FR2993363A1 - Method for detecting risk of dysfunction in separation unit of chemical components of e.g. air, involves predicting future profiles of concentrations of chemical components along distillation column with assistance of model - Google Patents

Abstract

The method involves implementing a model for estimating concentrations (201, 202, 204) of chemical components of a product along distillation columns of a separation unit, where the module exhibits an adjustment parameter for variations in an operation of the distillation column. Future profiles of the concentrations of the chemical components along the distillation column are predicted (200) with the assistance of the model. A profile of the concentrations of the whole or part of the chemical components along the distillation column is determined with the assistance of the model. Independent claims are also included for the following: (1) a device for detecting a risk of dysfunction in a separation unit of chemical components of a product (2) a computer program comprising a set of instructions to execute a method for detecting a risk of dysfunction in a separation unit of chemical components of a product (3) a recording medium comprising a set of instructions to execute a method for detecting a risk of dysfunction in a separation unit of chemical components of a product.

Description

The present invention relates to a method and a device for detecting a risk of malfunction in a unit for separating the chemical components of a product. It also relates to a corresponding air separation unit. Although the invention can find its applications in the detection of many types of dysfunction, it is more particularly intended to allow the detection of nitrogen flushes in an air separation unit. For a better understanding of things, this phenomenon, also called "trip" according to the Anglo-Saxon associated term, is illustrated in connection with Figures 1 to 3. Figure 1 schematically shows a cryogenic unit 2 of air separation . This unit makes it possible to obtain oxygen, nitrogen and argon from the air. In known manner, said separation unit 2 comprises a first distillation column 14, called a high-pressure column, and a second distillation column 26, located here in the vertical continuity of the high-pressure column 14.

It further comprises a third distillation column 34, called crude argon column, and a fourth distillation column 36, said pure argon column, said first, second, third and fourth columns communicating with each other. The high pressure column 14 is supplied with air at the bottom of the column, as illustrated by the arrows associated with the mention AIR 1 and AIR 2. Liquid nitrogen, almost pure, noted LIN in FIG. 1, is recovered at the top of the column. the high pressure column 14. Almost pure liquid oxygen, noted LOX, is recovered at the bottom 28 of the low pressure column 26. Argon-rich liquid, noted LAR, is recovered at the bottom of the pure argon column 36. The phenomenon known as nitrogen puffs concerns the crude argon column 34.

Figures 2 and 3 show the evolution of the concentration of nitrogen, oxygen and argon along the longitudinal axis of the crude argon column 34. Along the abscissa, the original point T corresponds to the top of the column and the point B at which the curves stop corresponds to the bottom of the column.

By "concentrations" is meant more precisely here and in the rest of the text, "molar fractions". The term "concentrations" is kept for the purpose of simplification. Figure 2 corresponds to nominal operating conditions. In this figure, it can be seen that the concentration 400 of oxygen is almost zero at the top of the column to close close to 1 at the bottom of the column whereas the concentration 402 in argon is close to 1 at the top of the column as it approaches from zero to the bottom of the column. The nitrogen concentration 404 at the bottom of the column is very low whereas this component tends to accumulate at the top of the column up to about 30 times its concentration in the bottom column. Under nominal operating conditions, the nitrogen concentration is about 0.1% at the bottom of the column, which makes it possible to limit its concentration at the top of the column to about 3%. Figure 3 corresponds to the emission of a puff of nitrogen, the amplitude of the concentration differences with Figure 2 having been exaggerated to make things more visible. In this figure, it can be seen that the concentration of nitrogen at the bottom of the column has slightly increased; resulting in a much larger increase at the head of the column, to the detriment of argon. In addition to the inconveniences associated with the production of argon, such a situation risks defusing the condensation phenomenon occurring at the top of the crude argon column and, consequently, can disrupt the operation of the whole of the reactor. 'unit. In this way, there is a need to avoid nitrogen flushes. A known solution for this purpose is to define risk situations, that is to say, situations known to be able to lead to nitrogen flushes and to parameterize the control systems used to control the separation units. to keep them away from such risky situations. It is further known to employ for this purpose predictive models for estimating the future concentration of components. However, these models do not satisfactorily account for the nonlinear phenomena occurring in the columns, and the relevance of their estimation is debatable, especially during transient regimes. In addition, future concentrations are evaluated only at certain points in the column, which also limits the relevance of the predictions made. They can thus lead to the conclusion of a risk of flushing with oxygen when this risk is not real or does not detect a real risk. In the first case, the control systems over-react which disturbs the operation of the unit. In the second case, a puff of nitrogen can intervene, without having been anticipated. It is understood that the operation of the air separation units requires precise knowledge of the concentrations inside the distillation columns.

There are sensors on the market, known as concentration analyzers, which measure the concentrations of chemical components at a specific point in a distillation column. However, in addition to their prohibitive cost, the use of these analyzers requires the provision of additional pipes to take samples in the columns.

