MXPA97008584A - Continuous colada process between rodil - Google Patents

Abstract

The present invention relates to a process for continuous casting between rolls to obtain thin metal products, especially those made of steel, where, during casting, the roller separation force (RSF) is measured continuously and a representative signal is obtained of the variations of the force of separation between rollers (RSF) in relation to time and where the separation of the rollers is modified especially in function of said signal to compensate the ovalization of the rollers, characterized in that in order to detect defects that they are not the rollers ovalization, the mentioned signal is broken into several harmonic components and these harmonic components are compared with reference harmonics of the corresponding order, the results of the aforementioned comparison being representative of a defective state of the casting process and, according to the results of the aforementioned comparison, rules are defined for control the queue process

Description

CONTINUOUS CASTING PROCESS BETWEEN ROLLERS FIELD OF THE INVENTION This invention relates to the continuous casting between two rolls of thin metal products, especially those made of steel.

BACKGROUND OF THE INVENTION According to this known technique, the manufactured product, for example, a thin strip of steel several millimeters thick, is obtained by pouring molten metal in a casting space defined between two rollers of parallel axes, cooled and rotated in opposite directions . The metal, when in contact with the cold walls of the rollers, called sleeves, solidifies and the outer layers or crusts of the solidified metal, rotated by the rollers, are joined in the neck between the rollers to form the said strip, the which is pulled down. The use of the roll casting process is subject to various limitations relevant both to the cast product and to the use of the casting facility.

REF: 26131 In particular, the section of the cast strip must correspond, in form and in dimensions, with the required section, directly depending on the actual section of the strip of the space - called separation distance - that is between the rollers in the neck. For this, a regulation process is known for the continuous casting between rollers, described in Patent Application FR-A-2728817, where the force of separation between rollers (RSF) is measured and the relative position of said rollers is modified so that it suits. This process makes it possible to modify the relative position of the rollers. These are moved away from each other if the force is too great, or they are moved closer together if the force is too small, especially in order to avoid burrs due to liquid metal outlet or even broken strapping, and also to avoid damage to the rollers in case of excessive solidification of the cast metal. It is also known that one can not completely avoid an ovalization of the rollers, on the one hand for mechanical reasons and, on the other hand, due to the thermal deformations to which the jacket is subjected when it first comes in contact with the metal melted at the start of the casting, and also later, during the rotation of the rollers. A compensation process for this ovalization is already known, which will be referred to hereinafter as "normal ovalization" (or also "mechanical ovalization" even though it is partly of thermal origin). This process consists of automatically acting on the position of the bearings of at least one of the rollers, depending on the angular position of these rollers, in order to keep the separation distance as constant as possible. As it is practically impossible to directly measure the separation distance, it has already been proposed to use as a representative parameter of the ovalization, a signal supplied by the means of measuring the separation force between rollers, the ovalization compensation system then being combined with a regulation system such as that described in the aforementioned document FR-A-2728817. However, the use of these processes does not enable the real-time detection of certain defects that may disturb the casting process, lead to its being stopped, or cause permanent damage to the rollers. There are already known defect detection methods, visual or otherwise, which make it possible to detect defects in relation to the casting process, for the thermal / dynamic characteristics of the molten metal, or also those known as "bright strips". . This last type of defect corresponds to a local decrease in the surface roughness of the rollers, which leads to variations in the cooling of the strip that can be detected by measurements of the temperature made on the cast strip. However, the observation of these defects can only be made after the event has occurred, on the strip already formed, and therefore after a certain time after they have appeared. However, these defects can damage the surface finish of the rollers, and this, in particular, when they are perceived at a late stage, in which case the damage can be irreparable. Certain defects can be detected a priori, by direct observation of the signal representing the force of separation between rollers. However, variations in this signal represent both variations in force due to normal ovalization and variations due to other parameters or events that may occur during casting. The direct observation of the force signal does not allow, therefore, determine the part that each of these causes plays in the vibrations of the signal to be determined. The purpose of this invention is to solve the aforementioned problems, and seeks to enable, by measuring the force of separation between rollers (RSF), the detection in real time of the defects, before an amplification of these defects originates irreparable damage, especially in the rollers. The purpose of this invention is also to make it possible to monitor the changes in these defects, in order to propose to the operator in charge corrective actions or the interruption of the laundry, depending on the severity of said defects.

