Method and system for monitoring gaps in a rolling mill

Abstract

A method for monitoring the wear of rollers of a rolling mill cage, in particular for bars or rods, comprises the following steps: a neural network (5) reads a plurality of data (1, 2, 3) related to initial conditions (3), settings (2, 3) and operation of one or more rollers, in particular one or more roller pairs, each of which belongs to a rolling cage, of a process (1); and the neural network (5) generates a signal (6) related to the wear state of the rollers.

Description

Technical Field

[0001] The object of the present invention is a method and a system for monitoring the gap in the rolling cage of a rolling mill, in particular a rolling mill for bars, rods or the like. Background Art

[0002] Laminators for the aforementioned continuous or discontinuous type of processing usually include a series of tandem cages, usually more than ten in number, which subsequently reduce the caliber of the workpiece and, if appropriate, the variation in the shape of the cross section. Therefore, the pair of rollers of the cage has a shaped channel. The workpiece passes between a pair of opposing channels, and the pressure of the rollers determined by their distance (gap) and the shape of the channel perform the variation in cross section that is the task of a single cage. The cage can also provide several stages and several pairs of adjacent channels that alternate in the travel of the subsequent workpiece, thereby increasing its working life.

[0003] A suitable type of motor, for example a synchronous type motor, can be adjusted accurately and continuously to ensure sliding of the workpiece, for example avoiding uncontrolled drag forces between one cage and the next, which would affect the cross section of the workpiece.

[0004] One problem associated with the operation of a laminating machine formed by several cages in series is associated with the significant wear of the rollers at the passage, which continuously changes their section and shape during a single processing section. This first requires the replacement of the rollers, requiring adjustment of the gaps between the rollers of the cages. The most modern cages are equipped with transducers and actuators that allow precise and automatic adjustment. However, it is difficult to determine when and to what extent the adjustment is to be made in order to maintain the characteristics of the workpiece. It must further be taken into account that the greater the number of cages working in series, the more serious the problem. In fact, changing the section of the exiting cage not only changes the shape of the exiting workpiece in the subsequent cages, but also changes the wear rate of the cages, with an effect that is spread and multiplied along the mill.

[0005] Different types of sensors can be used to monitor the cross section. Magnetic sensors are very widespread and based on the induction of eddy currents. Magnetic sensors are compact and relatively cheap and can be placed between two cages, thus helping, for example, to adjust the speed of the downstream cage. These sensors provide useful data about the cross section of workpieces with a circular cross section; however, in addition to a possible reduction in the reliability of the sensor, wear on the channel also leads to undetected shape changes. In the case of different types of cross sections, such as elliptical cross sections, the effectiveness of these devices is significantly reduced.

[0006] Other types of sensors, such as optical sensors, are expensive and require a lot of space, and if they are placed between two cages, the two cages should be spaced apart, increasing the size of the mill. Accordingly, their use is limited to, for example, the ends of a mill and possibly to one or a few units placed in selected points between two series of cages.

[0007] Therefore, monitoring the wear on the workpiece and the shape of the workpiece is an important aspect. Adjustment needs are usually met based on the experience of the rolling mill operator. The discontinuity of adjustment leads to discontinuity in the characteristics of the processed items, and the unpredictability of the wear state also reduces the processing and the length of the production section.

[0008] It would therefore be desirable to be able to accurately monitor the wear on several cages, and if possible on each single cage, preferably continuously or with a significant time frequency, and if possible intervene more frequently to correct the play in the cages, thereby monitoring the quality continuity of the workpieces, counteracting the cascading effect of wear on one cage on subsequent cages, and accurately envisaging the need to replace the rollers, all without reducing or even increasing productivity. Summary of the invention

[0009] According to the invention, the problem outlined above is solved by a method for monitoring the wear of rollers of a stand of a rolling mill, the method comprising the following steps:

[0010] A plurality of data relating to initial conditions, process settings and operation of one or more rolling rollers, in particular one or more roller pairs, each of which belongs to a rolling cage, are read in by a neural network;

[0011] A signal is generated by the neural network which is related to the wear state of the drum and preferably to the task to be performed accordingly.