The addition of pipes and drilling are not desirable because they deteriorate the insulation of the columns. As a result, the number of concentration analyzers and temperature probes that can be used in a column is limited, making it impossible to measure concentrations anywhere along the column. In the state of the art, three major families of solutions have been proposed for estimating concentrations of chemical components in a packed distillation column, based on existing sensors. The first family of solutions is to establish a model, called a black box, in which the concentration to be estimated is considered as a function of the measurements made by the sensors. This function can be linear, polynomial, structured by a neural network, etc. The parameterization of this model consists in adjusting all its coefficients on the basis of experimental data. The setup time of the model is very long, which can take several weeks for an air separation unit. In addition, the black box model does not reflect the physical phenomena occurring in the column and it does not generally account for nonlinearities in the column. However, the distillation columns exhibit a very strongly non-linear behavior all the more when they operate at high purity, for example to obtain oxygen of very high purity from a separation of air. This makes the black box model relevant only to the operating conditions of the column close to those used for its parameterization. Approaches have therefore been proposed to develop nonlinear models. Thus, the second family of solutions consists in establishing a so-called plate model in which a packed column is modeled as a plate column consisting of several theoretical stages. This model is quite satisfactory for a classical intensive simulation of the column, for example during its design, but it is insufficient for a dynamic simulation in real time. In addition, this model is complex in terms of calculation because it requires several sets of equations, each set corresponding to a theoretical stage of the column. In addition, this model requires the parameterization of a large number of parameters and its stability is not always guaranteed. The third family of solutions consists in establishing a model, called wave model, expressing the distribution of concentrations along the column. According to this model, the distillation columns are considered as continuous beds along which the concentration profiles move as waves propagating according to the flow of gas and liquid. This results in concentration profiles with a general S-shape along the column. The wave model is parameterized with a parameter, called form factor, which adjusts the shape of the wave, including the smoothing of its pace. Generally, the form factor is considered constant in existing wave models. Thus, with this model, the wave slides in the column while its shape remains constant. However, this assumption is false when operating conditions change, particularly at high purity. Although interesting because delivering a profile of concentration all along the column, this model remains insufficient. To improve the relevance of the wave model, corrective approaches have been developed linking the form factor to the operating conditions by using inferential models or online estimation to adapt the shape of the curve. Although associated with another approach, it has also been developed by the applicant a method using a model for estimating concentrations along the distillation columns, also using a setting parameter to take into account operating variations. of the column.

The present invention aims to improve the detection of a risk of malfunction in a unit for separating the chemical components of a product based on this type of model. The invention thus relates to a method for detecting a risk of malfunction in a unit for separating the chemical components of a product, in particular air, comprising a distillation column of said product, in particular with packings, said process involving a model is used to estimate the concentration of the components along the column, in particular from concentration measurements made by one or more sensors, said model using a setting parameter making it possible to take into account variations in the operation of the column said method comprising a step of predicting at least one future profile of the concentrations of said components along the column using said template. Using this type of model, we have a complete estimate of future concentrations, that is to say, all along the column, and reliable, because taking into account effectively the nonlinear dynamics of the column. In this way, a more accurate forecast of malfunctions can be made. Having an estimated profile all along the column makes it possible in particular to be able to follow a point of interest of this profile which is not necessarily directly accessible by measurements. According to various embodiments, which may be taken together or separately: said method comprises a step of determining a profile of the concentrations of all or part of said components along said column using said model, at the moment present, the concentrations at the present time serving as the first input data of the model to perform said prediction step, at a first iteration, the prediction step uses as first input data to obtain the profile of concentrations during an iteration n the value of one or more of the concentrations obtained during a previous iteration n-1, - said prediction step is performed using one or more other physical parameters of the unit as data of additional input of the model, - according to a first variant, said physical parameter has known future values and these future values are used are known as additional input data for performing said prediction step, at each iteration, - according to a second variant, said additional input data is derived from a determination step in which said parameter is varied, in particular linearly, from a present value, or past, of said physical parameter and using said future values thus determined, at each iteration, - an increasing or decreasing orientation of said physical parameter is determined at the present instant. , or at a past instant, and said step of determining an additional input data while maintaining said orientation, - according to a third variant, said additional input data is derived from a determination step in which one maintains fixed said physical parameter, in particular at its value at the present moment or past, and one uses said values was ures thus determined, at each iteration, - said prediction step is performed by maintaining fixed the setting parameter, in particular its value at the present moment, - it updates said prediction periodically, - said method comprises a step of comparing the concentration of at least one of said components, at at least one point of said column, originating from one of said future profiles, with one or more reference concentrations of said component and / or the concentration of the same component, at same point of the column, coming from another of said future profiles, said method comprises a step of comparing one of the future profiles of at least one of said components with a reference profile and / or with another of the profiles of said component, at least for certain longitudinal portions of the column, - said method comprises a step of processing results of said comparison step, - said step processing is performed by assigning a different weighting to the results corresponding to one or more points of the column to decide the issuance of an alert, - or one of said particular points is a drawpoint. That being so, according to one aspect of the invention, the model employed is of the type for estimating the concentration of the components as a function of time and position along the longitudinal axis of the column, provided for packings. More specifically, it may be a model exploiting a propagation term in relation to a convection of said components along the column and an axial diffusion term in relation to a diffusion of said components in the column, the setting parameter to weight the effects of diffusion in relation to the effects of propagation. More precisely, the setting parameter, of very low value, in particular much less than 1, makes it possible to weight the diffusion term. The diffusion term of the model of the invention results from exchanges between the liquid and gaseous phases. Indeed, in a packed distillation column, the gas flows, or steam, amounts are in contact with the descending liquid flow. This is modeled simply by a single stream of vapor in contact with a single stream of liquid, through a single contact interface. At the interface, the liquid and the gas are concomitant and satisfy at each instant the thermodynamic equilibrium, which imposes the concentrations of the components in the liquid phase and in the gas phase. The further one is from the interface, the less the fluids are coupled thermodynamically. Therefore, far from the interface, the concentrations are different from the concentrations at the interface. In each phase, diffusion flows towards the interface tend to rehomogenize the concentrations. By taking into account, not only the phenomenon of convection of the components along the column, but also their diffusion, the method of the invention provides a relevant estimate of the concentrations.

