MXPA00000568A - Absorbent composites comprising superabsorbent materials. - Google Patents

Abstract

The present invention is directed to absorbent articles containing superabsorbent materials. The superabsorbent material has a Gel Bed Permeability (GBP) value of greater than about 70 x 10-9 cm2 and an Absorbency Under Load (AUL) value at 0.6 psi of less than about 25 g/g. The present invention is further directed to fiber-containing fabrics and webs containing superabsorbent materials and their applicability in disposable personal care products.

Description

SYSTEM. BASED ON MULTIVARIATE STATISTICAL MODEL FOR VERIFY THE OPERATION OF A CONTINUOUS FOUNDER AND DETECT THE APPARATUS OF EMBEDDING BREAKS FIELD OF THE INVENTION This invention relates to a method for verifying the parameters of a continuous, specific steel casting machine and the use of this information to predict the possibility of a rupture occurring in a thin film of metal imperfectly bonded to the surface of the solidified acere before it actually occurs, such action can be taken to prevent rupture.

PREVIOUS TECHNIQUE Continuous steel casting, in the iron and steel industry, is the process of converting liquid steel into solid steel plates or strings. This transformation of the liquid to solid state is achieved through a process known as continuous casting. In this process, the liquid steel is continuously poured into an open copper mold. Cooling water is applied internally to the walls of the mold, so that the liquid steel in contact with the copper mold solidifies to form a thin film of imperfect metal attached to the surface containing liquid steel within the interior of the steel metallic cord. The thin film of metal imperfectly bonded to the solidified steel surface is continuously removed from the mold to the additional cooling chambers of the melter, where the remaining internal liquid steel solidifies under controlled cooling conditions. During the casting process, ruptures may occur in the thin film of metal imperfectly bonded to the surface that is solidifying because localized liquid steel does not solidify properly. When such rupture reaches the end of the mold, the molten steel leaks through the rupture and causes extensive damage to the melter. This phenomenon is known as a burr due to mold breakage. Burrs due to mold breakage result in high maintenance costs and production losses and can lead to hazardous conditions that have an adverse impact on the safety of production. Mold burrs can be avoided if the melting rate is reduced when the steel does not solidify properly. The reduction in casting speed gives more time for the steel to solidify and also reduces productivity. To prevent the occurrence of burrs due to mold breakage, it is critical to predict the improper solidification of the thin metal film imperfectly bonded to the surface of the steel with sufficient time to take corrective action. Casters in the steel production industry typically use mold burst detection systems that observe the specific patterns in the mold temperature readings. The pattern comparison systems are based on experiences of mold break burr from the previous melter. Rules were developed that characterize the patterns in the temperatures before the incidence of a burr due to mold breakage. If the patterns in the mold temperature readings follow these rules, there is a high likelihood that a burst will occur due to mold breakage. If the conditions of those rules are satisfied, typical mold break burst systems produce an alarm so that the operator takes the necessary action to prevent mold break burr or take action automatically. This usually means slowing down the casting speed. However, only a subset of all the data from the melter operation process is used in the development of those rules. Those rules typically comprise finding specific differences and rates of change variations for specific mold temperature readings. Typical rules are as follows: the change rate of thermocouple A is greater than X degrees Celsius for Y consecutive readings; the reading of the thermocouple B is greater than the reading of the thermocouple C for Z consecutive readings. Current industrial mold burst burst detectors generate an alarm only when a predetermined set of rules has been satisfied, indicating that the mold burst burst is imminent. These systems provide a binary signal as output, alarm or not. There are no indications of when the system is approaching the alarm or the severity of the alarm. In some cases, there is not enough anticipated time to react or prevent the mold break burr from occurring. This inevitably results in some mold break burrs occurring without detection. To date, there is no known system that has been able to detect each type of mold break burr. Having some burrs of mold breakage is considered part of the operation cost of a continuous melter. Comparison detection systems of this type are described by Yamamoto et al. in U.S. Patent 4,556 / 099, Blazek et al. in U.S. Patent 5,020,585, i Nakamura et al. in U.S. Patent 5,548,520, and by Adamy in U.S. Patent 5,904,202.

