Computer-implemented method for operation of a chemical production plant

Abstract

A computer-implemented method for operation of a chemical production plant (10), the chemical production plant (10) being configured to produce a chemical compound from at least one reactant in a reactor (12), wherein the chemical production plant (10) comprises means (38) for determination of process data (40), wherein such process data (40) are used to manipulate at least one influence parameter (54) influencing at least one target parameter, the method comprising the following steps: a) acquiring current values of the process data (40), b) feeding the current values of the process data into a trainable algorithm (42) which has been configured to determine a value of the influence parameter (54) from the current values of the process data (40) and historical values (52) of the process data, wherein the determined influence parameter (54) is suitable to achieve a desired value of the at least one target parameter, and c) applying at least one influence parameter (54) determined in step b) for the operation of the chemical production plant (10).

Description

[0001] 2024PF30130-Foreign Countries

[0002] - 1 -

[0003] COMPUTER-IMPLEMENTED METHOD FOR OPERATION OF A CHEMICAL PRODUCTION PLANT

[0004] The present teachings relate to computer-implemented methods for operation of a chemical production plant. Further, the present teachings relate to a computer software product.

[0005] US 2024 / 144043 Al discloses a method for training a machine-learning module of a computer-implemented prediction model for predicting product quality parameter values for one or more quality parameters of a chemical product produced by a chemical production plant. The production plant includes a plurality of sensors, each of which is configured to acquire process parameter values for one or more process parameters of a chemical process carried out by the production plant for producing the chemical product during operation of the production plant. A priori information about the production plant and the process carried out by the production plant is used, including chronological sequence information about a chronological sequence of the process carried out within the production plant, for which sensors sensor-specific time shifts between an acquisition time of training process parameter values and a production time of a product unit, during the production of which the corresponding training process parameter value was acquired.

[0006] EP 3924785 Bl discloses systems and methods for determining an operating condition of a chemical production plant including at least one catalytic reactor are provided. Via a communication interface operating data and a catalyst age indicator are received. At least one target operating parameter for the operating condition of a scheduled production run ora current production run are determined. The at least one target operating parameter for the operating condition may be used for monitoring and / or controlling the chemical production plant.

[0007] Various chemical production plants for producing chemicals on a large scale are known in the art. One specific field, where chemical production plants come into use is for the production of chemical compounds, in particular the production of polymers and more particularly the production of different kinds of polyesters. Large production plants for synthesis of polyesters usually allow for the production of various basic esters in different parallel reaction lines, in order to produce 2024PF30130-FC - 2 -

[0008] multiple, typically over 50, different basic esters and also at least the same number of different polyesters as final products.

[0009] The production of polyesters involves a series of chemical reactions, typically starting with the esterification or transesterification of dicarboxylic acids and alcohols e.g., diols. Additionally, the production plant is typically equipped with distillation systems for the distillation (evaporation) of water and purification of polyester and for alcohol distillation. Further the addition of a catalyst and / or additives may be possible. Finally, the gained polymer can be stored in storages or production tanks. The production of polymers and particularly polyesters is often performed in a discontinuous process having a non-steady-state operation and changing process conditions.

[0010] As used herein, a "discontinuous process" is understood as a process whereby during operation, at least one of the reactants does not enter or leave the system during a certain period of time. For example, a process in which all reactants are introduced into a reaction vessel which is then kept for a certain period of time at a certain temperature and / or temperature profile and pressure and / or pressure profile to allow for a chemical reaction to proceed, after which period of time the product is removed from the vessel, would be considered a "discontinuous process" within this meaning. A process as described in this example may also be referred to as a "batch process". Variants of this kind of batch process are known and referred to in the literature as "fed batch" or "semi batch" processes. For example, one reactant may be added to a reaction vessel and another reactant fed slowly thereto or all reactants may be added slowly to a reaction vessel followed by a period of time (which can be short) for completion of the reaction. It is also possible that parts of the product are removed during operation. All of these examples are "discontinuous processes" within the above definition. Such a discontinuous process is characterized by a non-stationary state during operation. The composition of the reaction mixture varies over time and is at a given point in time - at least in an ideal model representation - identical in any part of the reaction vessel.

[0011] In contrast thereto, in a "continuous process", a stationary state (or steady state) exists in production during which a defined flow of reactants enters a reaction vessel and a corresponding flow of products leaves the reaction vessel at the same 2024PF30130-FC - 3 -

[0012] time. With the exception of special circumstances such as start-up, shutdown and changes of production capacity, a continuous production is characterized by such a stationary state. In this stationary state, the composition of the reaction mixture does not vary over time, but changes as the reaction mixture passes through the vessel, i.e. at any point in time the composition of the reaction mixture is identical in a given infinitesimally small volume element of the reaction vessel.

[0013] As can already be derived from the above, the operation of such chemical production plants is highly complex. The complexity of a chemical production plant lies in particular in the need to precisely control reaction conditions such as e.g., temperatures, pressures, and reactant and catalyst flow to ensure a high-quality chemical compound output, in particular polymer output by maximizing the resource and energy efficiency and minimizing the manual workforce effort. The integration of these various processes and equipment requires sophisticated engineering and operational expertise to maintain efficiency and product consistency.

[0014] Already today, chemical production plants are automated up to a high degree. Usually, chemical production plants are equipped with multiple sensors, management systems for production planning, process systems, controlling systems, alarm management systems and more. However, mainly due to the complexity of the production plant, further optimization of the production process in order to save time and resources is very difficult and usually requires experienced personnel and a tightly meshed process control. Also, automation systems are optimal for repetitive and predefined tasks, which in particular is suitable only to a limited degree to discontinuous processes with non-steady process conditions.

[0015] Thus, the underlying object of the present invention is to provide computer-implemented methods and computer software products to optimize the operation of a chemical production plant in order to save resources and reduce energy and time consumption and manual workforce effort and yield a higher output at highest quality of the chemical compound, in particular at highest quality of the polymer.