In addition, by reducing the expression of the diffusion phenomenon along the same axis along which the concentrations vary due to the propagation term, namely the longitudinal axis of the column, the weight of the calculations to be performed is reduced. and the simulation can be implemented in real time. It may be noted that such a model comes from a microscopic analysis, in particular from an equation of some of the phenomena involving the concentration of the components in an infinitesimal section of the column. The diffusion term of the model results in this way, in particular, microscopic exchanges between the liquid and gaseous phases. Most commonly, the columns being oriented vertically, the position along the longitudinal axis of the column will represent the vertical position along the column. The invention will however also find its applications in columns having a different orientation, including horizontal. By way of example, the model in accordance with the invention delivers the value of the concentrations along an axis whose origin is placed at the end of the column at which the concentration of the least volatile compound is minimal and oriented in the increasing direction of the concentration of said least volatile compound. In other words, for air distillation columns, having a vertical orientation, the least volatile compound is oxygen and the origin is the top of the column, the axis being oriented according to the increasing concentration of oxygen, that is, from top to bottom. It should be noted that, by packed distillation column, distilling columns comprising elements in the form of metal sheets defining a network of liquid circulation channels and gas flowing through the column, said sheets being configured to said channels are strongly interlocked so as to promote contacting of the liquid phases and the gaseous phases circulating in said channels. However, the invention applies more broadly to any distillation column technology comprising elements defining such a network of fluid circulation channels.

Advantageously, the model uses two different time scales to take account of both the longitudinal circulation and the circulation in a direction normal to the interface of the components in the infinitesimal section considered, that is to say, according to a direction transverse to the longitudinal axis of the column, said circulation in a direction normal to the interface being faster than the longitudinal flow. Thanks to the use of these two scales, all the physical phenomena occurring in the column are taken into account in the model while allowing simplifications decreasing the weight of the calculations to be performed. According to this aspect of the invention, said model exploits, for example, a partial derivative differential-convection-diffusion equation linking a first derivative according to time, a first derivative and a second derivative according to the longitudinal position in the column of a value. in relation to the concentration of said components in the column, said setting parameter being associated with said second derivative.

Said model also exploits, for example, an approximate expression of the concentration of said components as a function of an intermediate value resulting from the resolution of said equation.

Said approximate expression is, in particular, a development limited with respect to the adjustment parameter, said limited development comprising a term of order 0, expressing the slow phenomena, and a term of order 1, perturbative. From said partial differential equation is derived a profile, according to the time and the longitudinal position in said column, of said intermediate value and a profile is determined, according to the time and the longitudinal position in said column, of the concentration of the components in the column, reporting said intermediate value in said approximate expression. By way of example, the partial differential equation has the following form: f X + Vk (X)] in which: t represents the time; z represents the position along the axis of the column oriented from top to bottom; - L and V represent the respective flow rates of liquid and gas in the column; X is a vector representing said intermediate value at time t, at position z; k is a matrix of functions, advantageously non-linear, of X expressing the thermodynamic equilibrium between the liquid and gaseous phases of the components; f and G are function matrices of X; and - this is the setting parameter. Function matrices are applications of the set [0; 1] A to real valued vectors in the interval [0; 1], of dimension A, where A is the size of the vector X, in the set M (R) AXN of the matrices with value in the set of real R, with A rows and N columns, N being 1 or A. According to a preferred embodiment, when the mixture contains M components, the size of the vector X is equal to M-1, since the sum of the concentrations of the components is equal to 1. For example, in an air separation unit, if the components of interest are oxygen, nitrogen and Argon, the size of the vector X is equal to 2.

In the case of a simplified binary mixture, for example a mixture of oxygen and nitrogen, the vector X is of size 1 and the partial differential equation is then purely scalar. Said diffusion term is, for example, a function of the flow of liquid and gas in the column. In other words, in the equation given above, the function matrix G is parameterized by L and V. The functions f and G may depend on a vector of parameters a representing the liquid and gas reservoirs. According to one embodiment, a and / or L and / or V depend on the time t and / or on the position z. Preferably, the method comprises a step of numerical resolution of the partial differential equation. For example, this digital resolution uses a technique of finite differences in time and space to ensure a fast calculation and low computational complexity. In one embodiment, the time step used for the digital resolution is between a value of the order of one second to a value of about one minute, and the spatial pitch is fixed at about 10 centimeters. The numerical scheme used to process the equation can be written implicitly or explicitly. To have calculated concentrations that are neither negative nor superior to 1, we may prefer the use of an implicit form method, even if it requires more calculations. According to one embodiment, the column comprises feed points and / or pulling the product and / or all or part of the product components. Said model divides said column into several sections, called homogeneous sections, each provided between two feed points and / or neighboring draw depending on the height of the column. Said convection term and / or said diffusion term are adapted to each homogeneous section. In particular, the vector of parameters a may be adapted. Adapting the sets of parameters used to each homogeneous section improves the accuracy of the model.