DESCRIPTION OF THE INVENTION With these objectives in mind, the subject of the invention is a continuous casting process between rollers to obtain thin metal products, especially those made of steel, where, during casting, the separation force between rollers is measured continuously and obtain a representative signal of the variations in the force of separation between rollers (RSF) measured in relation to time, and where the adjustment of the rollers is modified, especially in relation to said signal, to compensate for the ovalization of the rollers, This process is characterized in that, in order to detect defects other than roller ovalization, said signal is broken down into different harmonic components, and said harmonic components are compared with reference harmonics of the corresponding order, the results of which are the aforementioned comparison representing the state of the defect of the casting process, and, according to the In the results of the mentioned comparison, the control rules of the casting process are defined. Indeed, the inventors have been able to establish, after many tests carried out on an industrial scale, that there is a certain relationship between the variations in the signals representative of the force of separation and the appearance of defects during casting. For example, the appearance of a defect called "bright strip" on a roller is characterized by the presence of a disturbance in the signal of the measured separation force. This disturbance is cyclical and occurs for each roll revolution. This disturbance reflects excess solidification of the product when it passes through the neck, and leads to variations in force that are clearly faster than those that would be generated, for example, by variations in thickness of the solidified product. The inventors then imagined the decomposition of said signals into harmonics, in order to differentiate in those signals the part that could be assigned to the normal ovalization of the part due to other causes. The inventors thus verified, by comparing the harmonic components recorded during several runs, that, although the signals representative of the separation force vary in particular according to the ovalization, and even when this ovalization is compensated by the compensation system, variations in certain harmonic components corresponded to the appearance of defects during the castings. It was therefore evident that an analysis of these harmonic components, carried out continuously during the castings, would allow, in comparison with a reference obtained experimentally during the castings considered as flawless, to detect in near real time revealing deviations of such casting defects, much more quickly than by the known methods. One hypothesis that explains the relationship between the variations in the harmonic components and the presence of casting defects is that normal ovalization causes variations in the signal representative of the separation force between rollers (RSF) that are essentially slow and soft. In other words, the signal has, due to said normal ovalization, mainly a harmonic component of low order, of frequency equal to the frequency of the rotation of the rollers. However, real defects, such as the aforementioned bright strips, led mainly to abrupt changes of said signal and, therefore, to harmonics of a higher order. Typically, the spectrum of the signal representative of the separation force between rollers and resulting only from normal ovalization, is characterized by a high harmonic component, of order 0 (for example, of 70% of the total amplitude of the signal) and the harmonics that decrease rapidly for the highest orders (20% for the harmonic of order 1, 10% for the harmonic of order 2). Rarely is the presence of higher order harmonics observed. However, when bright strips are present, the distribution of the harmonics is different from that of the previous case; the presence of an excess solidified edge at the level of the bright strip generates higher harmonics. It is specified here that in the following the component of the signal of a frequency Fj = 2'F0 has been designated as a harmonic of order i, where F0 is the fundamental frequency corresponding to the speed of rotation of the rollers. Similarly, in the following, the amplitude of the harmonic components will be designated in order i_ by h, and will be designated by H; a representative value of the harmonics of order i taken along a predetermined number of revolutions of the rollers. According to a specific arrangement of the invention, when a separation distance regulation system has been installed, such as that described above, an associated signal used as the reference for bearing displacement of at least one can be used. a roller, as a signal representative of the variations in the force of separation between rollers (RSF) obtained by measuring said force. In other words, the signal that is then decomposed into the various harmonic components is directly related to the aforementioned displacement reference, which is generated by an ovalization compensation module, and therefore reflects the variations in the force of separation. To decompose the signal into its various harmonic components, a fast Fourier transform, applied to the signal representative of the roller separation force (RSF), could be especially used, this signal therefore being directly the signal of the measurement of the separation force, or a corresponding signal generated by the aforementioned ovalization compensation module. In a preferred arrangement of the invention, the representative value Hj of each harmonic of order i is calculated as a mean value of the amplitudes h, of each harmonic, determined along a given number of revolutions of the rolls. Since the value H_, representative of each harmonic, is calculated as a mean value of the amplitudes measured along a given number of revolutions, this allows to attenuate the effect of the random defects, which are located in time and in the space and that are not repetitive throughout several revolutions of the rollers. Therefore, if a defect is generated due to a lasting problem in a roller, the system will completely integrate this data after the mentioned number of revolutions, while the effect of the harmonics, which appear in only a small number of revolutions, especially less than the aforementioned number of revolutions, will be considerably attenuated. The comparison of the signal measured with a signal from a cast judged to be good can be done in various ways. It is possible simply to compare term by term the H_ values representative of each harmonic of the measured signal, with the reference values Hi. obtained from the measurements made on the casts judged as good, and it can be verified that the sum of the differences of the H_ values representative of each harmonic with the reference values HÜ- is not too high. Alternatively, the proportion of each harmonic can be compared to a proportional reference distribution. However, - preferably the comparison will be made on the basis of a barycenter of harmonics, this barycenter being calculated by weighting each harmonic with a predetermined coefficient, in order to give relative importance to the different harmonics by weighing the latter unevenly for it. This method of calculation has been justified by experimental observations: during a cast judged as good, the first harmonic is the most important, decreasing the importance of the different harmonics in function of the increasing order of the considered harmonics. Weighting the harmonics of the highest order with an appropriate coefficient, the variations of these harmonics of high order, it will be as if the variations of these harmonics of high order were amplified, making their appearance or their increase more easily perceptible in the result of the calculation of the barycenter. For example, a barycenter of frequencies Bf can be calculated by assigning a coefficient that represents the amplitude of the harmonic considered for each harmonic frequency: Bf (Hz) =? H¡ + F¡ /? H¡ and this barycenter can be regulated by the fundamental frequency to obtain a relation R = Bf / F "that could be compared with a predetermined reference value R0 to be free of any differences in the fundamental frequency and, consequently, of any differences in the effective speed of the rolls between the considered laundry and the reference. In addition, the derivative dR / dt could be calculated and the result also be compared with a second predetermined threshold, thus making it possible to follow the change of the relation R, over time, with a rapid change of R being a sign of a rapid aggravation of a defect. With the values of the various parameters: A representing the total amplitude of the variations: A =? H, R representing the importance of the defects in the signal, and E = dR / dt can be formed a decision table, as will be seen below, that could be used to propose to the operator in real time corrective actions on certain parameters of the laundry, with the aim of correcting the defects as quickly as possible after their appearance. Other advantages and features will become apparent from reading the detailed description that follows of examples of embodiments of the invention, given for informational purposes and in no way restrictive, to be read in relation to the accompanying drawings, among which: - Figure 1 represents a schematic view of a roller casting device with a regulation system of a known type per se, but using the harmonic decomposition of the ovalization compensation signal, - Figure 2 represents a table of decision that defines the procedure to be followed during casting, based on the various values of the parameters provided by the process according to the invention, - Figures 3a, 3b, 3c and 3d illustrate, in the form of graphs that represent the variations of the various parameters measured or calculated, the results obtained from a cast judged as good with the compensation process Ovalization, Figures 4a, 4c and 4d, represent the corresponding graphics obtained during a cast judged as deficient. The installation for casting, represented only partially in Figure 1, usually includes, as is known, two rollers 1, 2, with parallel axes, spaced apart at a distance called separation distance. This corresponds to the thickness required for the cast strip, minus the dimensional reduction resulting from the deformations due to the separation force between rollers. The two rollers 1, 2 are rotated in opposite directions, at the same speed. They are supported by the bearings 3, 4, represented schematically, in two supports 5, 6 installed on a frame 7. The support 5, and consequently the axis of the corresponding roller 1, is fixed relative to the frame 7. The other support 6 it can move with translatory movement on the frame 7. Its position is adjustable and is determined by jacks 9 acting in such a way that they move jointly or separately the supports 5, 6, each in relation to the other. Means for measuring the separation force between rollers (RSF), such as the scales 8, are located between the fixed support 5 and the frame 7. Sensors 10 are used to measure the position of the mobile support 6 and consequently the variations as regards to position relative to a predetermined reference position, according to the thickness of the required strip. During casting, the molten metal is poured between the rollers and the molten metal begins to solidify in contact with its cooled walls, to form solidified crusts which are driven by the rollers and which are more or less joined at the level of the neck 11 between the rollers, to form the solidified strip that is pulled down. The metal therefore exerts a separation force (RSF) on the rollers, measured by the scales 8, this force being variable, in particular according to the degree of solidification of the metal. To regulate this force, and guarantee the continuity of the casting, the installation for casting includes a regulation system. In this regulating system, the difference between the force reference signal and the force signal measured by the force sensor 8 is calculated by means of a first comparator 12. The signal corresponding to that difference is inputted into a force regulator 13, which determines a position reference signal input to a second comparator 14. The force signal measured by the force sensor 8 is also introduced into a system by the Force sensor 8 is also introduced in an ovalization compensation system 15, which decomposes the force signal into harmonics and generates the compensation signals Hl, H2, H3 of each of said harmonics. These signals, Hl, H2 and H3 are summed in an adder 16, which generates a reference signal for correction of the position that is transmitted to the second comparator 14. The output signal of the second comparator 14 is introduced in a third comparator 17 , together with a position signal from the position sensor 10. The output signal of the third comparator 17 is introduced into the position controller 18, which controls the jacks 9. The rotation of the rollers 1 and 2 is ensured, by motors 19 and 20, respectively, controlled by a speed controller 21. This speed controller 21 receives a signal from a thickness regulator 22, - - which in turn receives a reference signal of the thickness, the force signal transmitted by the position sensor 10. By means of this regulation system an action is automatically carried out on the jacks 9 which makes it possible, for example, to act on the jacks 9 in the sense that Use a roller separation to reduce the force of separation (RSF) or, conversely, in the direction that leads to approximate the rollers to increase the force. In a similar way, this system makes it possible to compensate, at least partially, for normal ovalization, that is, to compensate for a possible decentering existing between the axis of the sleeve and its axis of rotation, and the irregularities in the shape of a roller, whether those irregularities of mechanical origin or thermal origin. The regulation system then takes into account these defects of shape and coaxiality, to give a displacement reference for the push jacks 9 that control the distance of separation between the rollers, in order to keep that separation distance as constant as possible. possible during the rotation of the rollers. Next, a preferred method for determining the various parameters A, R and E, which will be used to inform the operator of the presence of defects and the severity thereof, will be described.