[0016] According to the present teachings a computer-implemented method for operation of a chemical production plant is provided. The chemical production plant is configured to produce a chemical compound, in particular a polymer and more 2024PF30130-FC - 4 -

[0017] particularly a polyester from at least one reactant in a reactor. A "chemical compound" as used herein is understood to mean any substance or substance mixture that can be produced by any chemical conversion process. Particularly preferred examples of chemical compounds within the meaning of the invention are polymers. A "polymer" is a substance or material that consists of very large molecules, or macromolecules, which are constituted by many repeating subunits, at least two subunits, derived from one or more species of monomers. The polymer is in particular a polyester. The chemical production plant can be designed to produce multiple different chemical compounds, in particular multiple different polymers and also even more particularly multiple different polyesters.

[0018] The chemical production plant may comprise a production line with a reactor with inlets and outlets for reactants, products, additives and / or catalysts. The reactor can be cooled or heated by means of a jacket or coil. The reactor may have an external product circulation externally cooling the reaction product. The chemical production plant may comprise a distillation for the distillation of water and alcohol, including a head condenser and a partial condenser and a distillate vessel. The production set up may include a vacuum station for introduction of the vacuum or a nitrogen inlet for pressurizing the reactor. Further downstream the reactor the chemical production plant may comprise a thin film evaporator and / or a post-pro-cessing device, in which additives are inserted into the reaction product. The final product can be stored in one or more storages. The product can be transferred to a storage / tank via a pigging system.

[0019] The chemical production plant can comprise at least one production line or also several production lines which are operated in parallel. In a non-limiting example, the chemical production plant comprises more than 10 and particularly more than 30 or more than 50 parallel production lines. Each production line is functionally connected to one or more reservoirs with reactants, catalysts and / or additives (in the further course also referred to as raw materials) and one or more storage for the final product. The chemical production plant may be designed to produce in particular multiple different monomers, dimers or trimers and even more particularly so-called basic esters to produce multiple different polyesters as final products. 2024PF30130-FC - 5 -

[0020] The chemical production plant further comprises means for periodical or continuous determination of process data. The process data are subsequently used to manipulate at least one influence parameter influencing at least one target parameter. Examples for process data, influence parameter and target parameter will be described in detail in the following.

[0021] Process data may be any data, which can be gained in a running process, and which describe the running process. Process data can for example be gained via measurements, from recipes and / or from controlling systems. Process data may be heterogenous and can be quantitative or qualitative, time-continuous or timediscrete.

[0022] Influence parameters are used to influence the operation of the chemical production plant and the running production process. An influence parameter can be for example a standard control parameter such as pressure or temperature inside the reactor. The influence parameter determines the behavior of the ongoing production process and thus the resulting product and resulting process.

[0023] Target parameters are defined which characterize a desired course of a production process and / or a desired quality of the final product.

[0024] The method comprises the following steps:

[0025] a) First current values of the process data are acquired from the running process. Here, process data gained from sensors, recipes and / or different controlling systems are gathered from the running production process.

[0026] b) These acquired current values of the process data are fed into a trainable algorithm which has been configured to determine values of the influence parameter from the current values of the process data and historical values of the process data. The determined influence parameter is suitable to achieve a desired value of the at least one target parameter. In other words, the trainable algorithm is capable of determining values of one influence parameter or values of a set of influence parameters, which will lead to an operation of the production process to achieve the defined target parameter(s). 2024PF30130-FC - 6 -

[0027] Historical values of the process data are defined as values of the process data, which have been recorded and acquired before the actually running process. Historical values may be acquired from one or more preceding batches. The historical values may include all available "older" process data or it is also possible to exclude specific historical process data, e.g., from a specific time range or which exhibit some kind of disturbance or anomaly.

[0028] The historical values may be updated by including process data of preceding operation of the production plant under application of the herein described method.

[0029] c) The at least one influence parameter determined in step b) is applied for the operation of the chemical production plant.

[0030] In a particularly preferred embodiment, the method comprises the following steps:

[0031] a) acquiring current values of the process data (40),

[0032] b) feeding the current values of the process data (40) into a trainable algorithm (42) which has been configured to determine a value of the influence parameter (54) from the current values of the process data (40) and historical values (52) of the process data, wherein the determined influence parameter (54) is suitable to achieve a desired value of the at least one target parameter, and

[0033] c) applying the at least one influence parameter (54) determined in step b) for the operation of the chemical production plant (10), preferably a discontinuous process,

[0034] wherein the at least one influence parameter (54) is a necessity of taking a sample, a necessity of calibrating a sensor and / or a discharge temperature of the chemical compound, preferably a necessity of taking a sample, wherein the necessity of taking a sample may be carried out at selected times within the running process,

[0035] wherein, under the assumption that each measurement taking during the course of a process has a particular inaccuracy, these inaccuracies of the processes are added in order to calculate a quality distribution of the chemical compound, in particular the polymer in the product, and, if the quality 2024PF30130-FC - 7 -

[0036] distribution lies at least partially outside a predefined confidence interval, then a necessity for taking a sample is determined,

[0037] and

[0038] wherein, if the sample is taken out of the reactor, the sample is analyzed and if the quality parameters of the sample are inside a predefined confidence interval, then the process is continued,

[0039] wherein, most preferably, the influence parameter (54) determined in step b) is applied to a discontinuous process in step c).

[0040] Using a trained algorithm as described above allows the transition from an automated operation of the chemical production plant to an essentially autonomous or even fully autonomous operation. With the present teachings the operation of a chemical production plant requires much less resources, as e.g., the amount of reactants or raw materials, the process time and consequently energy consumption is reduced, whereby the desired quality of the final product is maintained at a high level.

[0041] The trained algorithm may also be trained to detect an anomaly in the current process from the current process data. In case, an anomaly is detected by the trained algorithm, the application of the determined influence parameters in step c) can be accompanied by the generation of an alert. In addition or as an alternative, in such a case of an anomaly, a necessity to take a sample from the reactor to analyze the quality of the reaction product may be determined by the trained algorithm.