In addition to the prior art in the field of burst burrs detection systems for continuous melters, the Applicant is familiar with the prior art in the area of process verification and implicit detection. For example, a class of verification systems has been described in the Canadian Journal of Chemometries, Vol. 69, by Kresta, MacGregor, and Marlin in 1991 (and by others since then) based on the use of a multivariate process model to describe the normal operation of a process. In this method, new data is supplied to a model in real time, and calculations are made to determine a prediction error and summary, (latent), variables. These calculated data are then tested to determine whether the process is operating normally or not. This is basically the method adapted by Wang ei al to detect faults in disc manufacturing tools as described in U.S. Patent 5,859,964. A flowchart of the generic verification system as described by the published prior art is shown in Figure 1. Such a system is typically displayed on a computer with access to sensor signals from field instruments using a video monitor to present the exit. The system requires the process signals as input for a mathematical model and calculates output values as described in Block 10. Block 12 provides the calculation of test statistics, such as a prediction error to be used in the next step . The decision of whether or not the new observation is normal is made in Block 13. Threshold tests are performed on the test statistics to determine the probability that the new observation belongs to the normal operation set. If the new data is considered normal, the system repeats the Block 10 process in the next sampling interval, but if the probability is sufficiently low, a signal is issued to take a corrective action on the process, either manual or automatically. Block 14 is provided to determine the contributions to test statistics. The information is presented to direct the appropriate actions. The final block shown in the figure, Block 15, provides the corrective action that must be taken to remove or mitigate the previously detected failure. The system continues its cycle through the algorithm that starts again in Block 10. This method was tested to determine if it was applicable to a continuous casting process by Vaculij in 1995. The results of this off-line work showed the applicability of the technology to the particular process. What was not included in this work, however, are the details required to implement a viable online system. The work provides motivation for the development of an on-line system to detect abnormal operation, including burrs of mold breakage. Several significant innovations were required to make the system a reality in its current form. These developments deviate from the prior art and are integrated into the successful operation of the system; they are described later.

DESCRIPTION OF THE INVENTION The invention is a system for checking and detecting defects in line for a continuous casting process based on the application of a multivariate model of normal process operation. A further aspect of the invention deals specifically with an on-line system implementation and the development of the model not found in the prior art. In accordance with this invention, it is proposed to use an extended set of process measurements, beyond the standard molding temperatures, to develop a multivariate statistical model to characterize the melting process. The model is then used in the context of a verification system that detects exceptions to normal operation and predicts burrs of mold breakage in the continuous casting process allowing it to be taken 1 corrective action to avoid a burst of mold breakage. The system is implemented on a computer using sensor inputs from the casting process to provide input data. The invention relates to the prediction of the occurrence of improper solidification of the steel in a melter mold. This prediction process is based on a multivariate statistical model of the normal operation of the melter. The model was developed using the statistical modeling technique, principal component analysis (PCA). The PCA is a method to decompose a data matrix into a set of vectors and scalars. This method produces a model that projects the original data on less variables without loss of information. The results of the model are then used to calculate test statistics from which the condition of the melter can be inferred. If the condition guarantees, the system will generate warnings and alarms, so that corrective actions can be taken. This action can be taken manually by the operator or can be controlled automatically by the output signals of the system. The invention includes the following aspects that arise only in the case of online implementation; preprocessing of the input data in the form of specific filtering signals for non-stationary, or moving, addressing in the process; the ability to dynamically compensate for lost or invalid input data; ability to dynamically change models from one operating regime to another; consolidation of model results to facilitate verification in fewer dimensions; implementation of an alarm logic that works with the detection algorithm to reduce the percentage of false alarms; the presentation of information is organized using a hierarchical structure; the presentation of the results of the system is carried out using visual and audible indications; and the presentation includes a graphic indication of the influence of process parameters on the level of test statistics. In addition, the invention includes the process used to develop a model for the system, a prerequisite for successful implementation in lines. There are a number of aspects for this process that are critical for the operation of the system, including: selection of process parameters to be used in the model as inputs, these include the addition of outdated variables to add dynamic information to the model; selection of the data set to be used to adjust the parameters of the mold; selection of significant component numbers in the PCA model; and determination of appropriate detection thresholds for test statistics. A specific flowchart for this system and including the points described above is shown in Figure 2. The notable differences in Figure 1 include the development of the model and the implementation characteristics of the system. The implementation portion of the flow chart verification system in Figure 2 differs from the generic case as described in the prior art and as seen in Figure 1, with the addition of the following steps: data preprocessing between the acquisition of data and model calculations (step 32), consolidation of model output (step 34), alarm logic for more robust online decisions (step 36), specific output processing (step 37).

DESCRIPTION OF THE DRAWINGS In order to better understand the invention, a preferred embodiment is described below with reference to the accompanying drawings, in which: Figure 1 is a flow chart describing a typical implementation of a verification system based on the model; Figure 2 is a flow chart describing the application of a verification system based on the model for a continuous melter according to the invention; Figure 3 is a representation of a main verification screen according to the invention. Figure 4 is a representation of a screen that provides information on which of the process variables are contributing to the level of a test statistic HTl according to the invention. Figure 5 is a representation of a screen that provides information on which of the process variables are contributing to the level of an HT2 test statistic according to the invention. Figure 6 is a representation of a screen that provides information on which of the process variables are contributing to the level of an SPE test statistic according to the invention. Figure 7 is a diagram showing the basic components of an online system, according to the invention; and Figure 8 is a schematic of a continuous casting mold and provides an indicator of the locations of the thermocouples in the mold.

BEST MODE FOR CARRYING OUT THE INVENTION The invention is a system for verification and detection of online faults for a continuous casting process based on the application of a multivalent model of the normal process operation. As indicated above, additional aspects of the invention include the process by which the model was developed. The first step in this process, identified by the number 20 in Figure 2, is to determine which variables are included in the model.