In this method, the representative signal of the separation force between rollers will be decomposed, said decomposition being carried out in the ovalization compensation module 15, by means of a Fourier transform. This same operation would be carried out equally well not by using a Fourier transform but by using a Laplace transform, or any other signal processing or mathematical operation such as, for example, by using filters to obtain the same result, that is, the composition of the signal in several harmonic components. The various values H_ will then be calculated, as expressed above, that is, by obtaining an average of the amplitudes H i along a predetermined number of revolutions of the rolls, for example, of the last ten revolutions . Note that the above method for calculating the coefficients H, has been given as an example, and that it is by no means restrictive. It is also possible to calculate the H_ values representative of each harmonic of order i as the quadratic mean value of the amplitude Hj of the harmonics for any other calculated value that characterizes the aforementioned harmonics, this calculation being carried out by an arithmetic mean, by a method of least squares, or by any other method. Whatever the method of calculation, the values H i are representative of the amplitude corresponding to each harmonic of order i and of a frequency F i. The Bf criterion will then be calculated as a barycenter of frequencies of the different harmonics. That is to say, the barycentre of the frequencies of the considered harmonics is calculated, assigning to a value Fi a weight according to the corresponding value H_, that is to say: Bf =? Hx x Fx /? H_. In general, only harmonics of orders 0, 1 and 2 will be used. It is, however, obviously possible to consider other harmonics. In order to make valid comparisons for the various speeds of rotation of the rollers, the ratio Rf = Bf / F0 is calculated, with FD corresponding to the rotation frequency of the rollers. In the case given as an example, where only the first three harmonics are taken into account, we then obtain the following three criteria: - global amplitude of the signal variations: A = H_ + H2 + H3, - standard baricenter: Rf = (F? XH? + F2xH2 + F3xH3) / ((H1 + H2 + H3) xF0) - Change in Rf over time: E = dRf / dt. A comparison of these various criteria calculated during the casting with a predetermined threshold will then allow to detect if this or that defect appears in the current casting. As an example, in the case in which the signal representative of the roller separation force is the signal obtained from the ovalization compensation module, which is expressed as a displacement value of the mobile roller, and in the presence of only the ovalization normal, could be obtained from the following: H0 = 700 μm, H_ = 200 μm, H2 = 100 μm, where F0 = 0.2 Hz, Fi = 0.4 Hz and F2 = 0.8 Hz, and then Bf = 0 , 3 Hz and Rf = 1.5. If a bright strip appears, these values will be 350 μm, 350 μm and 300 μm, respectively, for H0, Hl and H2, and therefore Rf = 2.25. We can therefore see that simply by setting a suitable threshold for Rf, for example, R & mbrai = 1/6, the Rf step by that threshold can activate an alarm indicating a defect.