[0042] In a non-limiting embodiment, the at least one determined influence parameter is applied to a discontinuous process in step c). The chemical production plant may be configured to produce chemical compounds, in particular polymers discontinuously (batch-wise). In a discontinuous process, the chemical reactions occur over a determined reaction period, and the final product is removed at the end of the reaction period. Between two batches, the production plant and particular the reactor may be cleaned. A discontinuous process is characterized by its discrete, non-continuous nature, wherein each batch is processed separately. The behavior of a discontinuous process over time is non-linear, making it difficult to control and predict the course of the process and the quality of the final product. The 2024PF30130-FC - 8 -

[0043] production of chemicals in a discontinuous process is complex and ensuring consistent product quality across different batches is challenging. Variations in reactants, reaction conditions, and operator handling can lead to batch-to-batch variability. Discontinuous processes often involve significant downtime for cleaning, setup, and maintenance between batches. This can reduce overall production efficiency and increase operational costs. Maintaining optimal reaction conditions (temperature, pressure, reactant concentration) throughout the entire batch is very difficult. By means of the present teachings a highly reproducible operation of the chemical production plant can be achieved also in a discontinuous process.

[0044] In an alternative, the influence parameter may also be applied to a continuous process in step c). After a starting phase, in a continuous process, reactants, catalysts and / or additives are continuously fed into the reactor, and products are continuously removed. The chemical reactions occur in a steady-state manner, allowing for the constant production of the final product.

[0045] Further alternatively, the influence parameters may also be applied to a semi-discontinuous process.

[0046] In a further non-limiting embodiment, the determined parameter of step b) is applied to a production process of a chemical compound, in particular of a polymer and even more particularly of a polyester, preferably a hydroxyl group-terminated polyesterol. A polyester is a polymer that contains at least two ester linkages in every repeating unit of the main chain. Polyesters, preferably hydroxyl group-terminated polyesterols, are typically produced industrially through condensation reactions by reacting a polyol with a polycarboxylic acid and / or a cyclic anhydride, often in the presence of catalysts, forming the polyesterol and water, whereby the water is removed from the reaction system.

[0047] Typically, polyesters, in particular hydroxyl group-terminated polyesterols have calculated molecular weights of from 500 g / mol to 5000 g / mol.

[0048] Typical examples for polyols are ethylene glycol, diethylene glycol, triethylene glycol, 1,2-propylene glycol, dipropylene glycol, 1,3-propanediol, 1,3-butanediol, 1,4-butanediol, 1,6-hexanediol, 1,12-dodecanediol, neopentyl glycol, trimethylolpropane, triethylolpropane, castor oil, and glycerol. 2024PF30130-FC - 9 -

[0049] Typical examples for polycarboxylic acids are terephthalic acid, oxalic acid, malonic acid, succinic acid, glutaric acid, adipic acid, pimelic acid, suberic acid, azelaic acid, sebacic acid, undecanedioic acid, dodecanedioic acid, brassylic acid, tetradecanedioic acid, thapsic acid, maleic acid, fumaric acid, citraconic acid, mesaconic acid, and furandicarboxylic acid.

[0050] Typical examples for cyclic anhydrides are succinic anhydride, maleic anhydride, glutaric anhydride, itaconic anhydride, citric anhydride, cis-aconitic anhydride, 2-(2'-carboxyethyl)maleic anhydride, l-methyl-2-(2'-carboxyethyl)maleic anhydride, phthalic anhydride and trimellitic anhydride.

[0051] In order to form a polyester, the reactants are usually charged into the reactor and then converted to the product, although the reactor volume can typically only be partially utilized due to water formation.

[0052] The method steps a), b) and / or c) may be started and carried out during the whole process time of one batch.

[0053] As an alternative, the method steps a), b) and / or c) may only be started and carried out at single process steps, for example during a solvent (for example an alcohol) distillation phase of the production process of a chemical compound, in particular of a polymer and even more particularly of a polyester. In particular, for a target parameter such as a short process time and an influence parameter such as taking a sample from the reaction and analyzing it, the method steps are started and carried out in an alcohol distillation phase. In a non-limiting example the method steps a), b) and / or c) are started and / or performed when an acid value of the polymer below 2,5 and preferably an acid value below 2,1 and more preferably an acid value of 2 or smaller is reached.

[0054] The influence parameters in step b) are preferably continuously determined and continuously applied to the process in step c). It may also be possible to perform at least steps b) and c) only periodically, wherein values of the control parameters are applied only after defined time intervals. Newly determined influence 2024PF30130-FC - 10 -

[0055] parameters may be applied to the process in intervals of less than 10 minutes, preferably less than 5 minutes and more preferably less than 1 minute.

[0056] The acquired process data of the running process and / or historical process data may be any kind of data describing the production process. The process data may comprise one or more of the following data:

[0057] Measurement data from a sensor, in particular measurement data characterizing the polymer itself, measurement data characterizing the polymerization reaction, measurement data characterizing the batch distillation system or measurement data characterizing auxiliary subprocesses. The process data may be results from an in-line sensor, for example an NIR. (near-infra-red) sensor or results from a database offline of the production line, e.g. analysis results for a sample taken from the reactor. The process data may further comprise data acquired by a production controlling and supervision, such as a reactor pressure, a reactor temperature, reactants and / or catalysts or additive (raw material) flow rates, weight of the solid raw materials, pressure and flow rates in pneumatic conveying lines of the solid raw materials, temperature and pressure measurements over the batch distillation column packing, cooling and chilled water temperature in the condensers of the distillate vapors, level in the distillate vessels for both aqueous and organic distillate, reflux flow rate to the distillation column, temperature of the distillate, online and offline analytical measurement results of the both aqueous and organic distillate, pressure and pressure difference measurements for assessment of fouling in the heat exchangers and vacuum pumps, temperature of the operation fluid of the liquid ring vacuum pump, the temperatures in the heating steam distillate for assessment of the steam trap performance, as well as redundant measurements for estimation of the measurement errors. Further, process data may be the positions of the control valves, the operation modes of the controllers and information about process alarms and messages. Process data may be data from a recipe, a control system, a production management system and / or a production planning tool. Further process data may comprise amounts of raw materials, time stamps for start and end as well as duration of various batch steps and functions, recipe parameters like set points, operation modes, functional signals, information about the raw material supply like a number of tank or silo, scheduling information about the sequence of production and product transfer or shipment batches. 2024PF30130-FC - 11 -

[0058] The at least one influence parameter may be one or more of a temperature, a flowrate or a pressure of a coolant, of reactants, of the reactor, the catalysts and distillation and / or evaporation, preferably a temperature, a flowrate or a pressure of a coolant, of reactants, of the reactor, of a distillation and / or evaporation. An influence parameter can be every parameter, which influences the behavior of the chemical production plant and the course of the production process.