[0012] According to one aspect of the invention, the signal generated by the network can result in a series of information pieces that are made available to the operator, for example through a screen or in other known modes, related to the adjustment of the distance between the rollers to be operated, or to the need to replace one or more rollers. According to another preferred aspect, the generated signal is used to adjust the distance between the rollers of one or more cages. Alternatively or in combination, the generated signal is used to adjust the speed of the rollers of one or more cages.

[0013] According to another aspect of the invention, a database collects data used by the neural network and provides the data to the neural network. The database may collect preset data and continuously (this term also refers to periodic data collection at an appropriate sampling frequency) collect data from sensors and process data, as well as manually or automatically preset data.

[0014] According to a preferred aspect, a neural network is provided for several pairs of rollers belonging to one cage, preferably for all cages of a rolling mill. According to one aspect of the invention, the neural network is commanded by data obtained from a suitable model. The model can be a physical model based on theoretical calculations and observations, such as a model of a semi-empirical model. Models of this type can exist in the literature or be obtained from observations of a rolling mill. The modeling can be obtained using the finite element method. Preferably, the data of the first model are verified using a second model. For example, data are obtained using a first physical model, which is partially verified using data obtained from finite element modeling. In all cases, any model that is considered appropriate can be used. The neural network is commanded by the model using a series of pairs of input data and output data (output signals) associated with it. In this way, the neural network can perform interpolation and possibly extrapolation required for its operation.

[0015] The invention also relates to a system comprising at least one neural network as disclosed above and a rolling mill provided with a system as outlined above.

[0016] The preceding claims and those to be explained below can be used not only in the particular combination indicated but also in other combinations or alone without falling outside the scope of protection of the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In the following, reference will be made to the Figure 1 The present invention is explained in more detail. Figure 1 The structure and operation of the system according to the present invention are schematically shown. DETAILED DESCRIPTION

[0018] According to one aspect of the invention, for each cage provided with a neural network, preferably for each cage of a rolling mill, the neural network acts as a virtual sensor of roller wear by monitoring parameters related to this cage during rolling.

[0019] During monitoring of rolling, different data groups are considered as exemplified below. In each group, one or more or all of the data items exemplified below or other data whose accuracy is detected may be used, as required. At each stage of processing, the workpiece may be a bar, a rod, etc. In the following, it may be possible to talk about bars in general, especially considering the products supplied to the rolling mill, and the term also means bars, billets or other products, and depending on the context, the term also means products or semi-finished products at any stage of rolling, such as bars or bars of other cross-sections with any appropriate cross-sectional dimensions.

[0020] The first set of data concerns the parameters of the running process:

[0021] The start time of rolling of the workpiece;

[0022] The rotation speed of the drum;

[0023] If, for example, rolling forces and torques are measured in a known manner;

[0024] Temperature of the incoming workpiece;

[0025] the temperatures measured along the mill if and where suitable sensors are available;

[0026] The distance between the rollers (if in the cage there is a means, such as an encoder, for measuring the gap, otherwise this data can be a preset data which is corrected after a possible adjustment);

[0027] The length of the workpiece that has passed through the cage (or through the reference cage), which in the case of conventional machining can be the length of the supply bar, thereby resetting the data for each successive bar, or in the case of annular or semi-annular rolling mills can be the length of a portion of a bar that is appropriately defined. This length can also be obtained by time integration of the transport speed obtained by measurement.

[0028] The second set of input data consists of preset process data, which can be inserted into the system by the user through an interface, or obtained from a database or information technology system (the database or information technology system is related to the device), and can be:

[0029] characteristics, especially mechanical characteristics in the rolling condition, especially the temperature of the processed material (especially steel);

[0030] the distance between the rollers (if not measured in a device such as that seen above);

[0031] Dimensional and shape parameters of the workpiece, such as the height and / or width of the bar entering the first holder, the cross-sectional area of ​​the entering bar are monitored in particular by sensors.

[0032] The third set of data relates to the rollers of the monitored cage. Moreover, according to a particular aspect of the invention, these data are preset data. According to one embodiment, these data can be obtained from data identifying the roller, such as a serial number, which enables automatic acquisition of other database data or equipment system data. The data may include:

[0033] In the case of a multi-channel drum as indicated above, the channel in use;

[0034] Roller hardness;

[0035] Drum diameter;

[0036] Maximum permissible wear for each channel of the roller;

[0037] Wear thresholds, used to activate actions such as generating an alarm or adjusting the distance between rollers;

[0038] Other features, such as calibrated geometric features.