A better appreciation of the severity of the defects can be obtained taking into consideration simultaneously the three criteria mentioned above. For this, a decision table such as that shown in Figure 2 could be used to indicate directly to the user the state of the laundry in terms of defects, that is, to give the same an indication of the presence, the importance and the development of defects, and indicate the need to take corrective actions, such as modifying certain parameters of the laundry to try to correct defects that have appeared, or, in the worst case, the need to stop the casting to avoid irreparable damage to the installation for casting. This table presents, for example, the procedure that must be followed according to the corresponding values of the coefficients A, Rf and E: a "small" value of A is a sign of small variations in the separation force between rollers, with the wash in good conditions. - When the value of A is "medium", - and if the values of R and E are "small", this means few defects, or none, and then the casting is also carried out in good conditions, - if the value of R is " small "and the value of E is" large ", this may mean that, although there are no real defects, the operating point of the installation is unstable, for reasons essentially related to" normal "ovalization, and that it has exploded an alarm in the casting process to inform the operator of the need to modify, for example, the thermal conditions of the jacket (temperature or flow rate of the cooling water), - if the value of R is "large" and the value of E is "small", which indicates the presence of defects, without a notable tendency to its possible aggravation, an alarm is triggered in the casting process, - if the values of R and E are "large", which indicates the presence of defects and the aggravation of these, the stop is requested of the casting process, - when the value of "A" is "large", - and if the values of R and E are "small", no latent defect is indicated, the normal ovalization is correctly compensated, but the amplitude of the the displacements of the mobile roller to achieve this compensation is great, which is not serious for the laundry itself, but can relieve problems in terms of the geometric configuration of rollers, - if the value of R is "large" and the value of E it is "small", which also means the presence of defects, but without noticeable aggravation, an alarm is triggered in the casting process, - if the value of E is "large", regardless of the value of R, it indicates a significant worsening of the defects, and the quick stop of the casting process is requested. Note that the "small", "medium" and "large" values of the characters for the various criteria are determined by comparison with experimental data acquired during previous castings.To illustrate the possibilities of defect detection of the process according to the invention, let reference Figures 3a, 3b, 3c and 3d, in which the variations of the various parameters measured and calculated during a cast with the ovalization compensation process judged as good, and Figures 4a, 4b, 4c and 4d, in which a comparison of the curves obtained during a casting with bright strip defects is illustrated, Figures 3a and 4a illustrate the variations in the force of separation between rollers expressed in so many percent of the force of permissible roller separation for 40 minutes from the start of casting.