Selection of Variables The selection of the parameters of the process to be used in the model as input are based on the knowledge of the continuous casting process. The model was developed using the following variables: readings of the mold thermocouple; mold thermocouple readings, out of phase; temperature differences between vertical pairs of thermocouples; melter speed; melter's out-of-phase speed; mold width; mold level; frequency of oscillation of the mold; temperature differences of the cooling water of the mold (that is, between the entrance to the mold and the outlet of the mold); Mold cooling water flows; funnel weight; funnel temperature; Calculated obstruction index. In the list of variables above, the calculated obstruction index is the only input that is not measured directly. This is derived from the ratio of the actual position to the predicted position of the control valve that controls the flow of hot metal from the funnel into the mold to regulate the level of the mold. These variables define the operation of the continuous casting system. Each variable in the previous list contains information or has an impact on the state of the solidification process in the mold. Since the behavior of the material in the mold is critical to the integrity of the thin metal film imperfectly bonded to the surface, the mold is typically instrumented with numerous thermocouples. Figure 8 shows a typical thermocouple configuration of the mold. A mold 50 is drawn having a variable width w, a fixed length 1, and a fixed height h, the resulting casting having a cross section defined by the width w and the length 1, and the casting direction of the upper part of the mold The lower part of the mold is indicated by the arrow 52, parallel to the height h of the mold. At a minimum, the thermocouples 54 are distributed around the mold 50 in two rings, the thermocouples in an upper ring are indicated as 54u and the thermocouples in a lower ring are indicated as 541. The thermocouples 54u in the upper ring are equally spaced between the entire length and width of the mold 50 at a predetermined upper height and the thermocouples 541 in the lower ring are equally spaced apart from each other across the width and length of the mold 50 at a predetermined lower height, approximately 150 mm below 50u thermocouples in the top ring. The lower thermocouples 541 are placed directly below the upper ones 54u to form vertical pairs. The sampling rate of the data is not less than once per second for the online system to be effective, and preferably not less than twice per second. The availability of detection equipment and automation infrastructure varies among smelters. As a minimum requirement, a number of essential signals must be available to the system. Those essential signs are: width of the mold; thermocouple readings from the thermocouple mold, arranged in rings around the mold to form vertical pairs; out-of-phase readings of the mold thermocouple; temperature differences between vertical pairs of thermocouples; melter speed; melter's out-of-phase speed; mold level; Mold cooling water flow; temperature differences of the cooling water of the mold (that is, between the entrance to the mold and the outlet of the mold); measurement of the driver of the mold level controller (for example, clogging index). If more signals are available, they can be added to the quality of the mold and improve the operation of the verification system.

Development of the Model. The Principal Component Analysis (PCA) is a linear method and is not very suitable to explain the entire range of operations of a continuous melter. 'The biggest obstacles to be overcome in the implementation of an online system are the number of variables of active thermocouples due to the width of the casting, the effect of changes in speed on the characteristics of the signal and the behavior of the different steel grades in the casting process. In the course of the operation of the melter, the width of the mold (50) is modified by moving the narrow faces inwards and outwards. The relevance of the signals provided by the external thermocouples on the large surfaces of the mold depends on the width of the mold. In the normal operation of the melter, it is preferable to melt at a constant speed, for a casting speed it can, and, change for a variety of reasons. A change in speed effectively acts as a disturbance to the process and produces transient effects on the processed signals. The degree of steel or reception determines the process behavior due to changes in the properties of the material. The main concern is the effect of the peritectic grades of the foundry (Medium Charcoal) or process stability and how those grades affect the variability of the process signals. These aspects are not addressed in the previous work and only arise in the context of an online system. Since the number of active thermocouples changes with the width of the cast, that is, more thermocouples are active on a wider product than when the width of the cast is small, this effectively changes the number of entries to the mold and affects the structure of the model. Accordingly, separate PCA models are required for each interval within predetermined casting. The number of models is determined by the number of operating regimes, each with a different number of active thermocouples. Consequently, the number of models required for the system is determined by the number of models required to cover different operating regimes over the entire range of melter widths. In a specific case at the local # 1 Smelter of Dofasco Inc., Hamilton, Ontario, Canada, two molds were required, one for wide plates and one for thin plates, each covering the operating speed range. The selection of the model in the online system was based on the measured mold width. To compensate for changes in velocity, the addition of outdated variables was done to add dynamic information to the model to incorporate the dynamic behavior of the melter, which is essentially a steady-state model of the process. A novel method was developed to capture trends in the data and compensate for changes in speed. This was done by taking samples around the readings of the casting velocity measurements and using those as input parameters to develop the PCA model. Specifically, the velocity over the five previous consecutive samples (covering about 2.5 seconds of operation), were used 7 speed samples or 3.5 seconds before and the measured speed of 10 samples or 5 seconds before. This method contributes effectively to the dynamics of the casting process and allows the use of a single model to cover the entire range of operation speeds of the melter for each of the width, width and narrow intervals, respectively.