[0039] A neural network can provide a series of output data about a cage, such as:

[0040] Roller wear;

[0041] Width, height and cross-sectional area of ​​the bar leaving the cage;

[0042] The temperature of the bar leaving the cage;

[0043] Possible correction of the distance between the rollers to be performed.

[0044] The neural network can also generate wear-related alerts, indicating, for example, the need for manual correction of the distance between the rollers or replacement of the rollers.

[0045] The output data can be used as input data (first set of data) for a network associated with another cage, especially a cage downstream of the cage under consideration. The output data can be used to update the process database.

[0046] refer to Figure 1 , a wear control logic is shown as an example.

[0047] The process parameters 1 (first set of data) are detected and stored by the system, for example, in a process database 4, which is monitored, for example, at intervals of a few seconds, into which preset data of the second set 2 and the third set 3 are loaded. By processing the data available in the database 4, the neural network 5 calculates the wear data and, if foreseeable, corrects the distance between the rollers. The new distance 6 is reintroduced into the database as process data for continued monitoring.

[0048] The neural network can be constructed in any known manner, for example, a neural network can be provided for each cage and the neural network can receive as input data data about the previous cage, in particular the size and shape of the workpiece, also depending on the calculated wear. In this way, all incoming data are continuously updated and calculations are performed with maximum accuracy. This is equivalent to a single neural network that receives updated data calculated for each cage monitored and provides data for each cage, storing the data. Different configurations can also be found based on the need and the number and location of the cages monitored.

[0049] The neural networks can be of known types, with one or more hidden layers, depending on the requirements that can be defined during the network training step. They can be performed by programming (software) on a known type of machine (CPU), or they can be performed with specialized dedicated logic units (hardware) or by creating a combination of different solutions.

[0050] The neural network used in the development of the method according to the invention can be divided and trained according to the type of calibration: elliptical and / or circular. Furthermore, in one type of calibration, the subdivision can be made by the intervals of geometrical features: diameter of the groove bottom, depth of the calibration, wear angle, calibration and connection radius. It has been found that the same number of nodes can be obtained for different calibrations.

[0051] Normalization of the data of the neural network can occur in a conventional manner. For example, the network can be organized according to fuzzy logic, so the data can be normalized in a value interval between 0 and 1, for example, after defining the value interval that the parameter can take in normal operation (thereby assigning values ​​between 0.01 and 0.99) using a sigma-like normalization and denormalization function, while values ​​outside normal operating values ​​can take values ​​between 0 and 0.01 and between 0.99 and 1.

[0052] The neural network can be trained by running simulations from the model. For example, an operating interval of interest for the rolling process can be defined. A series of simulations is obtained by the model within the interval. For example, a series of simulations is obtained using a physical model of the semi-empirical type and a part is compared with the simulation obtained by the finite element method, which requires more time to run the simulation. In this way, accurate simulations are obtained, verified by both methods, and the accuracy of the model is tested. For example, the simulation for the cage can take into account the following parameters: hardness and diameter of the rollers, characteristics of the workpiece (calibration, shape and size), distance of the rollers, productivity, temperature of the incoming bar, speed of the rollers, power, length of the bar that has passed, characteristics of the steel or other material.

[0053] After having obtained the values ​​that the network must calculate, it is possible to train the network with simulations so that it can provide the output data seen above in the process. In particular, the wear is evaluated as a linear value of the depth variation of the runner, for example, in a preset point and as a value of the variation of the cross-sectional area leaving the bar. These two combined values ​​provide both data on the wear and data on its accuracy, the area being affected by the dimensional variations of the runner, these variations also in points that are not evaluated with linear measurements.

