A METHOD FOR PREDICTING COBBLE GENERATION IN A HOT STRIP MILL

FIELD OF THE INVENTION

The present invention relates to methodology of prediction of cobble generation during hot strip rolling using signal analysis techniques of mill signals.

BACKGROUND OF THE INVENTION

Hot strip mills in steel plants rolls slabs to coils in number of stands. Schematic diagram of a typical hot strip mill is shown in Figure-1. It comprises 3 Roughing stands and 6 Finishing stands. The first roughing stand is a combination horizontal stand and a vertical stand. The other two roughing stands are 4-hi horizontal stands. There is a delay table after last roughing stand, one coil box and a crop shear at the end of the delay table. There are six numbers of 6 high finishing stands. The mill rolls are rotated by mill motors. There are different field sensors in the mill to record different signals during rolling. Signals like roll force, roll gap, looper angles, motor speed, motor field current and motor armature current are recorded from the field sensors by Programmable Logic Control (PLC) systems installed in the mill.

Cobble generation is an important operational problem in the mill. Due to various operational reasons, the coils are struck in the mill stands and become cold. The rolling operation is halted for some time. The cobbled coil is cut into pieces and taken out of the mill. There is loss of material and precious rolling time due to cobble generation.

The art describes the methodology of prediction of cobble generation in the mill using signal analysis technique.

OBJECTIVE OF THE INVENTION

The objective of this invention is to develop a system which can predict the generation of cobble before entry of material into the mill stand using signal analysis. When the cobble generation is detected before the entry of coils into the mill stand, corrective actions are taken by the mill operator resulting in saving of precious operational time.

SUMMARY OF INVENTION

Therefore such as herein described there is provided amethod for real time prediction of cobble generation in a hot strip mill comprising thesteps of:
capturing of the data for motor speed signals - 6 signals, motor field current - 6 signals, motor armature current-6 signals, roll force -6 signals, roll gap -6 signals, inter stand Looper angle – 5 signals in real time of the operation of the rolling mitt employing a plurality of sensors placed at strategic locationsand classifying the said mill signals into a plurality of groups using limit values calculated from first quartile, 3rd quartile and inter quartile range of previously recorded signals.

The methodology of the invention primarily uses Support Vector Machine (SVM), a Machine Learning algorithm, to classify mill signals into 2 groups: Cobble and Non-Cobble. Machine Learning algorithms work in 2 stages: Training Stage and Simulation Stage. In the training stage, both input and output data are provided to the program. It updates its internal parameters. In the simulation stage, when input data is given, it predicts output. In the present case of cobble prediction, 35 nos. of mill signals were identified which affects generation of cobble in a hot strip mill having 6 finishing stands. Past recorded signal data of these 35 signals of cobble and No-cobble were used to train the support vector machine program. After the program trained itself, it is able to predict the formation of cobble for coils. A program code was written to display the SVM output to the operator on real time.

BRIEF DESCRIPTION OF ACCOMPANYING DRAWINGS

Fig. 1 illustrates schematic diagram of a typical hot strip mill wherein the cobble prediction system is installed in accordance with the present invention;
Fig. 2 illustratesa typical signal plot showing the effect of signal on cobblein accordance with the present invention;
Fig. 3 illustrates the principle of statistical classificationin accordance with the present invention;
Fig. 4 illustratesan example signal (Roll Force of Stand F1) for statistical classificationin accordance with the present invention;
Fig. 5 illustrates principle of classification of Cobble and No-Cobble using Support Vector Machine (SVM) Techniquein accordance with the present invention;
Fig. 6 illustrates Data Flow Diagram of Online Cobble Prediction System in accordance with the present invention;
Fig. 7 illustrates Output Screen for display of output to Operator into 3 groups Healthy (No-Cobble), Unhealthy (Cobble) and Normal (No-Prediction) in accordance with the present invention;