Selection of the training data set The offline data collection identified by the number 22 in Figure 2 and the processing to create a set of training data identified by the number 24 in Figure 2 is required to develop the models identified by number 26 in Figure 2 that characterize the normal casting conditions. The distribution of the data refers to the categorization of the data 22 in periods of normal and abnormal operation. Numerous trouble-free operation periods are included to determine the model for normal operation. Several specific criteria were used for normal operation, for the distribution. These include: sufficient temperature separation between the upper (54u) and lower (541) thermocouples (>; 10 degrees C); consistent temperatures in both thermocouple rings (+/- 10 degrees C) and stable mold level control (+/- 5 mm). The training data set was selected 24, so that it will span the operating range, that is, data of the entire range of width and speed conditions are included, unlike the restricted subset of data considered in the previous work. The data must also be balanced over the range of operation. This ensures that the model fits equally well over the entire operation window and does not deviate to specific conditions. For the narrow model, the width range ranges from 800 mm to 1200 mm; for the wide model, the interval is from 1200 mm to 1630 mm. For both, the speed range is 600 mm / min at the maximum speed recorded for the given width interval. In addition, the data of both stable and transitory operations were included to cover both static and dynamic operation conditions. Include operation period data where the melting rate was ascending or descending, provides additional information about the behavior of the process and allows the system to recognize this variation as normal. Similarly, the data from the periods of operation where the mold width changed were included in the training data set. That is, keeping with the underlying point that the training set contains only data characteristic of normal operation, that is, data that will not generate an alarm in the verification system. The individual data sets that satisfy the above criteria contain 300 samples taken over a window of 150 seconds of operation. These sets are then concatenated to build a large matrix of observations used for the development of the model.

Selection of the number of significant components As part of the activated development of model 26, the selection of the number of significant components in the log output vector T of the PCA model determines the operation of the system. The objective of selecting the number of components is to maximize the information content of the model with a smaller number of components. The number of significant components is determined by the training data 24, but is selected so that the model explains at least 80% of the variation in the data. A choice was made to select five components for the online system and this was based on the fact that more than 80% of the variation was explained and adding more components would increase the number significantly.

Determination of detection thresholds The detection thresholds identified by number 28 in Figure 1 for statistics were determined by means of process data 22. In theory, test statistics follow a known distribution, but, if uses the theoretical value, the system tends to be overly alarmed. Consequently, the detection values are derived using the offline simulation. For simulation purposes, it is necessary to distinguish between normal and abnormal operation and identify such data. The data is then used to simulate the operation of the melter and generate model results and subsequently test statistics. The simulation results indicate the level of the test statistics and allow the selection of threshold levels. The goal is to find levels such that the system does not emit an alarm under normal operating conditions and always emits alarms under an abnormal operation. Practically, this can not be achieved but an optimal level can be determined based on the relative costs of the radio on either side. Current thresholds provide a long-term alarm percentage of less than 2% for all test statistics.

Implementation of the system Once the offline models 26 have been developed, the implementation of the online verification system that contains the steps of the invention of how to pre-process inputs and use the results of the model to achieve the desired results is required. The online implementation includes the integration of offline models in a verification system that runs on a computer that has access to process data 31 of the type described above, in real time at a sampling rate of two times per second.

Process data 31 is preprocessed in step 32 to provide filtered values, outdated and calculated data. The verification system calculates the values recorded in step 33 with the model 26 developed using the training data set 24. This then takes the results of the model 33 calculations and calculates the test statistics from step 34. The statistics provide information on how the current operation conforms to the model, or the training data set 24 and, consequently, infers the condition of the melter. The results are presented graphically in step 37 on a computer screen and provide visible and audible signals to the equipment operator. The physical components of such system 60 are shown schematically in Figure 7. A continuous melter is generally indicated by reference numeral 62 and is coupled to diagnostic instruments which include inter alia thermocouples 54, water flow meters, and the like for provide process data 31, which is fed to the computer of a verification system 64. The verification system computer 64 has a data collection device for acquiring off-line measurements of the process parameters 22 used to create the multivariate statistical model 26 and for acquiring online measurements of the process data 31. The computer 64 has calculation devices configured to calculate a P matrix of coefficients using the principal component analysis (PCA), to generate a vector T of the registers, and to define a selected number n of significant components in such vector T. Additional calculations are made using the computer 64 to generate detection thresholds 28, to calculate test statistics 34 from the data of online process 34 for making comparisons 35 of the test statistics with the detection thresholds 28. The computer 64 is configured to generate an alarm according to the predefined criteria developed out of line 30, as indicated by step 36 in Figure 2, and the alarm comprises a visual display unit 66 and a horn 68 coupled to the computer 64. The data is obtained by continuously sampling and feeding the computer 64, which is also configured to store readings which may be used in the model calculations, and for the filter data, as required. The computer 64 also has control means whereby the speed of the melter can be adjusted automatically according to the predetermined alarm thresholds without requiring any intervention of the operator.