[0054] It is clear that the operating conditions of the rolling mill must be as close as possible to those for the calculation run, for example by avoiding pre-tensioning when supplying the first cage of the rolling mill. Example

[0055] Two rod production sections (I and II) were run in a rolling mill with several cages (from S6 to S11 in Tables 1 and 2) and the wear values ​​(depth variation of the wheels in the central part) (Table 1) and the cross-sectional variation of the exiting bars (Table 2) were obtained, for all cages distinguishing between the individual wheels (the diameter of the wheels is shown in the "Wheels" column for the same cage, where there are several wheels). These values ​​were obtained using a neural network trained on the basis of simulations obtained with the Oike model (ANN column) and measured at the end of production (Measurement results column).

[0056] The considered process data (first group) used by the neural network are the motor speed, the length of the workpiece that has been traversed, the temperature measured after the cage S7 (with a pyrometer), the bar diameter and the motor power. The preset data (second and third groups) are: the diameter and hardness of the rollers, the distance of the rollers (not adjusted in these tests), the production rate, the temperature of the bar entering the mill, the size of the bar entering the cage, the geometric characteristics of the runner (width and depth), the characteristics of the steel. The errors contained in all cases are lower than 4%, indicating the reliability of the neural network in predicting wear.

[0057] Reference numerals:

[0058] 1 The first set of data

[0059] 2 The second set of data

[0060] 3 The third set of data

[0061] 4 Database

[0062] 5 Neural Networks

[0063] 6 Data leaving the network.

[0064]

[0065]

Claims

1. A method for monitoring the wear of rollers of a rolling stand of a rolling mill for bars or rods, the method comprising the following steps: A plurality of data (1, 2, 3) related to initial conditions (3) of one or more rolling rollers, settings (2, 3) and operation of the process (1) are read by a neural network (5); generating a signal (6) related to the wear state of the roller by means of the neural network (5), and wherein the neural network is trained using simulations based on semi-empirical models and / or obtained using the finite element method, and wherein the wear is evaluated as a linear value of the depth variation of the runner in a preset point and as a value of the variation of the cross-sectional area leaving the bar, The plurality of data related to the initial conditions (3) of one or more rolling rolls are selected from a group consisting of at least the following characteristics: In the case of a multi-channel drum, the channel in use; Roller hardness; Drum diameter; Maximum permissible wear for each channel of the roller; Wear threshold, used to activate the operation; The geometric features of the calibration, The plurality of data related to the settings (2, 3) are selected from a group including at least the following characteristics: Mechanical characteristics in the as-rolled condition; The distance between the rollers, if not measured in the equipment; The size and shape parameters of the workpiece, The plurality of data related to the operation of the process (1) is selected from a group consisting of at least the following characteristics: The start time of rolling of the workpiece; The rotation speed of the drum; Rolling forces and torques; Temperature of the incoming workpiece; Temperatures measured along the mill; The distance between the rollers; The length of the workpiece that has passed through the rolling stand.

2. The method according to claim 1, wherein: A plurality of data (1, 2, 3) relating to initial conditions (3), settings (2, 3) and operation of a process (1) of one or more roller pairs, each of which belongs to a rolling mill stand, are read in by a neural network (5).

3. The method according to claim 1 or 2, wherein: The distance is adjusted between the rollers of one or more monitored rolling stands.

4. The method according to claim 1 or 2, wherein: The rotational speed of the rollers of one or more monitored rolling stands is adjusted.

5. The method according to claim 1 or 2, wherein: One or more neural networks monitor wear on several rolling stands of a rolling mill.

6. The method according to claim 1 or 2, wherein: The database (4) stores preset data and data leaving the neural network and is read by the neural network.

7. The method according to claim 1, wherein: The operations include generating an alarm or adjusting the distance between the rollers.

8. The method according to claim 1, wherein: The mechanical characteristics include the temperature of the processed material.

9. The method according to claim 8, wherein: The processing material is steel.

10. The method according to claim 1, wherein: The size and shape parameters of the workpiece include the height and / or width of the bar of the first rolling mill stand monitored by the sensor.

11. The method according to claim 1, wherein: The size and shape parameters of the workpiece include the cross-sectional area of ​​the incoming rod.

12. A monitoring system for monitoring the wear of rollers of a rolling stand of a rolling mill, the monitoring system comprising a neural network adapted to run the method according to any preceding claim.

13. A rolling mill provided with a system according to claim 12.

Images