METHOD AND SYSTEM FOR MONITORING THE GAP IN ROLLING MILLS

Abstract

Method for monitoring wear to cylinders of the cages of a rolling mill, in particular for bars or rods, including the following steps: reading by a neural network of a plurality of data relating to the initial conditions of one or more rolling cylinders, in particular one or more pairs of cylinders each belonging to a rolling cage, to the settings, and to the running of the process; and generation by the neural network of signals relating to the state of wear of the cylinders.

Claims

1-7. (canceled)

8. A method for monitoring wear to cylinders of mill stands of a rolling mill for bars or rods, comprising: reading, by a neural network, a plurality of data relating to: initial conditions of at least one rolling cylinder belonging to a corresponding rolling mill stand, settings, and a running of the process; and generating, by the neural network, signals relating to a state of wear of the rolling cylinders; wherein the neural network is trained with simulations at least one of: based on semiempirical models, and obtained with a finite elements method; and wherein the state of wear is evaluated as a linear value of a variation in depth of a runner in a preset point and a variation value of an area of a section of an exiting bar.

9. The method according to claim 8, wherein the at least one rolling cylinder comprises at least one pair of cylinders belonging to the corresponding rolling mill stand.

10. The method according to claim 8, further comprising adjusting a distance between the cylinders of one or more monitored mill stands.

11. The method according to claim 8, further comprising adjusting a rotation speed of the cylinders of one or more monitored mill stands.

12. The method according to claim 8, further comprising monitoring, by one or more neural networks including the neural network, wear to several mill stands of a rolling mill.

13. The method according to claim 8, further comprising: storing, in a database, preset data and data exiting the neural network; and reading, by the neural network, the preset data and the data exiting the neural network.

14. A monitoring system for monitoring wear to the cylinders of the mill stands of a rolling mill, comprising the neural network suitable for running the method according to claim 8.

15. A rolling mill provided with the system according to claim 14.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The invention will be explained in greater detail below with reference to attached FIG. 1 that shows schematically the structure and operation of a system according to the present invention.

DETAILED DESCRIPTION

[0021] According to one aspect of the invention, the neural network acts, for each mill stand for which it is provided, preferably for each mill stand of the rolling mill, as a virtual sensor of cylinder wear, by monitoring the parameters connected thereto during rolling.

[0022] During monitoring of rolling, different groups of data exemplified below are considered. In each group, depending on requirements, one or more or all the data items exemplified below can be used or still other data the appropriacy of which is detected. A workpiece can be a bar, a rod or the like, at each stage of processing. Here below, it will be possible to talk generally of bars, in particular in consideration of the product that is supplied to the rolling mill, and this term also means bars, billets or other products and, depending on the context also means products or semifinished products at any stage of rolling for example rods or other section bars of any suitable sectional dimension.

[0023] A first group of data relates to the parameters of running the process: [0024] Start time of rolling of workpiece; [0025] rotation speed of the cylinders; [0026] rolling forces and torque, if measured, for example in a known manner; [0027] temperature of the entering workpiece; [0028] temperature measured along the rolling mill if and where suitable sensors are available; [0029] distance between the cylinders (if in the mill stand there are means, for example encoders, for measuring the gap, otherwise, this datum can be a preset datum, which is corrected after possible adjustments); [0030] length of the workpiece that has passed through the mill stand (or through a reference mill stand), which can be the length of a supplied bar, resetting the data for each successive bar, in the case of traditional processing, or of a portion of suitably defined bar in the case of endless or semi-endless rolling mills. The length can also be obtained by integrating over time the transit speed obtained by the measurements.

[0031] A second group of input data consists of preset process data, which can be inserted into the system by the user by interface, or be taken from databases or information technology systems relating to the plant and can be: [0032] features, in particular mechanical features in the rolling conditions, in particular the temperature, of the processed material, in particular steel; [0033] distance between cylinders (if not measured in the plant as seen above); [0034] dimensional and shape parameters of the workpiece, for example, height and/or width of the bar entering, in particular the first mill stand monitored by the sensor, area of the section of the bar entering.

[0035] The third group of data relates to the cylinders of the monitored mill stands. Also these data are, according to particular aspect of the invention, preset data. According to one embodiment, these data can be obtained from a datum identifying the cylinder, for example a serial number, that enables other database data or plant systems data to be acquired automatically. The data may comprise: [0036] channel in use, in the case of multichannel cylinders, as indicated above; [0037] cylinder hardness; [0038] cylinder diameter; [0039] maximum permitted wear for each channel of the cylinder; [0040] wear thresholds to activate operations like the generation of alarms or the adjustment of the distance between cylinders; [0041] other features, for example geometric features of the calibration.

[0042] The neural network can supply a series of output data on a mill stand, for example: cylinder wear; [0043] width, height and area of the section of the bar exiting the mill stand; [0044] temperature of bar exiting the mill stand; [0045] possible correction of the distance between cylinders to be made.

[0046] The neural network can also generate alarms relating to wear, for example indicating the need to make manual corrections to the distance between cylinders or to replace the cylinders.

[0047] The output data can be used as input data (first group of data) for the network relating to another mill stand, in particular the mill stand downstream of the mill stand under consideration. The output data can be used to update a process database.

[0048] With reference to FIG. 1, a wear control logic is exemplified.