According to the indicated, there are a number of characteristics that are novel and not obvious in the realization of such a system. These characteristics are described in greater detail in the following text.

Preprocessing of the input data The preprocessing 32 is effected in the form of specific filtering signals to deal with stationary or deviations in the process. A method was required to compensate for deviations in the absolute temperature in the thermocouples for online implementation. The method devised to solve this uses an Exponentially Weighted Average Movement Filter, MA, to dynamically calculate the mean of the thermocouple readings. This calculated mean was then extracted from the thermocouple temperatures to generate a deviation signal used by the model. This method is also used on the flows and temperatures of the cooling water. The previous signals are the only ones that have deviation and if they have filtering applied. Other signals were not filtered as this would lead to loss of information.

Compensation of lost or invalid input data One of the features developed for the online system 60 was the ability to operate continuously in the absence of a complete set of input data. Occasionally, sensor signals are invalidated for a variety of reasons, including sensor calibration procedures, where the sensor is taken offline, sensor failure, sensor movement, and others. The system 60 can mark the entry as absent and work with the rest of the entries to provide verification and alarm as usual without the hassle of false alarms. This is done in the model by modifying the parameters of the model to ignore the contribution of the lost data and increase the contribution of the valid data to provide results that are consistent with a complete data set. The method for compensating the lost or 'invalid element in the input data vector involves setting the coefficients of the model corresponding to zero for each component. The remaining coefficients for each component are then inflated, so that the sum of their square values is equal to one. This can be through the use of the model to predict the lost value, then using that predicted value instead of the value lost in a verification system. Once the signal is restored to its normal state, it will be marked as such and used in the verification system. This can be done for any of the input signals and they are used very frequently for the thermocouples that have failed in the service. Other use cases include the treatment of the mold level signal as lost when the sensor is being calibrated and at the beginning of the casting sequence before reception of a valid funnel temperature, when the signal is marked as loss.

Programming the model As discussed above, more than one model is required to cover the entire operating range. The model to be used by the system at any given moment is determined by the current width of the metal cord. In fact, all models are calculated continuously, the width determines which of the outputs will be selected for detection, presentation and alarm. This method provides a uniform transition between the two different operating regimes. To maintain the definition of the training tests, the model changes between narrow and narrow to a cast width of 1200 mm.

Consolidation and testing of the results or outputs of the model To facilitate verification in smaller dimensions, represented by step 33 in Figure 2, the verification system calculates the number of records that form a vector T based on the preprocessed data input (inline) Z, from step 32, to model P, from step 26, according to the following formula: T = PTZ where: P is the coefficient matrix of the PCA model; Z is the vector of variables used as input of the model for the current observation; and T is the vector of records generated as model output. As a result of the use of the PCA, the records in vector T have known statistical properties and can be used to test the probability of an abnormal occurrence. As is typical with the PCA, the first record contains most of the information and, consequently, tests itself using a univariate statistical test, the Hotelling T test for univariate distributions. This signal is described as HT1 in the verification system and can be seen as the middle graph in the upper third of the main display screen shown in Figure 3. To summarize the information further, the subsequent significant records are combined to form a multivariate statistic that is tested against a single multivariate statistical distribution using another Hotelling T test adapted for multivariate distributions. This signal is described as HT2 in the system and can be seen as a graph further to the right in the upper third of the main display screen shown in Figure 3. In addition to these two tests, the Square Prediction Error (SPE) for the observation was also calculated and tested based on a known and univariate statistical distribution. This signal was described as SPE in the system and can be seen as the leftmost graph in the upper third of the display screen presented in Figure 3. In summary, the online verification system generates a vector of records 33 of a large number of variables in the data entry 31, based on the parameters of the PCA model 26. From those records and other internal calculations of the model, the verification system then generates summary univariate and multivariate statistics step 34) that are tested in step 35 against the developed threshold (step 28) during the development of the model using historical data 20.

Alarm logic The results of the detection of step 35 described above are passed to the alarm screens for further processing in step 36, if the minimum thresholds of normal behavior are not satisfied. The filtering logic is applied to the detection results to ensure the validity of the alarm. The alarm logic determines if the alarm is persistent before issuing a visual and / or audible warning. Any alarm condition that persists for at least five samples (2.5 seconds) before any alarm indication on the system operator's display. The system provides a quantitative signal of a range of casting conditions of "normal" "burst due to highly probable mold breakage". This gives the operator the maximum amount of information about the casting process and the maximum amount of time anticipated to take the appropriate actions when required. The system can be tuned to provide three specific alarm levels based on the results of the model, as follows: Level 1 - casting conditions are normal. The system could clean and readjust any previous alarm conditions; Level 2 - casting conditions have deviated from normal; conditions for a burst due to mold breakage are possible. The background color of the screen of the visual representation unit 66 presenting the results of the model becomes amber; Level 3 - smelting conditions have deviated significantly, so mold bursting is highly likely to occur. The system provides an audible alarm to the operator, who is asked to reduce the casting speed until the melting conditions improve. Also, the background color of the visual display screen 66 that the model results show turns red.

