input, spatio-temporal information from a radio frequency (rf) sensor array. The elements in the sensory array are LC resonators that are inductively coupled to ferritic steel rods moving past the array (Schafer, l995). In the case of the Limulus eye, the general dynamic response function of the neural plexus has been fixed by 400 million years of evolution (the response functions are, however, context sensitive; i.e. they can adapt to average available light levels to maintain optimal sensitivity and selectivity) (Hartline et al, 1961, 1956). In some insect species mechanical tuning of the sensory system can additionally complement neural adaptation to changing light levels (Schafer, 1978). In an electronic sensory network, the specific dynamic response function of the lateral inhibitory process is constructed through interaction with inductively coupled ferritic steel rods of varying mechanical and chemical properties on the production line. The goal or purpose of the genetic algorithm is to construct a lateral inhibitory process that gives a good and therefore useful prediction concerning the final mechanical properties of specific rod product. The fitness test for all constructions is degree of correlation with destructive mechanical properties testing on the finished product for the specific mechanical property of interest. Metallographic correlations can also be employed as a fitness criterion by the genetic algorithm.
For situations that involve new steels never seen before, the genetic
algorithm is seeded with expert knowledge, contextual information, and
stochastic distributions (Hinton et al 1995, 1996) to compress or expand
the search space for suitable lateral inhibitory processes as needed.
Contextual information is empirically derived information from mill process
control databases such as melt codes, mill setups, and temperatures (Schafer,
1996). A fully adapted lateral inhibitory process is not re-seeded
unless if fails to yield a criterion level of predictive utility.
LABORATORY TESTS ON ROD SAMPLES
Samples of plain low carbon steel rod were randomly gathered from coils rolled as part of a normal production sequence. No special preparation of the samples was required for the scanning. Measurements were performed in a laboratory setting to simulate a possible means by which the measurements would be performed in the rolling mill.
Shown in Figure 5 below are the ultimate tensile strength (UTS) predictions from the integrated electronic sensory and perceptual rendering engine measurements at various positions around the circumference of a single ring of a low carbon steel grade, identified here as Low Carbon 1. Initially, three measurement runs were made on the same ring in order to provide the genetic algorithm with some learning experience. The positions are 150mm (6) apart. One feature of the measurements which is evident in this figure is that locations on the ring which have peaks show up in the measurements consistently as peaks, whereas the intermediate values show more effects of noise and do not replicate consistently.
Next, a fourth run was made for another complete set of measurements around the ring, using the fully adapted process. The results, shown in Figure 6, show clearly that the system detects two definite peaks in the ring. Comparison with Figure 5 indicates that the peaks fall at the same locations as the two consistently high areas on the initial runs.
Figure 5 Rod measurements on initial runs
Figure 6 Runs with Fully adapted system
Figure 7 Tensile tests results from rod samples
The rod rings were then cut into small samples of a length suitable
for conventional tensile testing and sent to an independent laboratory.
The results of the tensile tests are shown in Figure 7. Although
a comparison with Figure 6 indicates that there is only a weak correlation
between the measurements, the differences in the two graphs highlight an
important aspect of both measurements. In the case of the destructive
tensile tests of Figure 7, the variation of values is relatively small
and happens to be within the measurement accuracy of the tensile testing
machine. The non-destructive tests presented in Figure 6 are considerably
more sensitive to property variations around the ring.
In addition to the mapping of mechanical properties for the grade Low
Carbon 1, an additional map was made on a ring of another plain carbon
grade with somewhat higher carbon content, identified here as Low Carbon
2. The Q measurements made with the fully adapted lateral inhibitory
process show two peaks in the same relative ring position as seen in the
final ring mapping of grade 1. There is, however, a notable offset
for the absolute Q values between corresponding extrema of the two plots
as seen in Figure 6. This difference is due to the interaction of
the chemistry and final microstructure, which correlates with destructive
tests of mechanical properties. In situations where the comparison
is across grade, the interaction of chemistry and microstructure is quite
complex and the expected relationship between mechanical properties measurements
and Q measurements would be a nonlinear one. Since there was only
one across-grade comparison made, the current experiment does not provide
any information as to the degree of the nonlinearity. For measurements
within a particular grade, previous experiments on steel strip have shown
a linear relationship between Q measurements, microstructural measurements
and mechanical properties as determined by destructive testing over a wide
range of grades and product types (Schafer, 1996).