DETAILED DESCRIPTION OF THE INVENTION

The primary objective of a rolling process is to reduce the thickness of incoming material by plastic deformation. If the rolling process is carried out at a temperature above its re-crystallization temperature, then the process is known as hot rolling. If the cross section of the final material after rolling is rectangular, then it is called flat rolling and if the cross-section is circular, I-section, L-Section or any other irregular section then it is called shape rolling. Hot Strip Mills are one kind of hot and flat rolling process in which the incoming slabs of thickness about 200-300 mm are rolled into strips of thickness about 1.5-10 mm. A slab is first rolled in roughing stands then it passes through a delay table and finally into the finishing stands as shown in Fig. 1. The coil box and crop shear are located at the end of the delay table. A crop shear is used to cut the head end and tail end of transfer bar. The shear is located after coil box and before entry to finishing stand. After finishing stands, Strip passes through Run out table and then finally goes to Coiler.

Fig. 2 shows a typical signal plot recorded at the Hot Strip Mill to illustrate how the formulation of cobble is manifested in mill signals. In this figure, seven mill signals are plotted on real time. High values of signals indicate that there is material in the mill. In this diagram, there are three phases where the mill signals are high indicating that rolling of three coils. There was no abnormality in any signal during the rolling of first coil. Before rolling of second coil, the Looped Angle Signal was found to be very high, may be due to malfunctioning of looper. But, the mill operator was unable to detect it. But fortunately, the coils were rolled in healthy condition. But this aggravated the looper problem leading to cobble in the third coil. If the operator had detected the signal beforehand and corrective measure was taken in the looper, then the 3rd coil could have been saved for forming a cobble. It is a very difficult job for operators to monitor all signals manually. So, this art will help them to instantly identify problem areas and take corrective actions for saving cobble.
The present method monitors following 35 mill signals on real time basis for prediction of cobble for 6 finishing stands:
• Motor speed signals - 6 signals
• Motor field current-6 signals
• Motor armature current - 6 signals
• Roll force - 6 signals
• Roll gap - 6 signals
• Inter stand Looper angle – 5 signals

The classification of mill signals is done in two stages: Statistical classification and Support Vector Machine (SVM) classification. The principle of statistical classification is shown in Figure-3. Each of the 35 signal data of previously rolled coils (more than 10000 coils) are statistically analysed by the disclosed method. It calculates the median value, first quartile and third quartile of the data. The Inter Quartile Range (IQR) is determined by subtracting the first quartile value from the third quartile. It calculates one upper limit value of Q3+1.5 * IQR and a lower limit value of Q3-1.5 * IQR. If the signal value of the next recorded data falls outside the limits, those signals are considered as unhealthy which may lead to a possibility of cobble.

A typical signal data, signal of roll force of Stand F1 is shown in the Fig. 4 for illustration. It was calculated that the Q1, Median, Q3 values of the previously recorded roll force signal data was calculated as 16, 18 and 20 MN respectively. The algorithm decides that if the roll force signal is above 26.87 MN or below 8.87, then the case will be considered as a case of cobble and further analysis will be done by SVM technique.

The principle of Support Vector Machine (SVM) based classification of signals is shown in Fig. 5. The SVM is trained in off-line from previously recorded signals of Cobble and No-Cobble cases. During the training, it first converts the signal from time domain to frequency domain using Fast Fourier Transformation (FFT). An optimum hyper plane and two support vector planes for Cobble and No Cobble cases are identified and the values are stored in the online prediction system as disclosed herein. When the statistical technique identified a signal for a possible case of cobble, the SVM further analyses it using the previously stored support vector values and finally predicts the possibility of cobble.
The Data Flow Diagram of the On-line Cobble Prediction System is shown in Fig. 6. The signals are obtained from PLCs and other primary data like coils ID and steel grade are obtained from VAX and ERP systems. The program analyses the data, make online statistical analysis and SVM analysis (when required) and send the prediction output to operator for display.

A typical screenshot of online predictive model output is shown in Fig. 7. It shows three cases of health status to mill operators: Healthy (No-Cobble), Unhealthy (Cobble) and Normal (No-Prediction) in color code for quick attention. The critical signal values are also displayed to operator.