Said model can still exploit boundary conditions describing the principle of conservation of mass between two sections of the column, in particular between two homogeneous sections, and at the ends of said column. The boundary conditions thus complete the model at the ends of a homogeneous section of the packed column. Two cases arise. The first case is where the section is at one end of the column. In this case, the limiting condition translates a partial or total recycling of the flow leaving the column to obtain steam at the high end of the column and the liquid at the lower end of the column. The second case is where the section is connected to another section. In this case, the limit condition reflects a sampling or injection of liquid and / or gas between the two adjacent sections. However, said method may further comprise, for the time being, the steps of: measuring the concentration of at least one of said components in at least one point of the column; and - setting the model using the setting parameter determined from the measured concentration. More precisely, according to said method, it will be possible, in particular iteratively: - to estimate the concentration of said component at said location of the column where the measurement takes place using said model with a first value of said adjustment parameter, - to establish an error between the estimated value and the measured value of the concentration, establishing a second value of the setting parameter as a function of said error, - replacing the first value of the setting parameter with the second value in said model. Thus, comparison of the concentrations measured at particular locations in the column, for example with a concentration analyzer, and concentrations determined at the same locations using the model of the invention provides estimation errors which are then used to adjust the model. The measured concentration may in particular be a concentration at the top and / or bottom of the column, a place of high purity of one of the components of the mixture.

The invention also relates to a device for detecting a risk of malfunction in a unit for separating the chemical components of a product, in particular air, comprising a distillation column, in particular a packed column, said device comprising means for implementation of a model making it possible to estimate the concentration of the components along the column, in particular from concentration measurements made by one or more sensors, said model using an adjustment parameter making it possible to take into account variations of operation of the column, said device further comprising means for predicting at least one future profile of the concentrations of said components along the column using said model. Said device is in particular configured for implementing the method described above. The invention also relates to an air separation unit comprising at least one air distillation column and a device, as described above, for detecting a risk of malfunction, in particular nitrogen puffs, in the air separation unit. or one of said columns. The invention also relates to a computer program comprising instructions for implementing the method mentioned above, when the program is executed by a processor. The invention also relates to a recording medium in which said program is stored. Exemplary embodiments of the invention will now be described more precisely, but not limitatively, with reference to the appended drawings in which: FIG. 1, already mentioned, is a block diagram illustrating the structure and operation of a air separation unit according to one embodiment of the invention; FIGS. 2 and 3, already discussed, illustrate the concentration of nitrogen, oxygen and argon along the longitudinal axis of the crude argon column of the unit of FIG. 1, respectively, in FIG. nominal operating conditions and in the event of nitrogen flushes; FIG. 4 is a diagram illustrating an infinitesimal section of a column of the separation unit of FIG. 1; FIG. 5 is a diagram illustrating the phenomena involved in the infinitesimal section of FIG. 4; FIG. 6 is a diagram illustrating the updating of the adjustment parameter according to one embodiment of the invention; FIG. 7 illustrates an example of implementation of the method according to the invention, by means of a representation of the value of physical parameters serving as input data, of the adjustment parameter of the model used and of future concentrations. predicted, this as a function of time and for a first value of the present moment, - Figure 8 shows Figure 7 for a posterior value of the present moment; FIG. 9 illustrates a future concentration profile obtained during the implementation of the process according to the invention. As illustrated in FIG. 1, the invention will find its applications, for example, in an air separation unit 2, in particular a cryogenic unit. Such installations are also called ASU ("Air Separation Units" or "air separation devices"), according to the acronym usually used. To resume and complete the description already made above, said separation unit 2 comprises one or more distillation columns, including packed. It is advantageously configured to obtain substantially pure oxygen, nitrogen and argon from air, whether in liquid or gaseous form. It comprises here a first distillation column 14, said high pressure column, and a second distillation column 26, located here in the vertical continuity of the high pressure column 14. It further comprises a third distillation column 34 packed, called column crude argon, and a fourth packed distillation column 36, said pure argon column.

The high pressure column 14 comprises several homogeneous sections 16, 18, 20, here three. The low pressure column 26 also comprises several homogeneous sections, here five. The crude argon column here comprises a single homogeneous section. The pure argon column comprises several homogeneous sections, here two.