[0059] Further, an influence parameter may also be the necessity (yes / no) of taking of a sample from a reactor in order to analyze the quality of the polymer. Taking of samples usually has a non-negligible impact on the process time, as taking a sample and analyzing it takes several minutes time and during this time, some further process steps cannot be performed.

[0060] The determination of the necessity to take a sample in step b) and taking a sample in step c) may be carried out at selected times within the running process. Steps b) and c) may be performed during the distillation of alcohol. In a non-limiting example step c) and / or d) are performed, when an acid number of the polymer is between 1.8 and 2.2 and preferably 2. In addition or alternatively, steps b) and / or c) are performed, when an acid number lies between 0.33 and 0.37 and preferably is 0.35.

[0061] In a further non-limiting embodiment, the necessity (yes / no) of calibrating a sensor, in particular an in-line NIR. sensor, may be an influence parameter, which influences the process time as target parameter. If a calibration of a sensor is necessary, this is usually done by taking a sample from the polymer and analyzing it offline.

[0062] Further, an influence parameter may be a discharging temperature of the chemical compound, in particular of the polymer and even more particularly of the polyester. After the reaction, and particular the solvent (for example: alcohol) distillation phase, the chemical compound, in particular the polymer and even more particularly the polyester has to be cooled before discharging it into a storage. The discharging temperature may be dependent on different parameters, such as the filllevel and temperature in the storage. A lower discharging temperature and 2024PF30130-FC - 12 -

[0063] corresponding cooling parameters for the cooling process, will then influence the overall process time as one target parameter.

[0064] The at least one target parameter may be a target parameter of the chemical compound, in particular of the polymer and even more particularly of the polyester. The target parameter can be a value which reflects the desired quality of the final product, e.g. of the polyester. In particular, the at least one target parameter is a defined acid value of the chemical compound, in particular of the polymer and in more particular of the polyester. The target parameter may additionally or optionally be a defined hydroxyl value and / or a defined viscosity of the chemical compound. These target parameters are in particular preferred if the chemical compound is a polymer and is in particular a polyester. Any other result of a quality measurement or a soft sensor may be used as target parameter, regardless whether the chemical compound is a polymer or not. The trainable algorithm is preferably trained to determine the influence parameters in order to achieve the desired values of multiple even interdependent target parameters.

[0065] The at least one target parameter can be a target parameter of the process, in particular a process time. As one target, the process time can be minimized, to reduce consumption of resources and energy. Already by increasing the autonomy of the production process, the intervention of personnel is drastically reduced, leading to a lower process time. In order to achieve a desired short process time, the trainable algorithm has been trained with historical process data. Particular historical data may be excluded, for example historical process data belonging to processes with particular long process times and / or with particular disturbances. Particularly those processes with a process time above a specified maximum value are excluded, which process time can be attributed to failure of equipment of the chemical production plant, where no optimization of the process is possible, but rather a repair of the chemical production plant is necessary. The trainable algorithm can also be configured to determine the necessity of a repair, being trained on these process data with particular long process times.

[0066] In a further non-limiting embodiment, the trainable algorithm is an ensemble of supervised and unsupervised machine learning algorithms. The trainable algorithm 2024PF30130-FC - 13 -

[0067] may be based on an Isolation Forest algorithm (anomaly detection) and / or a XGBoost algorithm (process outcome prediction).

[0068] As already described above, in step c) additionally to the application of determined values of the influence parameters a warning may be generated if the trainable algorithm determines an anomaly in the production process. The trainable algorithm has been trained to determine anomalies in the production process from historical process data. Such an anomaly may be a failure of heating or cooling supply, a blockage of a solid raw material pneumatic or mechanical conveying line, a disturbance of a motor or pump, interlock based on exceeding of one of the upper or lower limits of any process variable, disturbance of a raw material supply etc.

[0069] The present teachings also relate to a further computer-implemented method for operation of a chemical production plant, in particular a chemical production plant as described above. The chemical production plant is configured to produce a chemical compound, in particular a polymer and even more particularly a polyester from at least one reactant in a reactor. For further details with respect to the chemical production plant it is referred to the description above.

[0070] The chemical production plant comprises means for periodical or continuous determination of process data. For the definition and embodiments of process data it is referred to the description above.

[0071] Further the chemical production plant comprises a sample taking system to take samples of the chemical compound, in particular of the polymer, and even more particularly of the polyester in order to analyze the quality of the chemical compound, in particular of the polymer, and even more particularly of the polyester. A sample which is taken out from the reactor is much easier to analyze than inside the reactor and therewith connected pipes and equipment.

[0072] The method comprises the following steps:

[0073] a) acquiring current values of the process data,

[0074] b) feeding the current values of the process data into a trainable algorithm which has been configured to determine the necessity to take a sample of 2024PF30130-FC - 14 -

[0075] the polymer from the current values of the process data and historical values of the process data,

[0076] c) taking a sample if a necessity of taking a sample has been determined by the trainable algorithm.