Inventive Step

(i) Classification of mill signals to 3 groups: No-Cobble, Cobble and No-Prediction using limit values calculated from first quartile, 3rd quartile and inter quartile range of previously recorded signals.
(ii) Application of a Machine Learning algorithm Support Vector Machine (SVM) for further identification of cobbles. The SVM model is trained off-line and the offline training vectors are used for on-line prediction.
(iii) Development of Flow charts and Data Flow Diagrams (DFDs) of the Visual Basic.Net based Cobble Prediction method and system to predicts Healthy (no-Cobble), Normal(No Prediction) and Unhealthy (Cobble) cases on real time for a coil using signal analysis of previous coils.

Industrial Applicability

Cobble generation is a major problem in rolling mills not only to hot strip mill for hot rolling of flat products but also for hot rolling of long products like rods, bars, channels, rails. This combined methodology of statistical analysis and support vector machines is found to be an effective tool for prediction of cobble generation using signal analysis.

Although the foregoing description of the present invention has been shown and described with reference to particular embodiments and applications thereof, it has been presented for purposes of illustration by way of examples and description and is not intended to be exhaustive or to limit the invention to the particular embodiments and applications disclosed. The particular embodiments and applications were chosen and described to provide the best illustration of the principles of the invention and its practical application to thereby enable one of ordinary skill in the art to utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. All such changes, modifications, variations, and alterations should therefore be seen as being within the scope of the present invention as determined by the appended claims when interpreted in accordance with the breadth to which they are fairly, legally, and equitably entitled.

We Claim:

1.A method for real time predictionof cobble generation in a hot strip millcomprising the steps of:
capturing of the data for :
a) motor speed signals - 6 signals
b) motor field current - 6 signals
c) motor armature current - 6 signals
d) roll force - 6 signals
e) roll gap - 6 signals
f) interstandLooper angle – 5 signals
inreal time of the operation of the rolling mitt employing aplurality of sensors placed at strategic locations; and
classifying the said mill signals into a plurality of groups using limit values calculated from first quartile, 3rd quartile and inter quartile range of previously recorded signals.

2. The method for real time prediction of cobble generation as claimed in claim 1, wherein the said classification of mill signals is done for preferably 3 groups namely, No-Cobble, Cobble and No-Prediction.

3. The method for real time prediction of cobble generation as claimed in claim 1, wherein the said classification of mill signals is done in two stages: Statistical classification and Support Vector Machine (SVM) classification.

4. The method for real time prediction of cobble generation as claimed in claim 3, wherein the statistical classification includes analysis of each of the 35 signal data of previously rolled coils (more than 10000 coils)

5. The method for real time prediction of cobble generation as claimed in claim 4, wherein the said statistical analysis includes step of calculating the median value, first quartile and third quartile of the data and further calculating Inter Quartile Range (IQR), which is determined by subtracting the first quartile value from the third quartile.

6. The method for real time prediction of cobble generation as claimed in claim 5, wherein the step of calculation includes obtaining of one upper limit value of Q3+1.5*IQR and a lower limit value of Q3-1.5 * IQR.

7. The method for real time prediction of cobble generation as claimed in claim 6, wherein under condition that,If the signal value of the next recorded data falls outside the limits, those signals are considered as unhealthy and lead to a possibility of cobble.

8. The method for real time prediction of cobble generation as claimed in claim 3, wherein the Support Vector Machine (SVM) based classification of signals is trained in off-line from previously recorded signals of Cobble and No-Cobble cases.

9. The method for real time prediction of cobble generation as claimed in claim 8, wherein during the training, it first converts the signal from time domain to frequency domain using Fast Fourier Transformation (FFT) and an optimum hyperplane and two support vector planes for Cobble and No Cobble cases are identified and the values are stored.

10. The method for real time prediction of cobble generation as claimed in claim 9, wherein the said statistical technique identifies a signal for a possible case of cobble, the SVM further analyses it using the previously stored support vector values and finally predicts the possibility of cobble.