At each end of a homogeneous section of one of the columns takes place either an introduction of air or an introduction of one or components, resulting from the distillation in one of the other columns, or a draw of one or components derived from distillation in the column in question, this in the liquid phase and / or in the gas phase. In this way, distillation of the type known as reflux distillation is provided. Here, the high pressure column 14 is supplied with air in the liquid phase, as illustrated by the arrow associated with the mention AIR 1, and in the gas phase, as illustrated by the arrow associated with the mention AIR 2. The air supply in The gas phase is carried out at the bottom of the column and the air supply in liquid form is carried out above the homogeneous section 16 located just above the bottom 24 of the high pressure column 14. Liquid nitrogen, almost pure, noted LIN in Figure 1, is recovered at the top of the high pressure column 14. Liquid oxygen, almost pure, noted LOX, is recovered in the bottom 28 of the low pressure column 26. Argon-rich liquid, noted LAR is recovered at the bottom of the pure argon column 36. The separation unit 2 may further comprise a heat exchanger 30 at which some of the flows introduced and / or drawn from the high and / or low pressure columns 14, 26 exchange heat. Said high and low pressure columns may also be configured to allow a heat exchange between the top of said high pressure column 14 and the bottom of said low pressure column 26 to respectively allow liquefaction and vaporization of the components at this level. Advantageously, three oxygen concentration analyzers 41, 42, 43 are placed at specific locations in the low pressure column 26 and two analyzers 44, 45 are placed at predetermined locations in the high pressure column 14. , the separation unit 2 comprises a device 50 for determining the concentrations of the chemical components of the air at any point in the high-pressure 14 and low-pressure columns 26. This device 50 comprises in particular a processor making it possible to operate a model describing the variation in the concentrations of the components as a function of time and position along each of these columns 14, 26. This model uses the measurements made by the analyzers 41, 42, 43, 44, 45, this by exploiting a setting parameter to take into account changes in the operation of the column. The operation of this device 50 is detailed in the following description, in relation with embodiment of said model using a propagation term in relation to a convection of said components along the column and an axial diffusion term in relation to a diffusion of said components in the column, the setting parameter for weighting the effects of diffusion with respect to the effects of propagation. In this description, the term column without further specification will designate one of the columns 14, 26 of Figure 1.

The model used by the device 50 comprises the following partial differential equation (1): X + Vk (X)} + G in which: t represents the time; z represents the position along the axis of the column oriented from top to bottom; - L and V represent the respective flow rates of liquid and gas in the column; X is a vector representing an intermediate value related to the concentration of the components at time t, at the z position; k is a matrix of functions expressing the thermodynamic equilibrium between the liquid and gaseous phases of the components; f and G are function matrices of X; and - this is the setting parameter. The function k can be non-linear. By way of example, it is expressed as follows: where a is the relative volatility of the component in question relative to the compound of rank M whose concentration is not calculated but deduced from the calculated concentration of other compounds, as explained above. The first term of equation (1) represents the term of propagation. The second term of equation (1) represents the term of axial diffusion. It comes from the taking into account of a rapid microscopic phenomenon, namely transverse microscopic diffusion, this after simplification. The microscopic origins of equation (1) will now be detailed with reference to FIGS. 4 and 5. For the sake of clarity, the mixture is considered as binary oxygen / nitrogen, so that equation (1) is scalar , x and y corresponding to the oxygen concentration, respectively in the liquid phase and in the gas phase. (1) Figure 4 represents an infinitesimal section S of height dz of the column 14, 26 in which the phenomena of convection and diffusion are studied. In the packed distillation column 14, 26, the rising gas streams are in contact with the downward liquid streams. With reference to FIG. 5, this physical phenomenon can be modeled simply as a single gas flow 80 in contact with a single flow of liquid 82 through a single contact interface 84. The arrows 86, 88 show the vertical displacement of low at the top of the gas flow 80 and the arrows 90, 92 show the vertical displacement from top to bottom of the flow of liquid 82, the abscissa axis representing the distance at the liquid / gas interface and the X and Y ordinate axes representing the concentrations of liquid and gas respectively. At the interface 84, the liquid and the gas are concomitant and thermodynamic equilibrium at all times, which imposes the concentrations at the interface according to the relation (2): (2) in which the asterisk "*" indicates that it is a variable evaluated at interface 84.

Far from the interface, the fluids are no longer thermodynamically coupled so that the concentrations are different from the concentrations at the interface. The downward movement of liquid represented by the arrows 90, 92 is described by the relation (3): a (Lx) AL has t OZ + (3) in which a represents the liquid phase retention of the component in question and AT represents a diffusion coefficient in the liquid phase. Similarly, the upward movement of gas represented by the arrows 86, 88 is described by the relation (4): Dy O (Vy) c) t az (4) wherein ov represents the vapor phase retention of the component cause and Δv represents a vapor diffusion coefficient. k (x Remarkably, the model is not satisfied with being placed in a slow time scale, in which the phenomenon of radial diffusion allowing a mass exchange between the liquid and gaseous phases is neglected because it is too fast.

On the contrary, the model of the invention also uses a fast scale to describe this circulation phenomenon represented by the arrows 94, 96, 98. In each phase, diffusion streams 94, 98 tend to rehomogenize the concentrations. The diffusion then ends up affecting the interface 84 which can not accumulate or create matter. Thus, a mass exchange flux 96 must pass through the interface 84 and thus makes it possible to couple the diffusion flows of each phase. This mass exchange between the two phases is expressed by the relation (5): E (17 * -17) -F -1 -X) = 0 (5) The setting parameter c is very small, in particular much less than 1 The term Δv / c is comparable to the diffusion coefficient associated with the diffusion fluxes in the gas phase, and the term Mc is comparable to the diffusion coefficient associated with the diffusion fluxes in the liquid phase. The assumption that diffusion coefficients are very large is reasonable when column packing is effective. With this assumption, the system of equations (2) to (5) can be simplified.