[0077] With the method according to the present teachings, the degree of autonomy of the operation of the chemical production plant is increased further. The number of samples taken out of the reactor can be reduced, leading to several advantages. First, the process time is significantly reduced, as taking samples and analyzing the same usually takes 15 minutes or more. By reducing process time, also energy consumption and resources are saved. Second, the samples taken from the process usually have a high temperature, sometimes of about 200 °C, which is in inherent risk for the personnel taking the sample to suffer burns. By reducing the number of samples, this risk is also minimized.

[0078] Regarding the definition of process data and also historical process data it is referred to the description above. Process data in particular comprise measurement results of an in-line NIR. Historical data may comprise measurements results of an NIR and corresponding measurement results of a sample taking from the same batch. Basically, when an anomaly of the process or the measurement results is detected by the trainable algorithm, the necessity to take a sample is detected.

[0079] The necessity to take a sample may be regarded as an influence parameter, which influences at least the process time. The necessity is chosen to achieve target parameters, such as short process duration and / or good quality of the polymer.

[0080] The determination of the necessity to take a sample in step b) and taking a sample in step c) may be carried out at selected times within the running process. Steps b) and c) may be performed during the distillation of alcohol. In a non-limiting example step c) and / or d) are performed, when an acid number of the chemical compound, in particular of the polymer and even more particularly of the polyesters is between 1.8 and 2.2 and preferably 2. In addition or alternatively, steps b) and / or c) are performed, when an acid number lies between 0.33 and 0.37 and preferably is 0.35. 2024PF30130-FC - 15 -

[0081] In a further non-limiting embodiment of the present teachings, the trainable algorithm belongs to the class of a supervised machine algorithm. In particular, the trainable algorithm is an algorithm based on binary trees. The trainable algorithm may e.g., be an isolation forest algorithm. The isolation forest algorithm is suitable for the detection of anomalies within the measurement data, which makes it necessary to take a sample.

[0082] If in step b) a necessity of taking a sample is positively determined and a sample is taken in step c), then the sample is subjected to an analysis to quantify the chemical compound, in particular the polymer and even more particularly the polyester quality. If the analysis reveals that the quality of the chemical compound, in particular of the polymer and even more particularly of the polyester is within a defined range, then the process is continued with unchanged influence parameters. In particular, a cooling phase can be started after a quality of a final product is confirmed by the analysis of the sample. If the analysis reveals that the quality of the polyester lies outside of a defined range, then the influence parameters are adapted accordingly. This correction of the influence parameters can be determined by a trainable algorithm designed to determine influence parameters to achieve a predetermined target parameter and trained with historical process data. The correction of the influence parameters may also be determined by an equation-based control circuit.

[0083] Further, differences of results of an in-line sensor, e.g., NIR. sensor, and results from the analysis of the sample may be compared and possible deviations may be taken into account in the further course of the process, when acquiring measurement data based on this in-line sensor.

[0084] The steps b) and c) may also be performed when the final product is filled in a storage in order to determine whether the product quality has a desired value and the final product is ready to deliver or whether a sample should be taken.

[0085] If in step c) a sample is taken from the final product and subjected to an analysis and the analysis results in a target parameter outside of a defined range, then in a next batch, a correction batch may be initiated, in order to equalize deficiencies of the current sample and / or the final product is corrected by means of a post- 2024PF30130-FC - 16 -

[0086] reaction. Does the analysis reveal a final product which lies within the specifications, the final product is ready for further processing.

[0087] The necessity of taking a sample is further determined under the aspect of enhancing the accuracy of the prediction of the quality distribution in the final product. Quality parameters are in particular an acid value and / or a hydroxy value. The final products of several batches are usually mixed and stored together on one storage. Under the assumption that each measurement taking during the course of a process has a particular inaccuracy, these inaccuracies of the processes are added in order to calculate a quality distribution of the chemical compound, in particular the polymer in the product. If the quality distribution lies at least partially outside a predefined confidence interval, then a necessity for taking a sample is determined. The predefined confidence may be at least 90% and preferably at least 95%. The sensor may be an in-line NIR. sensor, which is known to have a high and also fluctuating measurement accuracy. In particular, the trainable algorithm may be further configured to determine the accuracy of the present prediction. As an alternative or additionally, the accuracy may be determined by means of a Monte-Carlo simulation.

[0088] The invention also relates to a further computer-implemented method for operation of a chemical production plant, wherein the chemical production plant is configured to produce a chemical compound, in particular a polymer and even more particularly a polyester from at least one reactant in a reactor, wherein the chemical production plant comprises means for periodical or continuous determination of process data. For details with respect to the chemical production plant it is referred to the description above. The chemical production plant further comprises a sensor in order to prove the quality of the chemical compound, in particular of the polymer and even more particularly of the polyester.

[0089] The method comprises the following steps:

[0090] a) acquiring current values of the process data,

[0091] b) feeding the current values of the process data into a trainable algorithm which has been configured to determine the necessity of calibrating the sensor from the current values of the process data and historical values of the process data, 2024PF30130-FC - 17 -

[0092] c) calibrating the sensor if the necessity of calibration of the sensor has been determined by the trainable algorithm.

[0093] The sensor is in particular an in-line NIR-sensor. Such in-line NIR. sensors are known for their fluctuating measurement accuracy. To reduce the insecurity in quality control, a calibration of the sensor is necessary. Usually, calibration may be performed by means of samples are taken of the chemical compound, in particular of the polymer and even more particularly of the polyester and analyzing these offline. The offline results are then used to calibrate the sensor. However, the taking of samples is costly and takes time. With the method described above, the amount of sample takings can be significantly reduced, as the trainable algorithm is capable of detecting such fluctuations in measurement accuracy. The application of the present teaching further improves the autonomy of the chemical production plant operation.

[0094] For details regarding the process data, it is referred to the description above.

[0095] The determination of the necessity to calibrate and sensor and possibly to take a sample may be carried out at selected times within the running process. Steps b) and c) may be performed during the distillation of alcohol. In a non-limiting example step c) and / or d) are performed, when an acid number of the polymer is between 1.8 and 2.2 and preferably 2. In addition or alternatively, steps b) and / or c) are performed, when an acid number lies between 0.33 and 0.37 and preferably is 0.35.