MILL INTEGRATION AND SENSORY NETWORKS
Strategies for mill process monitoring and control are continually evolving to take advantage of developments in microprocessors, networking and new open system technologies. More reliable sensors such as loop scanners, dimension gauges, flaw detectors and pyrometers in the mill environment, coupled with more cost-effective computing power now enables large volumes of information to be gathered, arranged and presented to the mill operator. A hierarchical structure of various conventional control and monitoring function in a rod mill is shown in Figure 8.
The various parts of such an all-encompassing mill monitoring and control network are continually conducting on-line micro-experiments at different stages of the process. Unfortunately, the results of these valuable experiments can only be collected and stored for future reference in case of questionable product results. Due to subtle interactions of variations in melt chemistry, process parameters (i.e. mill set-up) and temperature control, some obviously important dependent variables in the processes go unobserved, unrecorded, or not connected in functional relationships to their causal antecedents.
In order for the system to be in total control of the process, a sensory network with inputs from the melt chemistry to inspection of the final product must be integrated throughout the entire production process. The effectiveness of this technology is limited only by the quantity and quality of process relevant data provided to the system. If this knowledge based system is provided with high quality information accumulated from the vast amount of data waiting to be collected in the mill, the steel producer can be virtually assured of being able to offer his customers high quality steel, produced under optimum conditions in the mill.
Figure 8 Structure of mill control and monitoring functions
SUMMARY
The system described here would incorporate innovative sensing technology
with an inherent means not only to quantify but also to learn from the
final product properties. It will add an essential piece of information
to the process, regarding both internal and superficial condition of the
finished product which, when using existing technology, is either not collected
or not available on a timely basis.
The advantage of on-line measurement will complete the fully integrated
intelligent control system and allow highly competitive mills to consistently
push the technological margin of their equipment, producing better products
using more robust processes. As this system is further tested and refined,
it will be integrated into mill control structures for the next generation
of knowledge-based intelligent control over critical operations in high
speed rod rolling mills.
A smart mechanical properties inspection system for steel products should include the following design guidelines for optimal properties measurement resolution:
REFERENCES
BEKESY, G. von, 1967.
Sensory inhibition. Princeton University Press, Princeton, New Jersey.
HARTLINE, H. K., RATLIFF, F. and MILLER, W. H., 1961.
Inhibitory interaction in the retina and its significance in vision.
In E. Florey (Ed.) Nervous inhibition, Pergamon, New York, pp. 241-284
HARTLINE, H. K., WAGNER, H. G. and RATLIFF, F., 1956.
Inhibition in the eye of Limulus. Journal of General Physiology,
Vol. 39, pp.651-673
HARTLINE, H. K.,1949.
Inhibition of activity of visual receptors by illumination of nearby
retinal elements in the Limulus eye. Federation Proceedings, Vol.
8, p. 69
HINTON, G. E. and FREY, B. J., 1996.
Using neural networks to monitor for rare failures. 37th Mechanical
Working and Steel Processing Conference Proceedings, ISS, Warrendale, Pennsylvania,
Vol. 33, pp. 545-548
HINTON, G. E., DAYAN, P., FREY, B. J. and NEAL, R., 1995.
The wake-sleep algorithm for self-organizing neural networks.
Science, Vol. 268, pp. 1158-1161
KAVANAGH, L. W., 1996.
AISI Testimony to the Subcommittee on Interior and Related Agencies
of the Committee on Appropriations, U.S. House of Representatives, March
7, 1996.
SCHAFER, K. L., 1996.
Genetic algorithms for optimizing and automating process control in
critical-cycle continuous annealing. 37th Mechanical Working and Steel
Processing Conference Proceedings, ISS, Warrendale, Pennsylvania, Vol.
33, pp. 611-613
SCHAFER, K. L., 1996
Presentation of MicroCert Research and Development Program
to the American Iron and Steel Institute, Advanced Process Control Program,
Project D, Online Mechanical Properties, I.M.I. Boucherville, Quebec, Oct.
10, 1996.
SCHAFER, K. L., 1995.
A sensor and method for the in-situ monitoring and control of microstructure
during rapid metal forming processes. U.S. patent #5,420,518.
SCHAFER, K. L., 1978.
Receptor sensitivity and pigment migration to ionizing radiation and
light in the noctuid moth. In Armington, J. C., Krauskopf,
J., and Wooten, B.R.(Eds.) Visual psychophysics and physiology.
Academic Press, New York.