To achieve this simplification, a so-called invariant manifold technique is used here, in particular a so-called center manifold technique. This technique makes it possible to preserve an overall mass balance. It also makes it possible not to make a phase predominant with respect to the other in the structure of the model, in particular from the point of view of the liquid / vapor retentions and from the point of view of the thermodynamic equilibrium. The reduction then leads to equation (1), in which function G makes it possible to relate the operating conditions of the column to the effects of diffusion. The function G can be expressed in the following way: k '(X) 2, (X) G (X) = AL 1- Av L ± v 1-avici (X) y- (6) in which k' is the function derived from the function k. It makes it possible to highlight the local effects of L and V on the diffusion.

The function f can be expressed in the following way: f (X) = crL + (X) (7) It should be noted that the parameters a, L and V may depend on the time t and on the position z.

The model also describes the concentrations in each phase. More precisely, an approximate expression of the concentration of said components as a function of the intermediate value resulting from the resolution of equation (1) is used for this purpose. Said approximate expression is, for example, a development limited with respect to the setting parameter, said limited development comprising a term of order 0, expressing slow phenomena, and a term of order 1, which is perturbative. By this we mean that the term of order 0 expresses an operation of the system in which the fast phenomena are considered as instantaneous and the term of order 1, which is perturbative, takes into account at least partially the non instantaneousness of said fast phenomena. The concentration x in the liquid phase can be expressed, for example, as follows: (8) The concentration y in the gas phase can be expressed, for example, as follows: G (X) OX Y (Z, k X + EUL vL + o-LV (9) Thus, to estimate the liquid phase and gas phase concentrations of a component, we can derive from equation (1) a profile, depending on the time and the vertical position in the column, of said intermediate value X and then determining a profile, according to the time and the vertical position in said column, of the concentration x in the liquid phase and y in the gaseous phase of the components in the column, by reporting said intermediate value X in said approximate expression (8) and / or (9) This is the approach that is implemented here in the device 50. In addition to equation (1), the model includes boundary conditions describing here the principle of conservation. of the mass between two sections, in particular between two homogeneous sections, of the column e and at the ends of said column. More X - Eav 0_ L + o-LV X G (X) especially, the effects of diffusion in equation (1) must be preserved at these boundary locations. FIG. 6 illustrates the operation of the device 50 for determining the oxygen concentration profile in the air separation unit 2.

Known data 100 are provided to the determination device 50. These include temperatures and / or pressures and / or flow rates of liquid and / or gas at particular locations in the separation unit 2. In addition, Concentration analyzers 41, 42, 43, 44, 45 provide the determination device 50 with discrete oxygen concentration measurements 102 in specific locations of the columns 14, 26. An initial version of the model is thus established. Alternatively, one can also choose arbitrary starting values. From these data, the determination device 50 provided with the model represented by the equation (1) iteratively estimates the oxygen concentration profile in the columns 14, 26. During the first iteration, the adjustment parameter c is set to a certain value. The determining device 50 estimates an oxygen concentration profile 104 using the model incorporated therein with this value of the setting parameter.

Then, the determination device 50 compares the estimated concentrations at the determined locations with the discrete measurements and deduces from them estimation errors (block 106 in FIG. 4) which it uses to adapt the adjustment parameter c (block 108 on the Figure 4). Thus, initially, during the first few iterations, the concentration profile can be very imprecise. After a while, the c parameter is set correctly. The determination device 50 then provides a precise concentration profile. Preferably, each column 14, 26 has its own setting parameter E. When estimating the concentration profile, the determining device 50 digitally resolves the partial derivative equation (1).

It uses a technique of finite differences in time and space to ensure a fast calculation of low complexity. The time step chosen for the digital resolution is set here to about one second and the spatial pitch is fixed at about 10 centimeters.

The numerical scheme used to process the equation is written so that the calculated concentrations are neither negative nor greater than 1, for example using an implicit scheme. According to a preferred embodiment, the adaptation principle 108 of the adjustment parameter is as follows: if the adjustment parameter c has a correct value, then the mathematical model of the equation (1) is realistic. In this case, the estimation errors are zero; - if the estimation errors are not zero, then the mathematical model is not correct. Therefore, it is necessary to change the value of the adjustment parameter E. Here, the determining device 50 uses an additional equation (10): dt M (10) in which M is a function of the estimation errors and possibly of other 15 parameters. Equation (10) allows the setting parameter c to be changed continuously in order to keep the estimation errors as low as possible. The function M may, for example, directly use one or more estimation errors and may take into account other parameters such as liquid and / or gas flow rates, pressures, etc. A simple linear function M solely dependent on a single estimation error can be used. It is also possible to use a more complex structure for the M function in order to accelerate the decline in estimation errors. However, as illustrated in FIG. 7, the invention relates to a method for detecting a risk of malfunction in a unit for separating the chemical components of a product, in this case air, such as that described above. . The method employs a model for estimating the concentration of components along one or more of the columns of the unit, as discussed in the preceding paragraphs. In other words, said model exploits an adjustment parameter, such as the parameter c, making it possible to take into account variations in the operation of the at least one column. In addition, said method comprises a step 200 of predicting at least one future profile of the concentrations of said components along the column using said model.