Presentation of information The system operator's screen has a hierarchy of presentations. The highest level, of which an example is shown in Figure 3, includes graphs of test statistics HT1, HT2 and SPE along with some operating parameters. The distribution of the screen is such that the statistical summaries are plotted on the upper third of the screen in separate line graphs. Above each graph is a selection area that provides access to the screens of the respective second level as described with reference to the subsequent figures. The lower two thirds of the screen are dedicated to presenting information of the mold level sensor in the form of a line graph and the individual thermocouples of the mold in graphics forms distributed in pairs around the perimeter of the mold. In addition, other data are displayed in the upper part of the screen numerically. These data include casting speed, metal bead width, mold level, casting length. There are also selection areas that allow direct access to trend graphs of the sensor signals. The following screens of the second level include the diagnostic screens for each of the test statistics, Figures 4, 5 and 6, which show the contribution of the process variables used as inputs to the model. Contributions are displayed in the form of color-coded bar graphs that indicate the level and meaning of the contribution. The distribution of the screen is such that the thermocouples of the mold and the differences between the vertical pairs are distributed around the perimeter of a mold scheme. Other variables are grouped on the screen to provide an organized presentation. The screens of the second level also include a selection area to return to the first level alarm screen shown in Figure 3. The selection areas are also included to provide access to trend graphs of the process variables. The next third level (not shown) is comprised of time traces of the process variables, either grouped or individually. The system presents the information in a simple, graphic way. On the highest level screen, as shown in Figure 3, each of the test statistics is presented as a graphic trend graph on the upper third of the computer screen. These graphs show values around the statistics and provide an indication of the evolution of the signal over time. The graphs have been normalized so that the value of 1.0 corresponds to the individual threshold. For each of the signals, SPE, HT1 and HT2, if the value is between 0 and 0.8, the condition of the melter is considered normal. If the value falls between 0.8 and 1.0, a warning is issued and the background of the graph changes to amber. If the value is greater than 1.0, the background of the graph turns red and an audible alarm message is issued. In the case of an alarm, diagnostic information can be collected by questioning the model. According to the invention, the contribution graphs are generated in step 38 for each of the three statistics. The graphs form the second level of the screen hierarchy and indicate which original process variables contribute to the warning or alarm condition. This information is presented, graphically and you have access to it by selecting the graph of the offending statistics. Figures 4 to 6 show examples of contribution graphs for HT1, HT2 and SPE respectively. This information is very useful in locating a burr due to imminent mold rupture and in determining the causes of an abnormal condition. This information is used to direct the operator to the appropriate trend graph of the original variables at the next level of the screen hierarchy, and to take the corrective action, as necessary, also to decrease the speed of the smelting machine to avoid the occurrence of a rupture of the thin metal film and perfectly attached to the surface that leaves the mold (step 39). The system can also generate output signals automatically to accommodate or mitigate the alarm condition, typically by controlling the speed of the melter.

INDUSTRIAL APPLICABILITY The realization of a system for predicting burrs by breaking the mold of the melter using a multivariate model of the process requires the availability of the process measurements described above to a computer. The computer is used to calculate the result, and model statistics are compared with the thresholds to generate a result that can be used to take evasive action to avoid burrs due to mold breakage. An example of such a system, and the steps required to develop this, are shown in Figure 2.

The development of the model is done using historical data offline. The alarm thresholds are also determined during this development (step 30). The system that the melter verifies uses the previously developed model to calculate values that are verified against alarm thresholds and generates the appropriate output. To develop the PCA model, a data matrix, X, is constructed with each row containing an observation, that is, values of the process variables for the same moment in time. These observations are taken from several periods of normal operation. The PCA is then used to decompose the X, TX matrix and determine the number of significant components. A wide range of normal operation data is used, including different casting speeds and steel plate widths to generate the model. The resulting model provides sets of weights that are used to generate the values of the main component for each multivariate observation. In the online implementation, the mold rupture prediction system based on the multivariate statistical model operates as follows. The measurements of the process are read every half second. Using these and previous measurements of the process, the data is used as input for a multivariate statistical model. The results from the model are used to calculate statistics and provide a quantitative measure of the state of the melting conditions. The results are presented graphically to the operator as continuous trends. These trends provide a quantitative signal on the state of the melting conditions over a period of time as shown in Figure 3. Another characteristic of the mold rupture prediction system based on the Multivariate Statistical Model is the ability to provide Diagnostic information about the casting operation. When the system detects these conditions for a mold burst burst is increasing, presents graphical information that indicates which measurements of the process (or combination thereof) are more different from normal operation. Figures 4, 5 and 6 show the graphic diagnostic presentation for the three test statistics. You have access to those diagnostic presentations by selecting the appropriate option on the main alarm screen, shown in Figure 3. The system can also generate output signals to automatically zoom out or mitigate an alarm condition, typically by controlling the speed of the melter . It will be understood that various variants may be made to the embodiment of the invention described above, within the scope of the appended claims. Those skilled in the art will appreciate that the method can be applied to operations other than those of the continuous smelter and that other multivariate statistical models other than the principal component analysis (PCA) may be suitable for such applications and could also provide test statistics. significant when applied to the verification of the operation of a continuous smelter. In addition, it will be understood that the verification of a continuous melter can be made to take corrective actions to prevent a burst due to mold rupture, but it could also be carried out to allow the analysis of the effect of changing the input parameters such as the steel composition and , therefore, allow the operation to be carried out without undue experimentation.