Figures 3b and 4b show the changes during that time in parameter A, that is, in the mean amplitude over 10 revolutions, in μm, of the displacement of the bearings of the mobile roller controlled by the compensation module of ovalization. Changes in the R parameter over time are illustrated in Figures 3c and 4c. Figures 3 and 4d show, in the same graph, the changes in the course of time of the values H0, H_ and H2, representative of the harmonic amplitudes of orders 0, 1 and 2, having illustrated the first (H0) at the bottom of the diagram, the second (Hx) at the center, and the third (H2) at the top. We can see that, for a laundry judged as good, an increase of A for approximately the first 20 minutes corresponds to a similar increase in H0 and essentially reflects the evolution of the ovalization compensation until the stability of A is obtained in approximately 50 um, which indicates an almost perfect ovalization compensation. Note also the stability of the parameter R after about 10 minutes, after the excursion of R towards higher values, corresponding to a relatively high amplitude of H2 during the same period, at the beginning of the casting. By comparison, the graphs of Figures 4b, 4c and 4d corresponding to a cast whose embodiment was very disturbed, have large amplitudes for H_. and H2 for about 40 minutes, with a high value of A during the same period and especially a high value of R. From these records it is easy to understand that a comparison, made in real time during the casting process, of the values of A and especially those of R, with predetermined thresholds, would have allowed to quickly detect the defects corresponding to the large amplitudes of the harmonics H_ and H2, and take an immediate action on the parameters of the casting to avoid its aggravation. The invention is not limited to the calculation methods of the various parameters given above, only as examples. In particular, following the use of the same values H_ representative of the amplitude of each harmonic, another barycentre B of the harmonic spectrum of the representative value of the roll separation force could be calculated, for example, by assigning each value H_ at a time carefully chosen weighting coefficient to accentuate in the calculated value of this barycenter the influence of the harmonics of the highest orders, which are those that reveal defects. Regardless of the type of calculation of the barycenter used, representative values of the different harmonics and of the weighting coefficients corresponding to each harmonic will be used so that it is easy to follow the evolution of the barycenter value and compare it with experimental values, in order to determine in real time a level of defects, by comparison with the condition in terms of defects (free casting of defects, disturbed casting, poor casting leading to a stop or damage to the rollers, etc.) of the previous castings. To compare the harmonics, it is also possible to define a reference distribution of the amplitudes of the harmonics as a percentage of each harmonic in relation to the total signal, for example assuming a priori that the first harmonic represents 66% of this signal , the second 17% and the third also 17%. It would then be possible to follow the evolution of this distribution during each casting and, by comparison with the distribution during each casting and, by comparison with the distribution during each casting and, by comparison with the reference distribution, easily determine any deviations. This comparison could be done, for example, - by calculating a sum of Ra of the differences between the proportion H_ / A of each harmonic component in the measured signal representative of the separation force and the reference proportion a_: Ra = pos (a0- H0 / A) + pos (H? / Aa?) + ... + pos (H¡ / A-a_), (ie, that each element of this sum is counted only if it is positive). Thus, if the proportion of the harmonic of order 0 is greater than the reference proportion, or if the proportion of a harmonic of an order equal to or greater than 1 is lower than the reference proportion, it is not taken into consideration the difference corresponding to the harmonic considered. For example, if the first harmonic represents, for example, 98% of A, the second 2% and the third 0%, which would correspond to an almost total absence of harmonics of an order greater than 0, and so Therefore, absence of defects, we would have that Rd = 0. If the installation for continuous casting between rollers does not include a system for regulating the separation distance as a function of the ovalization, one could of course use the process according to the invention, described above, taking for it as a signal subject to decomposition in harmonics the direct measurement of the variations in the force of separation between rollers (RSF), however the use of the values H i obtained from the compensation module of the ovalization remains practical when such compensation module already exists in the installation and already performs, within the scope of its usual operation, the required harmonic decomposition.