Thus, according to the method in accordance with the invention, the components 202, 204, ... along the column are established, component by component, at one or more future times t1, t2, tN in order to have a profile. concentrations of said components at each or of said instants, this for all or part of the components. This kind of information is available on the evolution of the unit on which one can work to detect risks of malfunction of said unit. FIG. 7 shows that, according to one embodiment of the invention, it will also be possible to use the concentration profile established for the present moment t. More precisely, concentrations 201 determined at the present moment, notably by virtue of said model, may serve as the first input data 207 of the model for performing said prediction step 200, in particular the concentration (s) found at the level of the at least one sensor. measured. Similarly, iteratively, the prediction step may use the value of one or more of the concentrations obtained at iteration n-1, that is, at time tn_i, as first data. input 207 to obtain the concentration profile at the instant following tn. It will be advantageous to use at each iteration the one or more points of the column as concentrations serving as first input data 207.

In this regard, it should be noted that FIG. 7 shows the principle of operation of the method according to the invention at a given point of the column to which the concentrations 201, 202, 204 respectively correspond to the moments t, t1 and t2. . However, thanks to this process, what is obtained is a complete profile of the concentrations along the column, of the type illustrated in FIG. 6. This gives particularly rich and reliable information. The pitch chosen between the instants t, t1, t2, ... tN for the prediction is, for example, greater than the pitch used for solving the equations mentioned above. It is, in particular, 1 to 10 minutes. It can be constant. The time range illustrated by the arrow marked 250 corresponds to the prediction horizon used, prediction horizon divided into as many future times t1, t2, tN as desired, for example of the order of ten. As more particularly illustrated in the upper graph of FIG. 7, said prediction step is performed by performing a step of determining a second input data item 208 of the model from a physical parameter 212 of said unit, called parameter variable, determination step in which said variable parameter 212 is linearly varied from a present value 252 of said variable parameter 212. An increasing or decreasing orientation of said variable parameter 212 may be determined. present instant t and performing said step of determining the second input data 208 while maintaining said orientation. Here it can be seen that said variable parameter 212 is increasing at the instant t, with a steering coefficient 3, and the second input data 208 coming from said variable parameter 212 will be located on the line 214 starting from the value 252 and 3. This said prediction step 200 may also be performed by performing a step of determining a third input data 210 of the model from another physical parameter 216 of said unit, said fixed parameter, determining step wherein said fixed parameter 216 is held fixed, in particular at its value at the instant t. Here it is found that said fixed parameter 216 which serves to determine the third input data 210 is stabilized along a line 218 located at a point 209 corresponding to the value of said fixed parameter at the present moment and said thirds input data will be located on this line 218.

However, the physical parameter (s) used as additional input data may have known future values, and the known values of said physical parameters will be used at each iteration. This is the case, for example, of a production flow that may have been planned. According to an advantageous embodiment, with respect to the physical parameter (s) employed as additional input data for which the future value is unknown, one of the above-mentioned methods for determining the value can be used. future, while for the physical parameter (s) presenting known future values, said known values will be used. As more particularly illustrated in the lower graph of FIG. 7, said prediction step 200 is performed by holding the setting parameter fixed, in particular at its value at the instant t. As can be seen from the foregoing, it is of course also used as input data 220 of the model to determine future concentration profiles.

At the first iteration t + 1, a first profile, comprising the said concentration 202, is obtained by virtue of the said prediction step 200, which here exploits the said model with the input data coming from the concentration 201 at the instant t, the present values 252, 209 of the variable parameter 212 and the fixed parameter 216 and the setting parameter 220. At the second iteration t + 2, a second profile, comprising said concentration 204, is obtained by said prediction step 200 which, at Again, this model is used, this time with the input data from the concentration 202 at time t1, the future values 208, 210 of the variable parameter 212 and the fixed parameter 216 and the setting parameter 220. And said step prediction continues iteratively until the last iteration N, only the first two iterations have been illustrated here. As represented in FIG. 8, it will be possible to update said prediction periodically. Thus, in an updated version, the present and future concentrations 202, 204 and the other data mentioned above, but shifted in time, are found again. It is found in particular that the value of the second fixed parameter 216 is updated and the new line 218 is lower than the old one. Similarly for the setting parameter for which the value 220 is here also lower. Said updating takes place, for example, according to a frequency ranging from one minute to ten minutes. In a particular embodiment, the update takes place at time t1 of the previous prediction. In other words, the new instant present t reference is the t1 of the previous prediction. That being so, according to a first variant, not illustrated, said method comprises a step of comparing the concentration of at least one of said components, in at least one point of said column, coming from one of said future profiles, with one or more reference concentrations of said component at the same point of the column. It may also be a comparison with the concentration of the same component, at the same point of the column, from another of said future profiles. This is called a local approach to determining whether a malfunction exists or not. It is concluded that a malfunction exists, for example, if the predicted concentrations are too far from nominal concentrations or too close to limit concentrations and / or remain distant or close to said concentrations for too long a time. The same may be true if the predicted concentrations deviate too quickly from said nominal concentrations or come too close to said limit concentrations. According to another variant, illustrated in FIG. 9, said method comprises a step of comparing one of the future profiles 300 of at least one of said components with a reference profile 302, 304 and / or with another of the profiles. said component, at least for certain longitudinal portions of the column, here all along the column. We are talking about a global approach. In this case, it is possible to work on integrating the error between the future profile or profiles 300 and the reference profile or profiles, one of them 302 corresponding, for example, to a nominal profile and / or the other. 304 to a profile associated with the existence of a dysfunction, here a puff of nitrogen. It will also be possible to work by correlation, in particular by studying the location of the correlation peak. It may also be used to establish a material balance of one or more sections of the column; a negative or positive balance sheet potentially indicating a destabilized situation. Figure 9 corresponds to a concentration profile of argon in a low pressure column. This is a bell-shaped profile having a central lobe defined, for example, as the portion of said bell-shaped profile located between two points of inflection of said profile. In such a case, it will be possible to carry out said comparison step by working, in particular, with measurements of the recovery of the lobe of a reference profile by the estimated lobe. We can also quantify the centering, symmetry or flattening of the estimated lobe, or its distortion with respect to a reference lobe, through the same quantities, among others. It will also be possible to carry out said comparison step by working on S-shaped profiles, in particular by using their point of inflection, particularly with regard to the oxygen profile in the crude argon column. The point of inflection is indeed advantageous in that it is strongly related to the position of the wavefront concentration, wave whose movement summarizes much of the dynamics of the process.