[0096] In a further non-limiting embodiment of the present teachings, the trainable algorithm belongs to the class of a supervised machine algorithm. In particular, the trainable algorithm is an algorithm based on binary trees. The trainable algorithm may e.g., be an isolation forest algorithm. The isolation forest algorithm is suitable for the detection of anomalies within the measurement data, which makes it necessary to calibrate the sensor.

[0097] The present teachings further relate to a computer-implemented method for operation of a chemical production plant, the chemical production plant being configured to produce a chemical compound, in particular a polymer and even more 2024PF30130-FC - 18 -

[0098] particularly a polyester from at least one reactant in a temperature-controlled reactor and to discharge the polymer into a storage. For further details with respect to the structure of the chemical production plant it is referred to the description above.

[0099] The reactor for the synthesis of the chemical compound, in particular of the polymer and even more particularly of a polyester, is a temperature-controlled reactor, which can be heated and / or cooled. Usually, the reaction is exothermic and after the distillation process the final product has a temperature of about 200°C. Before discharging the final product and for mixing it with reaction products from previous processes and batches, the final product of the current batch usually has to be cooled.

[0100] The method comprises the following steps:

[0101] a) acquiring a first temperature of a heated chemical compound, in particular of a heated polymer and even more particularly of a heated polyester in the reactor. The heated chemical compound, in particular the heated polymer and even more particularly the heated polyester may be the final product.

[0102] The first temperature may be the temperature of a polyester directly after the esterification has been finalized. This may be at the end of an alcohol distillation phase when the acid value and / or hydroxy value reach predetermined values.

[0103] b) acquiring a second temperature of a chemical compound, in particular of a polymer and even more particularly of a polyester in the storage,

[0104] c) determining a discharge temperature and / or a cooling parameter by inputting the measured first temperature and the second temperature into a trainable algorithm configured to determine the discharge temperature and / or a cooling parameter to minimize process duration and energy consumption, wherein the trainable algorithm has been configured to determine the discharge temperature and / or a cooling 2024PF30130-FC - 19 -

[0105] parameter by acquiring historical first temperatures and historical second temperatures, and

[0106] d) applying a discharge temperature and / or a cooling parameter determined in step c) for the operation of the chemical production plant.

[0107] Further to the first temperature and the second temperature, further parameters are taken into account by the trainable algorithm, which may be the fill level of the storage and / or and the ambient temperature.

[0108] As one constraint, a maximum allowable temperature in the storage is defined.

[0109] The present teachings also relate to a computer software product comprising instructions which when executed by a suitable processor cause the processor to perform one or more of the method steps of the above method claims. In particular, the computer software product is configured to acquire process data according to step a), determine influence parameters according to step b) and transmit these process data to a chemical production plant. The computer software product can be run in a cloud, as a web-application, on a desktop computer and / or a mobile device.

[0110] Further embodiments and advantages are described in connection with the following figures:

[0111] Fig. 1 shows a chemical production plant in a schematic representation,

[0112] Fig. 2 shows a schematic representation of a computer-implemented method for operating a chemical production plant according to a first embodiment,

[0113] Fig. 3 shows a flow chart of a computer-implemented method for operating a chemical production plant according to a second embodiment,

[0114] Fig. 4 shows a flow chart of a computer-implemented method for operating a chemical production plant according to a third embodiment, and 2024PF30130-FC - 20 -

[0115] Fig. 5 shows a flow chart of a computer-implemented method for operating a chemical production plant according to a fourth embodiment.

[0116] In Fig. 1 a chemical production plant 10 for producing a chemical compound, in particular a polymer and even more particularly a polyester, is shown. The production plant 10 shown here is designed to produce polyesters in a discontinuous process.

[0117] The chemical production plant 10 comprises a reactor 12 with a first inlet 14 and a second inlet 14 to introduce the reactants (raw materials) in the reactor 12. Here, via the first inlet 14 a polycarboxylic is fed into the reactor 12 and via the second inlet 16 a polyol is fed into the reactor 12. Inside the reactor 12, the reactants react by forming an ester and water.

[0118] The reaction products, ester and water, leave the reactor via an outlet 18 and are feed into a distillation column 20 for water distillation. The distillation column 20 comprises an inlet 22 and an outlet 24 for heating agent, configured to heat up the reaction products ester and water and leading to an evaporation of water. The evaporated water leaves the reactor via a steam-outlet 26. The residual ester is fed back from the distillation column 20 via an outlet 28 to the reactor 12. Further in the course of the reaction, the excess polyols are evaporated in an alcohol distillation step inside the reactor 12.

[0119] A basic ester is then- after cooling the reactor 12 and the basic ester therein -discharged over an outlet 30 to a storage 32. Alternatively, the basic ester may be supplied to a thin film evaporator 34 before being fed into the storage 32. Alternatively, the basic ester may be subjected to a post-treatment device 36 before being fed to the thin film evaporator 34 or also the storage 30.

[0120] Outside of the reactor 12, a jacket 31 is arranged in order to control the temperature inside the reactor 12, for example the reactor 12 can be cooled before discharging the basic ester into the storage 32.

[0121] During the reaction, several process data are acquired. In Fig. 1 only some of the multiple of possible of process data are depicted. Amongst others, process data 2024PF30130-FC - 21 -

[0122] comprise the pressure p inside the reactor 12, the flows f of the reactants, the temperature T inside the reactor 12 via the temperature of the jacket 31 and also the temperature T of the heating agent of the distillation columns 20.

[0123] In addition, the chemical production plant 10 comprises a sensor 38 - here an inline NIR. sensor - to evaluate the acid number and hydroxy number of the produced polymer inside the reactor 12. The measurement results of the sensor 38 are also process data.

[0124] Influence parameters for influencing the course of the process and also the quality of the reaction product / final product are amongst others the pressure p inside the reactor 12, the flows f of the reactants, the temperature T inside the reactor etc.

