Data processing: generic control systems or specific application – Specific application – apparatus or process – Product assembly or manufacturing
Reexamination Certificate
2000-07-14
2004-08-24
Picard, Leo (Department: 2125)
Data processing: generic control systems or specific application
Specific application, apparatus or process
Product assembly or manufacturing
C700S032000, C700S089000, C702S105000
Reexamination Certificate
active
06782304
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a learning control apparatus for a reversing rolling mill that rolls strip, etc. using the reversing method.
2. Description of the Related Art
In a rolling mill that rolls strip, etc. using the reversing method, highly accurate maintenance of the set up values in each pass is essential for stable operation, and set up calculation systems and learning controls that use various mathematical models are widely employed for that purpose. In the case of rolling using the reversing method, the set up calculation is generally executed by a process computer; an optimum pass schedule calculation for the strip delivery thickness for each pass is carried out for processing optimisation, and the various set up values required in the operation of the mill for each pass are calculated. For example, such settings as the roll gap for each pass, the strip threading speed, the running speed, the tail-out speed, and the side guides and roll coolant flow are calculated as these setting values, and these data are transmitted to lower level controllers.
In order to achieve stable operation at such a time, highly accurate prediction of the various quantities by mathematical models is a basic requirement. However, actual rolling operations are affected by all kinds of disturbance
factors due to intermittent conditions. Thus, however carefully constructed a mathematical model may be, it will not be possible completely to grasp the actual conditions and numerically to express them.
Therefore, in order to reflect in model prediction various disturbance factors such as disturbances that vary in a time series fashion, learning control is generally applied to the mathematical models. With relation to learning control for reversing rolling mills, various methods have been proposed up to the present time, and each has had some effect.
For example, a “Plate Rolling Method” that rolls while adding corrections so that the settings will be optimum for the next pass by measuring the actual values at each pass midway through the rolling pass and making learning
calculations using those results has been disclosed in Laid-Open Patent No. Heisei 7-60320 Gazette. In the method stated in this Gazette, since rolling is continued while measuring actual values midway through a rolling pass and sequentially adding corrections so that the set up for the next pass are optimum, the measurement results of the actual values for each pass in rolling the relevant strip can be reflected in the calculation of the setting values for the next pass. However, there is the problem that errors that are not sequentially present in a pass in time series fashion cannot be reflected in the rolling of the next strip.
Also, a “A Reversing Type Rolling Method That Excels In Configuration And Strip Thickness Control” that, for the second pass and thereafter, performs rolling while repeating the learning, during or immediately after the rolling of each pass, of corrections to the pass schedule up to the final pass based on load forecast expressions learned from the previous pass, in almost the same way as in the above Gazette, has been disclosed in Laid-Open Patent No. Heisei 8-243614 Gazette.
With this Gazette also, learning calculations can be performed based on the calculation results of the actual values for each pass and these results can be reflected in the settings for the next and subsequent passes. However, there is the problem that, even though they may be time series-wise, errors that do not continue in every pass cannot always appropriately be taken into consideration.
At the same time, methods are disclosed in Laid-Open Patent No. Heisei 2-137606 Gazette and Laid-Open Patent No. Heisei 4-367901 Gazette that assimilate model errors that depend on the material being rolled or processed, its target dimensions, etc. out of changes of state that are not time series dependent by storing them in tables prepared group by group. However, even though intrinsic model errors that depend on the material and target dimensions can be assimilated by these methods, this is nothing more than the assimilation of errors as statistical results, and there is the problem that time series errors that do not sequentially depend on passes cannot be assimilated.
As explained above, with the methods disclosed in the above several Gazettes, it is possible effectively to learn model errors that arise sequentially in each pass and intrinsic model errors dependent on materials, target dimensions, etc. when using each method, and there may be cases when product qualities such as strip thickness, strip crown and flatness are well maintained. However, there was the problem of how to assimilate factors other than the above-mentioned model errors, that is to say errors that, though they are time series type errors, are not sequentially dependent in each pass, in order further to maintain stable model forecast accuracy and improve product quality.
The reasons why a satisfactory product quality cannot be achieved solely by assimilation of model errors arising sequentially in each pass and intrinsic model errors depending on the material and the target dimensions using learning calculations for each individual error, as mentioned above, are as follows.
That is to say, a pass schedule in a reversing rolling mill composed of 1~N passes, taking N as an integer of 2 or more, is generally made up of several parts such as initial stage passes—intermediate passes—final stage passes or rough passes—finishing passes, and the essential points for the operations in the passes pertinent to each part differ. For example, with the initial stage passes, the pass schedule is determined so that the roll force is increased for a few passes in order to improve productivity while, conversely, with the final stage passes, it is normal to correct the pass schedule in order to satisfy different aims from those of the initial stage passes, such as ensuring the surface quality of the product.
Consequently, for example, operating conditions such as the reduction (percent draft) rate limits for each pass differ in each part and, naturally, the behavior of the model errors will differ according to those parameter limits. Therefore this means that, with learning calculations that are performed at each relevant pass, the problem will remain that it is not possible satisfactorily to assimilate time series model errors. In fact, since variations with passage of time, such as roll surface state, appear as behavior such as the coefficient of friction gradually varying in time series fashion with the progress of a pass, a method that reflects in the model calculation for the next pass the results of learning by measurement of actual values in a pass will be effective. On the other hand, since the behavior of model errors for material deformation resistance and the like is not simple, cases will often be observed in which the results of time series-wise learning calculations using the actual values of the previous pass do not necessarily operate toward assimilation of the model errors of the next pass.
Also, while on the one hand it is possible, by taking some amount of time, to assimilate the medium-term and long-term fluctuations of model errors by the method of storing the results of learning calculations in tables divided into groups for every material and product dimension, it is not possible to assimilate the minute fluctuations of model errors that occur in successive products in the operations of one day. That is to say, in reversing rolling, rolling is based on pre-stored pass schedules or based on optimum pass schedules that are generated by logic. If, at that time, there are not sufficiently many opportunities for renewing one by one the learning calculation values that belong to the same group division, and also if group division tables that are subdivided to that extent are not prepared, there will, conversely, be many cases when the fluctuations of model errors in the medium term or long term cannot be stabil
Hioka Masaki
Sano Mitsuhiko
Tsugeno Masashi
Yamaguchi Naoki
Foley & Lardner
Kabushiki Kaisha Toshiba
Picard Leo
Rao Sheela S
LandOfFree
Learning control apparatus for a reversing rolling mill does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Learning control apparatus for a reversing rolling mill, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Learning control apparatus for a reversing rolling mill will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3305542