However, said method according to the invention may comprise a step of processing results of said comparison step, said processing step being performed by assigning a different weight to the results corresponding to one or more particular points of the column to decide to issuing an alert. The one or one of said particular points is, for example, a withdrawal point.

The device for determining the concentrations mentioned above may of course be configured to allow such detections of malfunction risks. It comprises in this sense means for predicting at least one future profile of the concentrations of said components along the column using said model. Said means may exploit for this purpose a computer program executed, for example, by the processor of the device 50 for determining the concentrations along the column. Said computer program is possibly stored on a recording medium.

It may be noted that, according to one aspect of the invention, the detection of the risks of nitrogen flushes is done by monitoring the concentration profiles of one of the high or low pressure units while the nitrogen puff is an intervening phenomenon. in the crude argon column. In general, according to this aspect of the invention, predictions are made on one of the columns of the unit to monitor a malfunction occurring in another of the columns of said unit.

Claims (15)

REVENDICATIONS1. A method for detecting a risk of malfunction in a unit (2) for separating the chemical components of a product, in particular air, comprising a column (14, 26) for the distillation of said product, said process using a model for estimating the concentration of the components along the column (14, 26), said model using a setting parameter for taking into account changes in the operation of the column, said method comprising a step (200) of prediction at least one future profile of the concentrations of said components along the column (14, 26) using said model. The method of claim 1 comprising a step of determining a profile of the concentrations of all or part of said components along said column using said model, at the present time, the concentrations at the present time serving first input data (207) of the model for performing said prediction step (200), during a first iteration. The method of claim 2 wherein said predicting step (200) is performed using one or more other physical parameters (212, 216) of the unit as additional input data (208, 210) of the model. The method of claim 3 wherein said physical parameter has known future values and said future values are used as additional input data to perform said prediction step. 5. The method as claimed in claim 3, in which: said additional input data item (208) comes from a determination step in which said physical parameter (212) is linearly varied from a present value (252) said physical parameter (212) and / or, the additional input data (210) is derived from a determining step in which said physical parameter (216) is fixed. The method of any one of claims 3 to 5 wherein said predicting step (200) is performed by holding the setting parameter fixed. A method according to any one of the preceding claims comprising a step of processing results of a comparison step, said processing step being performed by assigning a different weight to the results corresponding to one or more particular points of the column (14). , 26) to decide to issue an alert. 8. The method of claim 7 wherein the or one of said particular points is a drawpoint. The method according to any of the preceding claims, further comprising, for the time being, the steps of: - measuring the concentration of at least one of said components in at least one location in the column (14, 26) ); and - setting the model using the setting parameter determined from the measured concentration. 10. The method of claim 9 wherein: - the concentration of said component is estimated at said location of the column (14, 26) where the measurement takes place using said model with a first value of said adjustment parameter, - an error is established between the estimated value and the measured value of the concentration, - a second value of the adjustment parameter is established as a function of said error, - the first value of the adjustment parameter is replaced by the second value in said model. The method according to any one of the preceding claims wherein said model uses a propagation term in relation to a convection of said components along the column (14, 26) and an axial diffusion term in relation to a diffusion of said components. in the column (14, 26), the setting parameter for weighting the effects of diffusion with respect to the effects of propagation. 12. Device for detecting a risk of malfunction in a unit (2) for separating the chemical components of a product, in particular air, comprising a distillation column (14, 26), said device comprising means for placing implementing a model for estimating the concentration of the components along the column (14, 26), said model using an adjustment parameter making it possible to take into account variations in the operation of the column (14, 26), said device further comprising means (50) for predicting at least one future profile of the concentrations of said components along the column using said model. 13. Air separation unit (2) comprising at least one air distillation column (10, 14, 26) and a device according to claim 12, for detecting a risk of malfunction, in particular puffs of air. nitrogen, in the one or one of said distillation column. 14. Computer program comprising instructions for carrying out the method according to any one of claims 1 to 11, when the program is executed by a processor. Recording medium in which the program according to claim 14 is stored.