[0125] Fig. 2 is a schematic figure showing a computer-implemented method to operate a chemical production plant 10.

[0126] The chemical production plant 10 can be a chemical production plant as described in combination with Fig. 1. Here, the chemical production plant 10 is represented by rectangle 10. During operation of the chemical production plant 10, several process data 40 are acquired, which are supplied to a trainable algorithm 42. Process data 40 may comprise measurement results 44 from an in-line NIR 38, a temperature 46 of the reactor 12, a pressure 48 inside the reactor 12, data 50 from a recipe and several more.

[0127] The trainable algorithm 42 is trained with historical process data 52, comprising process data recorded during preceding processes in order to determine influence parameters, to control the chemical production plant 10, to achieve a desired target parameter.

[0128] The trainable algorithm 52 determines influence parameters 54 which are applied to the current process in order to achieve a predetermined value of the target parameters. The determined influence parameter may be a temperature 56 of the reactor 12, a temperature 58 of a distillation column 20 and / or a pressure 60 inside a reactor or a flow 62 of a reactant, etc. 2024PF30130-FC - 22 -

[0129] Fig. 3 is a detailed flow chart of another embodiment of a computer-implemented method to produce a chemical compound, in particular a polymer and even more particularly a polyester in a chemical production plant 10.

[0130] In step SI a polyol is let into the reactor 12, followed by step S2, wherein a polycarboxylic is fed into the reactor 12. Step S3 involves the steps of the water distillation in the distillation column 20.

[0131] Step S4 represents the alcohol distillation. During the alcohol distillation in step S4, at a measured acid number of 0,8, with step S5 the trainable algorithm determines in a first step the necessity to take a sample. The determination is performed based on current process data and historical data. In a second step S6, the determination is further considered under the aspect, whether the predicted distribution of the quality parameter of the final product inside the storage 32 lies within a predefined confident interval.

[0132] If in step S7 it is determined, that neither in step S5 nor step S6 it is determined to take a sample (n), then the process is continued with the course of step S4.

[0133] If in step S7 it is determined, that step S5 and / or S6 resulted in a necessity to take a sample (y), then a sample is taken out of the reactor in step S8 and analyzed. If in step S9 it is analyzed that the quality parameters of the sample are inside a predefined range (y), then the process is continued with step S4.

[0134] Otherwise (n) in step S10 new influence parameters are determined, which are applied to the polyol distillation in step S4. Here for example, further polyol can be added to the reaction or the duration of the alcohol distillation can be extended.

[0135] The steps S5 to S10 can also be performed at another point in time in the course of the process, for example, when an acid value of 0,35 is measured.

[0136] In Fig. 4 a further embodiment of a method is depicted, which is similar to the method described in combination with Fig. 4. Instead of determining the necessity to take a sample of the product in the reactor 12, the necessity to take a sample out of a storage 32 is determined. 2024PF30130-FC - 23 -

[0137] In step SI a polyol is let into the reactor 12, followed by step S2, wherein a polycarboxylic is fed into the reactor 12. Step S3 involves the steps of the water distillation in the distillation column 20.

[0138] Step S4 represents the alcohol distillation. And after the alcohol distillation and after cooling the final product, the product is discharged into a storage and in step Sil stored into a storage 32.

[0139] During the time of storage in step Sil, in step S5' the trainable algorithm determines in a first step, based on the current process data and historical data the necessity to take a sample. In a second step S6', the determination is further considered under the aspect, whether the predicted distribution of the quality parameter of the final product inside the storage 32 lies within a predefined confident interval.

[0140] If in step S7' it is determined, that neither in step S5' nor step S6' it is determined to take a sample (n), then the process is continued with storage step Sil.

[0141] If in step S7' it is determined, that both step S5' and / or S6' resulted in a necessity to take a sample (y), then a sample is taken out of the storage in step S8' and analyzed.

[0142] If in step S9' it is analyzed that the quality parameters of the sample are inside a predefined range (y)', then the process is continued with step Sil.

[0143] Otherwise (n) in step S10' new influence parameters are determined, for example, a subsequent batch can be started having parameters to correct the parameters of the current batch or a mixture with another storage is performed.

[0144] In Fig. 5 a further embodiment of a method is depicted. Here, the method basically relates to the determination of a discharge temperature of a final product from the reactor 12 before being discharged into a storage 32. 2024PF30130-FC - 24 -

[0145] In step SI a polyol is let into the reactor 12, followed by step S2, wherein the polycarboxylic is fed into the reactor 12. Step S3 involves the steps of the water distillation in the distillation column 20.

[0146] Step S4 represents the alcohol distillation. And after the alcohol distillation and after cooling the final product in step S12, the product is discharged into a storage and in step Sil stored into a storage 32.

[0147] During the cooling in step S12, step S13 is performed, wherein the temperature of the chemical compound, in particular of the polymer and even more particularly of the polyester in the reactor 12 and a temperature of a chemical compound, in particular of a polymer and even more particularly of a polyester in the storage 32 are acquired and a trainable algorithm determines a discharge temperature of the chemical compound, in particular of the polymer and even more particularly of the polyester and / or cooling parameters to cool the chemical compound, in particular the polymer and even more particularly the polyester, which are then applied to the cooling process in step S12. 2024PF30130-FC - 25 -

[0148] Reference list

[0149] 10 chemical production plant

[0150] 12 reactor

[0151] 14 first inlet (reactant)

[0152] 16 second inlet (reactant)

[0153] 18 outlet

[0154] 20 distillation column

[0155] 22 inlet (heating agent)

[0156] 24 outlet (heating agent)

[0157] 26 steam outlet

[0158] 28 outlet

[0159] 30 outlet (basic ester)

[0160] 31 jacket

[0161] 32 storage

[0162] 34 thin film evaporator

[0163] 36 post-treatment device

[0164] 38 sensor (NIR. sensor)

[0165] 40 process data

[0166] 42 trainable algorithm

[0167] 44 measurement result (NIR)

