Metal deforming – With use of control means energized in response to activator... – Metal deforming by use of roller or roller-like tool element
Reexamination Certificate
2001-09-24
2004-04-20
Tolan, Ed (Department: 3725)
Metal deforming
With use of control means energized in response to activator...
Metal deforming by use of roller or roller-like tool element
C072S007100, C706S016000, C706S023000
Reexamination Certificate
active
06722173
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to a method and a device for determining the rolling force in a roll stand for rolling metallic material to be rolled, the rolling force being determined by means of at least one neural network which is trained by using measured values for the rolling force under different operating conditions with a view to improving the determination of the rolling force.
BACKGROUND OF THE INVENTION
It is known to determine the rolling force by means of a rolling-force model which has a neural network. DE 197 27 821 A1 describes a method and a device for controlling and/or presetting a roll stand or a rolling train as a function of at least the rolling force. The rolling force is determined by means of a rolling-force model, and at least one neural network. To improve the quality of the rolling-force model, the neural network is trained while a rolling train which it is intended to control is operating. In this context, the term quality of model is to be understood as meaning the deviation between the actual rolling force and the rolling force which is determined by means of the rolling-force model. To achieve a high quality of model, this deviation should be as low as possible across the entire range of metals which are to be rolled by means of the roll stand.
SUMMARY OF THE INVENTION
It is an object of the present invention to increase the accuracy of the determination of the rolling force in a roll stand for rolling metallic material which is to be rolled compared to the known method. Accordingly, the rolling force in a roll stand for rolling metallic material to be rolled is determined by means of at least one neural network which is trained, using measured values for the rolling force under different operating conditions, and thereby improving the determination of the rolling force. The neural network is preferably trained using values for the rolling force and values for the different operating conditions for roll stands of different rolling trains.
The term operating conditions is to be understood as meaning, for example, the width of the metallic material which is to be rolled by means of the roll stand before it enters the roll stand, the thickness of the metallic material to be rolled before it enters the roll stand, the relative reduction in thickness of the metallic material which is rolled in the roll stand, the temperature of the metallic material to be rolled when it enters the roll stand, the tension in the metallic material which is to be rolled upstream of the roll stand, the tension in the metallic rolled material downstream of the roll stand, the radius of the working rollers of the roll stand, and the levels of iron, carbon, silicon, manganese, phosphorus, sulfur, aluminum, copper, molybdenum, titanium, nickel, vanadium, niobium, nitrogen, boron and/or tin in the metallic material to be rolled and, if appropriate, the modulus of elasticity of the rollers.
Although in known methods for determining the rolling force in a roll stand using a neural network, the rolling force is specially matched to the roll stand by training, the present invention leads to a more precise determination of the rolling force for the roll stand across the entire range of metals which are to be rolled by means of the roll stand.
In a preferred embodiment of the present invention, the neural network is trained using values for the rolling force and values for the different operating conditions for at least one roll stand from a roughing train and at least one roll stand from a finishing train.
Other preferred embodiments of the present invention are disclosed herein below, the first of which is wherein the neural network is trained using values for the rolling force and values for the different operating conditions for roll stands of rolling trains of different rolling mills. This embodiment leads to particularly accurate determination of the rolling force in a roll stand, even though roll stands of different rolling mills, i.e. different plants, are used for training.
In the next embodiment, the neural network is trained using values for the rolling force and values for the different operating conditions for roll stands of rolling trains of at least five different rolling mills. In this embodiment, the accuracy of determination of the rolling force can be increased even further.
In the next embodiment, the neural network is used to determine a correction value for correcting (e.g., by multiplication) a value for the rolling force which is determined by means of an analytical rolling-force model. The linking of analytical models and neural networks is disclosed, for example, in DE 43 38 607 A1, DE 43 38 608 A1 and DE 43 38 615 A1. In particular, a structure as described in DE 43 38 607 A1 has proven particularly advantageous in connection with linking by multiplication.
In a further preferred embodiment, a stand-specific correction value for correcting the value for the rolling force which is determined by means of the analytical rolling-force model is determined by means of a stand network. The stand network is in the form of a neural network, as a function of physical properties of the metallic material which is to be rolled (e.g. the width of the metallic material which is to be rolled by means of the roll stand before it enters the roll stand, the thickness of the metallic material which is to be rolled before it enters the roll stand, the relative reduction in thickness of the metallic rolled material in the roll stand, the temperature of the metallic material which is to be rolled when it enters the roll stand, the tension in the metallic material which is to be rolled upstream of the roll stand, the tension in the metallic rolled material downstream of the roll stand and the radius of the working rollers of the roll stand, and of physical properties of the roll stand (e.g. the modulus of elasticity of the rollers).
A further advantageous embodiment of the present invention utilizes a chemistry-specific correction value for correction of the value for the rolling force. The rolling force is determined by means of the analytical rolling-force model. The correction value is determined by means of a chemistry network. The chemistry network is in the form of a neural network, as a function of chemical properties of the metallic material which is to be rolled (e.g. the levels of iron, carbon, silicon, manganese, phosphorus, sulfur, aluminum, chromium, copper, molybdenum, titanium, nickel, vanadium, niobium, nitrogen, boron and/or tin in a metallic material to be rolled).
This division into a stand-specific correction value and a chemistry-specific correction value, in conjunction with the invention, leads to a further improved accuracy in the determination of the rolling force. Further, under certain operating conditions of the roll stand, the accuracy of determination of the rolling force is significantly improved if, in combination with the invention, a microstructure-specific correction value for correction of the value for the rolling force which is determined by means of the analytical rolling-force model is determined by means of a microstructure network, which is in the form of a neural network, as a function of chemical properties of the metallic material which is to be rolled and the temperature of the metallic material which is to be rolled. In this context, chemical properties are, for example, the levels of iron, carbon, silicon, manganese, phosphorus, sulfur, aluminum, chromium, copper, molybdenum, titanium, nickel, vanadium, niobium, nitrogen, boron and/or tin in the metallic material which is to be rolled. This applies in particular if a distinction is drawn between a stand-specific correction value, a chemistry-specific correction value and a microstructure-specific correction value.
In yet another preferred embodiment of the present invention, the correction value of at least one neural network is multiplied by a confidence value. The confidence value forms a statistical measure for the reliability of the correction value. The confide
Broese Einar
Feldkeller Bjoern
Peuker Thomas
Poppe Thomas
Baker & Botts LLP
Siemens Aktiengesellschaft
Tolan Ed
LandOfFree
Method and device for determining the rolling force in a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and device for determining the rolling force in a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and device for determining the rolling force in a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3258321