It is noted that in relation to this date, the best method known by the applicant to carry out the aforementioned invention, is that which is clear from the present description of the invention. Having described the invention as above, the content of the following is claimed as property

Claims (10)

1. A process for continuous casting between rollers to obtain thin metal products, especially those made of steel, where, during casting, the separation force between rollers (RSF) is measured continuously and a signal representative of the variations in force is obtained of separation between rollers (RSF) with respect to time, and wherein the separation of the rollers is modified especially in function of said signal to compensate for the ovalization of the rollers, characterized in that, in order to detect defects other than the the ovalization of the rollers, the aforementioned signal is divided into several harmonic components and these harmonic components are compared with reference harmonics of the corresponding order, the results of said comparison being representative of a defective state of the casting process, and, In agreement with the results of the aforementioned comparison, rules are defined to control the casting process.

2. A process according to claim 1, characterized in that said representative signal obtained by measuring the variations of the roller separation force (RSF) is an associated signal used as a displacement reference for the bearings of a roller in a loop of regulation of the separation between the mentioned rollers.

3. A process according to any of the preceding claims, characterized in that a Fourier transform is used to decompose said signal representative of the roller separation force (RSF) into several harmonic components.

4. A process according to any of the preceding claims, characterized in that to make the comparison, the value used as a representative value of each harmonics of order i is a value H i corresponding to the mean of the amplitudes H_ of the harmonics of this order measured over a given number of revolutions.

5. A process according to any of the preceding claims, characterized in that, to make the comparison, a barycenter of the harmonics is used, this barycenter being calculated by weighting a representative value of each harmonic with a predetermined coefficient.

6. A process according to claim 5, characterized in that a barycentre of the frequencies (Bf) = (? (H.xF,) / (? H_) is calculated, where the representative value of each harmonic is its frequency F_, and the Weighting coefficient H_ represents the amplitude of the harmonic considered.

7. A process according to claim 6, characterized in that the comparison is made on the basis of a relation Rf = Bf / F0, where F0 is the frequency corresponding to the speed of rotation of the rollers.

8. A process according to claim 1, characterized in that the comparison is made using as a comparison criterion the proportion H / A of each harmonic component with respect to the signal representative of the separation force, representing H_ the amplitude of the harmonic. order i_ and where A =? H¡.

9. A process according to claim 8, characterized in that the result of the comparison is represented by the sum Rd = pos (? O-H0 / A) + pos (Hi / A-cti) + ... + pos (H¡ / A-a_).

10. A process according to a series of claims 7 or 9, characterized in that a decision table is used to determine the procedure to be followed for casting, according to the values of the criteria: - A =? H¡, - R (Rf or Rd), - E = dR / dt.