[0168] 46 temperature of reactor

[0169] 48 pressure

[0170] 50 data from recipe

[0171] 52 historical process data

[0172] 54 influence parameter

[0173] 56 temperature of reactor

[0174] 58 temperature of a distillation column

[0175] 60 pressure inside a reactor

[0176] 62 flow of a reactant

Claims

2024PF30130-FC - 26 -Claims1. A computer-implemented method for operation of a chemical production plant (10), the chemical production plant (10) being configured to produce a chemical compound from at least one reactant in a reactor (12), wherein the chemical production plant (10) comprises means (38) for determination of process data (40), the process data (40) being used to manipulate at least one influence parameter (54) influencing at least one target parameter, the method comprising the following steps:a) acquiring current values of the process data (40),b) feeding the current values of the process data (40) into a trainable algorithm (42) which has been configured to determine a value of the influence parameter (54) from the current values of the process data (40) and historical values (52) of the process data, wherein the determined influence parameter (54) is suitable to achieve a desired value of the at least one target parameter, andc) applying the at least one influence parameter (54) determined in step b) for the operation of the chemical production plant (10).

2. Computer-implemented method according to the preceding claim, wherein the influence parameter (54) determined in step b) is applied to a discontinuous process in step c).

3. Computer-implemented method according to any one of the preceding claims,wherein the method steps a), b) and / or c) are started and carried out during a solvent distillation phase of the production process of the chemical compound.

4. Computer-implemented method according to any one of the preceding claims,wherein process data (40) and / or historical process data (52) comprise one or more of the following data: measurement data from a sensor, measurement data from an inline-sensor, measurement data from a NIR. sensor, data acquired by a production controlling and supervision, a reactor pressure, a reactor temperature, a cooling jacket temperature, a temperature of an2024PF30130-FC - 27 -evaporator, a flow rate of at least one reactant, a flow rate of a catalyst, a flow rate of an additive, data from a recipe, data from a control system, data from a production management system and / or data from a production planning tool.

5. Computer-implemented method according to any one of the preceding claims, wherein the at least one influence parameter (54) is a temperature, a flowrate or a pressure of a coolant, of reactants, of the reactor, of catalysts, of a distillation and / or evaporation, preferably a temperature, a flowrate or a pressure of a coolant, of reactants, of the reactor, of a distillation and / or evaporation.

6. Computer-implemented method according to any one of the preceding claims, wherein the at least one influence parameter (54) is a necessity of taking a sample, a necessity of calibrating a sensor and / or a discharge temperature of the chemical compound.

7. Computer-implemented method according to any one of the preceding claims, wherein the at least one target parameter is a target parameter of the chemical compound.

8. Computer-implemented method according to any one of the preceding claims, wherein the at least one target parameter is a target parameter of the process, in particular a process time.

9. Computer-implemented method according to any one of the preceding claims, wherein the trainable algorithm (42) is an ensemble of supervised and unsupervised machine learning algorithms.

10. Computer-implemented method according to any one of the preceding claims, wherein in step c) additionally a warning is generated if the trainable algorithm (42) determines an anomaly in the production process.

11. Computer-implemented method according to any one of the preceding claims, wherein the chemical compound is a polymer, and the at least one reactant is a first monomer.2024PF30130-FC - 28 -12. Computer-implemented method according to claim 11,wherein the influence parameter (54) determined in step b) is applied to a synthesis process of a polyester in step c).

13. Computer-implemented method according to claim 11 or 12,wherein the at least one target parameter is a defined acid value, a defined hydroxyl value and / or a defined viscosity of the polymer.

14. A computer-implemented method for operation of a chemical production plant (10), the chemical production plant (10) being configured to produce a chemical compound from at least one reactant in a reactor (12), wherein the chemical production (10) plant comprises means (38) for periodical or continuous determination of process data (40) and a sample taking system to take samples of the chemical compound in order to analyze the quality of the chemical compound,the method comprising the following steps:a) acquiring current values of the process data (40),b) feeding the current values of the process data (40) into a trainable algorithm (42) which has been configured to determine the necessity to take a sample of the polymer from the current values of the process data (40) and historical values (52) of the process data, andc) taking a sample if the necessity of taking a sample has been determined by the trainable algorithm (42).

15. A computer-implemented method for operation of a chemical production plant (10), the chemical production plant (10) being configured to produce a chemical compound from at least one reactant in a reactor (12), wherein the chemical production plant (10) comprises means (38) for periodical or continuous determination of process data (40) and a sensor in order to analyze the quality of the chemical compound,the method comprising the following steps:a) acquiring current values of the process data (40),b) feeding the current values of the process data (40) into a trainable algorithm (42) which has been configured to determine the necessity of2024PF30130-FC - 29 -calibrating the sensor (38) from the current values of the process data (40) and historical values (52) of the process data, andc) calibrating the sensor (38) if the necessity of calibration of the sensor (38) has been determined by the trainable algorithm (42).

16. A computer-implemented method for operation of a chemical production plant (10), the chemical production plant (10) being configured to produce a chemical compound from at least one reactant in a temperature-controlled reactor (12) and to discharge the chemical compound into a storage, the method comprising the following steps:a) acquiring a first temperature of a heated chemical compound in the reactor (12),b) acquiring a second temperature of a chemical compound in the storage, c) determining a discharge temperature and / or a cooling parameter by inputting the measured first temperature and the second temperature into a trainable algorithm (42) configured to determine the discharge temperature and / or a cooling parameter to minimize process duration and energy consumption, wherein the trainable algorithm (42) has been configured to determine the discharge temperature and / or a cooling parameter by acquiring historical first temperatures and historical second temperatures,d) applying a discharge temperature and / or a cooling parameter determined in step c) for the operation of the chemical production plant (10).

17. Computer-implemented method according to any one claims 14 to 16, wherein the chemical compound is a polymer, and the at least one reactant is a first monomer.

18. A computer software product comprising instructions which when executed by a suitable processor cause the processor to perform one or more of the method steps of any one of the above method claims.

Images