Claims (1)

1. CLAIMS 1. A method to verify the operation of a continuous melter that operates a predetermined casting speed, the method includes the following steps: acquire off-line measurements of the process parameters by selecting training data from the offline measurements of the process parameters to represent the normal operation of a continuous melter; develop a multivariate statistical model that corresponds to the normal continuous melter operation with the input of the training data; generation of a detection threshold of the multivariate statistical model and off-line measurements of the process parameters; acquire online measurements of the process parameters of the continuous smelter operation; and to determine if the parameters of the online process are consistent with the normal operation of the continuous melter d, according to the off-line multivariate statistical model, characterized in that the multivariate statistical model is used to calculate an output vector of records with the input of the online process parameters; the test statistics are calculated from the vector of records including the calculation of at least one univariate test statistic on one of the records, and a multivariate test statistic on a selected number of records; and comparing at least one univariate test statistic and one multivariate test statistic with the detection thresholds to generate a detection signal, the detection signal is indicative of whether the continuous melter is operating normally and whether a corrective action to control the melter speed. 2. The method according to claim 1, characterized in that the multivariate statistical model is a Principal Component Analysis (PCA) model. 3. The method according to claim 1, characterized in that at least one univariate test statistic is selected from the group consisting of the Square Prediction Error (SPE) and a Hotelling T. 4. The method of compliance with the claim 1, characterized in that the multivariate test statistic is a Hotelling T. 5. The method of compliance with the claim 2, characterized in that the record output vector includes a first statistically significant record and a univariate Hotelling T statistics calculated from that first record. 6. The method according to claim 2, characterized in that the record output vector is limited to n significant components and a multivariate Hotelling T statistics is calculated from n-1 registers. The method according to claim 1, characterized in that the record output vector is limited to a number of components n which is sufficient to explain at least 80% of the variation in training data. The method according to claim 1, characterized in that the parameters of the process include the width of the mold, the readings of the thermocouples of the mold, the level of the mold, the speed of the melter, the flow of cooling water of the mold, temperature differences of the cooling water of the mold and a cooling measurement of mold control. The method according to claim 8, characterized in that the readings of the thermocouples of the mold are taken from the thermocouples separated throughout the width and length of a continuous melting mold at a first determined height to define an upper ring and a second predetermined height to define a lower ring, the upper and lower rings are vertically spaced in a casting direction for the continuous melter. The method according to claim 9, characterized in that the process parameters additionally include temperature differences between pairs of thermocouples vertically spaced in the upper ring and the lower ring, respectively. The method according to claim 8, characterized in that the process parameters additionally include readings around the thermocouples. The method according to claim 8, characterized in that the process parameters additionally include readings around the speed of the melter. The method according to claim 12, characterized in that readings around the speed of the melter include measurements taken from five consecutive samples, a melter speed from a previous seventh test and a melter speed from a previous tenth test. 3.4. The method according to claim 8, characterized in that the measurement of the control of the mold level control is calculated from a ratio of a measured position of a control valve to control the flow of hot metal from a funnel to a mold to a predicted position of the control valve. 15. The method of compliance with the claim 8, characterized in that the process parameters additionally include variables selected from the following group: frequency of oscillation of the mold, weight of the funnel and temperature of the funnel. The method according to claim 1, characterized in that the measurements of the process parameters are subjected to sampling at a speed not less than once per second. The method according to claim 16, characterized in that the measurements of the process parameters are subjected to sampling at a speed of not less than twice per second. 18. The method of compliance with the claim 9, characterized in that the measurements of the off-line process parameters are categorized into periods of normal operation and abnormal operation of the continuous smelter and a set of training data of the measurements, taken during periods of normal operation of the continuous smelter in where: there is a temperature separation of at least 10 ° C between the readings of the upper and lower mold thermocouples; there is a consistent temperature of +/- 10 ° C from the readings of the upper and lower thermocouples, respectively; and there is a stable mold level control of +/- 5 mm. The method according to claim 18, characterized in that the training data set corresponds to a normal operation period of a continuous melter that lasts at least 60 seconds. The method according to claim 18, characterized in that the training data set corresponds to a normal operation period of a continuous melter lasting at least 150 seconds. 21. The method according to claim 18, characterized in that several sets of training data are concatenated to construct a large input data matrix to develop the multivariate statistical model. 22. The method according to claim 1, characterized in that the detection thresholds are selected to generate a long-term alarm percentage of less than 2% for all test statistics. 23. The method according to claim 1, characterized in that the selected on-line measurements of the thermocouple readings, mold cooling water flow, and temperature differences of the mold cooling water are filtered to compensate for deviations in the Readings The method according to claim 23, characterized in that the filtering is performed using an Exponentially Weighted Movement Average filter, EWMA, to dynamically calculate an average of the selected on-line measurements. 