[0049] The process parameters 1 (first group of data), monitored, for example at intervals of a few seconds, are detected and stored by the system, for example in a process database 4, into which the preset data of the second 2 and third 3 group are loaded. By processing the data available in the database 4, the neural network 5 calculates the wear data and makes the correction of the distance between the cylinders, if envisaged. The new distance 6 is reintroduced as a process datum into the database, to continue monitoring.

[0050] The neural network can be structured in any known manner, for example a neural network can be provided for each mill stand and, as input data it can receive the data on the preceding mill stand, in particular the dimensions and shape of the workpiece, also depending on the calculated wear. In this manner, all the input data are updated continuously and the calculation occurs with maximum precision. This is the equivalent to a single neural network that receives the updated data calculated for each monitored mill stand and supplies the data for each mill stand, storing the data. Different structures can also be found on the basis of the needs and number and position of the monitored mill stands.

[0051] The neural networks can be of known type, with one or more hidden layers, depending on requirements, that can be defined during the network training step. They can be made by programming (software) on a machine (CPU) of known type, or be made with specifically dedicated logic units (hardware) or by creating a combination of the different solutions.

[0052] The neural networks used in the development of the method according to the present invention can be divided and trained according to the type of calibration: oval and/or round. Further, within one type of calibration, the subdivision can be by intervals of geometric features: diameter of groove bottom, depth of calibration, wear angles, calibration and connection radius. It has been found that it is possible to obtain the same number of nodes for different calibrations.

[0053] Normalization of data for the neural network can occur in the traditional manner. For example, the network can be organized according to a fuzzy logic, so the data can be standardized in value intervals between 0 and 1, for example, after defining a value interval that a parameter can adopt in normal operation, assigning a value between 0.01 and 0.99, whereas values outside the normal work value can assume values between 0 and 0.01 and 0.99 and 1, with normalization and denormalization functions of Sigma type.

[0054] The neural networks can be trained by running simulations by models. For example, the operation intervals of interest for rolling processes can be defined. A series of simulations is obtained by the models within the interval. For example, a series of simulations is obtained with a physical model of semiempirical type and a part is compared with simulation obtained by the finite elements method, which requires more time to run a simulation. In this manner, precise simulations are obtained, validated by two methods and the precision of the models is tested. For example, the simulations for a mill stand can take account of the following parameters: hardness and diameter of the cylinders, features of the workpiece (calibration, shape and dimensions), distance of the cylinders, productivity, temperature of the bar entering, speed of the cylinders, power, length of the bar that has passed through, features of the steel or other material.

[0055] After the values have been obtained that the network has to calculate, the network can be trained with the simulations to enable the network to supply, in a process, the output data seen above. In particular, wear is evaluated as a linear value, for example of the variation in depth of a runner in a preset point and a variation value of the area of the section of the exiting bar. The two combined values provide both a datum on wear, and on the precision thereof, the area being influenced by variations in dimensions of the runner also in points that are not evaluated with linear measurements.

[0056] It is clear that the work conditions of the rolling mill have to be as close as possible to those run for the calculation, for example by avoiding pretensioning in supplying the first mill stand of the rolling mill.

Example

[0057] Two rod-production sessions were run (I and II) in a rolling mill with several mill stands (from S6 to S11 of Tables 1 and 2) and the wear values were obtained (depth variation of the runner in the central part) (Table 1) and variation of the section of the exiting bar (Table 2), for all the mill stands, distinguishing the individual runners (the diameters of the runners are shown in the “runner” column of the same mill stand, where there are several runners. The values are obtained both with a neural network trained on the basis of simulations obtained with the Oike model (ANN columns) and measured at the end of production (measurements column)

[0058] The considered process data (first group), used by the neural network were speed of the motors, length of the workpiece that has passed through, temperature measured after mill stand S7 (with pyrometer) diameter of the bar and power of the motors. The preset data (second and third group) were: diameter and hardness of the cylinders, distance of cylinders (not adjusted in these tests), productivity, temperature of bar entering the rolling mill, dimensions of the bar entering the mill stand, geometrical features of the runner (width and depth), features of the steel. The errors contained, in all cases below 4%, indicate the reliability of the neural networks in predicting wear.

NUMERICAL REFERENCES

[0059] first group of data [0060] second group of data [0061] third group of data [0062] database [0063] neural network [0064] data exiting network

TABLE-US-00001 TABLE 1 Measurements ANN Error Mill [mm] [mm] [%] stand Channel I II I II I II S6 100 0.58 0.60 0.57 0.60 1.78 0.43 S7 31.5 0.58 0.59 0.56 0.58 3.63 2.04 S8 75 2.38 2.34 2.37 2.33 0.67 0.45 S9 22 24.5 1.20 1.10 1.20 1.08 0.58 1.66 S10 78 2.74 2.54 2.72 2.53 0.51 0.56 S11 16.5 1.50 1.52 −1.38 19.5 1.92 1.98 −3.12 21.5 0.89 0.88 1.02

TABLE-US-00002 TABLE 2 Measurements ANN Error Mill [mm.sup.2] [mm.sup.2] [%] stand Channel I II I II I II S6 100 57.42 59.26 55.70 57.90 3.09 2.35 S7 31.5 40.98 41.86 40.02 40.13 2.41 4.31 S8 75 111.65 108.03 111.34 107.32 0.28 0.67 S9 22 24.5 61.85 56.31 61.49 55.46 0.60 1.53 S10 78 114.90 104.56 114.11 103.83 0.70 0.70 S11 16.5 41.02 41.68 −1.59 19.5 52.07 54.06 −3.69 21.5 36.50 36.20 0.83