25. The method according to claim 1, characterized in that any inline process parameters are lost and marked and compensated by adjusting the appropriate model parameters to zero and proportionally scaling the remaining parameters to provide a valid output or result. 26. The method according to claim 25, characterized in that the lost online process parameters include failed thermocouple measurements. The method according to claim 1, characterized in that the multivalent statistical models are developed from respective training data sets, each of which corresponds to the normal operation of a continuous melter having a width range of preferred casting. The method according to claim 1, characterized in that the detection signal associated with the abnormal operation of the continuous melter and persisting for a predetermined minimum number of on-line measurement samples of the process parameters activates an alarm. 29. The method according to claim 28, characterized in that the alarm is a visual alarm. 30. The method according to claim 28, characterized in that the alarm is an audible alarm. 31. The method according to claim 1, characterized in that the test statistics are presented graphically. 32. The method according to claim 31, characterized in that the graphical representations of the test statistics are associated with graphic graphical representations of contribution graphs indicative of whether the process parameters that are in line are consistent with the normal operation of the continuous melter. 33. The use of a method according to any of claims 1 to 32 to improve the diagnostic information about a continuous casting operation. 3 . The use of a method according to any of claims 1 to 32 to control the speed of the predetermined melter on a continuous melter and avoid a burst due to mold rupture whereby the molten metal bursts through a thin film of Metal perfectly bonded to the surface of the solidifying metal. 35. A system for verifying the operation of a continuous smelter operation at a predetermined smelter speed, the system is characterized in that it has a data collection device to acquire off-line measurements of the parameters and processes selected to represent normal operation of a continuous smelter to create an array of X matrix training data; a calculation device for decomposing an XTX matrix and determining a selected number of significant components to define a multivariate statistical model corresponding to the normal continuous melter operation; a calculation device to generate detection modalities from the detection of the multivariate statistical model and the off-line measurements of the process parameters, "a data collection device for acquiring online measurements of the process parameters during the operation of the continuous melter and create a real-time input data vector Z; a computing device for calculating an output vector T of records using a multivariate statistical model and the input data vector Z; a calculating device for calculating test statistics of vector T of records including calculating at least one univariate test statistic on one of the records, and a multivariate test statistic on selected number of records; a calculation device for comparing univariate test statistics and multivariate test statistics with two detection thresholds and generating a detection signal; and display means associated with the detection signal to signal the abnormal operation of the continuous melter. 36. The system in accordance with the claim 35, characterized in that the multivariate statistical model is a Principal Component Analysis (PCA) model. 37. The system in accordance with the claim 36, characterized in that at least one univariate statistic is selected from the group consisting of the Square Prediction Error (SPE) and a Hotelling T test. 38. The system according to claim 36, characterized in that the multivariate statistical tests are a Hotelling T test. 39. The system according to claim 35, characterized in that the data collection device includes a calculation device for calculating outdated parameter values using approximate readings. 40. The system according to claim 35, characterized in that the data collection device is configured for samples of the process parameters at a speed not less than once per second. 41. The system according to claim 35, characterized in that the data collection device is configured for samples of the process parameters at a speed not less than two times per second. 42. The system according to claim 35, characterized in that it has a calculation device for distributing the off-line measurements of the process parameters in periods of normal operation and abnormal operation to create a set of training data according to a predefined criterion consistent with the normal operation of a continuous melter. 43. The system according to claim 42, characterized in that it includes a calculation device for concatenating several sets of training data and constructing an input data matrix X. 44. The system according to claim 35, characterized in that the Data collection device to collect process parameters online is associated with a filtering device to compensate for deviations in the readings of the parameters of selected processes. 45. The system according to claim 35, characterized in that it has a tool for marking data to mark predetermined on-line process parameters of some of the real-time input data vector Z and to compensate the Z-vector for such lost measurements. 46. The system according to claim 35, characterized in that it has initiation means corresponding to a predefined casting range and adapted to select a multivariate statistical model associated with the predefined cast width range. 47. The system according to claim 35, characterized in that it has an alarm, deactivated by the detection signal to show that an abnormal operation of the continuous melter is occurring. 48. The system according to claim 47, characterized in that the alarm is a visual alarm comprising a visual display screen configured to change the color of its background according to the magnitude of the detection signal. The system according to claim 47, characterized in that it has an audible alarm device. 50. The system according to claim 35, characterized in that it has a visual display screen for displaying the test statistics. 51. The system according to claim 50, characterized in that the display screen is configured to present graphs of contributions associated with a respective statistical test. 52. The system according to claim 50, characterized in that the display screen is configured to present time traces of the on-line process parameters. 53. The system according to claim 35, characterized in that it has control means operatively connected to a detection signal to automatically adjust the speed of the melter, predetermined.