AI systems for flow chemistry

Abstract

A computer-implemented method is disclosed for determining at least one set of target parameters for a flow chemistry setup (110) for flow chemistry in a slug. This method is a self-learning method. The method includes the steps of: a) determining at least one process variable using at least one sensor (122) of the flow chemistry setup (110); b) training at least one machine learning model (126) based on the process variable; c) determining the target parameter set by applying an optimization algorithm for at least one optimization objective to the trained machine learning model (126); d) providing and / or considering the determined target parameter set for evaluating the flow chemistry setup (110) and / or evaluating at least one flow chemistry product.

Description

Technical Field

[0001] This invention relates to a computer-implemented method, computer program, computer-readable storage medium, and automatic control system for determining a target set of parameters for flow chemistry settings in a slug. Specifically, this invention can be applied to the industrial production of chemical products and chemical manufacturing. Background Technology

[0002] Methods and apparatus for flow chemistry (also known as continuous flow chemistry) are well known. Generally, flow chemistry involves chemical reactions carried out in a continuous flow. Reactants are combined by pumped fluids (including solutions of reagents) through pipes at known rates. The relative proportions of reactants are controlled by their concentrations and relative flow rates, see https: / / www.nature.com / subjects / flow-chemistry.

[0003] The research and development process involves conducting experiments and analyzing data, which requires a significant amount of manual work. Specifically, chemists formulate research questions and plan experiments, lab technicians conduct the experiments in the laboratory, and chemists or data scientists analyze the experimental data and determine which additional experiments are necessary. Particularly in the first part of a research project, the focus is on screening the parameter space, a process that can be accelerated by using machine learning models for flow chemistry. Machine learning methods for continuous flow chemistry are known from, for example, the following: “Machine learning meets continuous flow chemistry: Automated optimization towards the Paretofront of multiple objectives”, Artur M. Schweidtmann et al., Chemical Engineering Journal, Volume 352, 15.11.2018, pages 277-282; “Efficient multi-objective optimization employing Gaussian processes, spectral sampling and a genetic algorithm”, J. Global Optim. 71(2018)407-438, DOI:10.1007 / s10898-018-0609-2; and “A multi-objective optimization including results of lifecycle assessment in developing bio-renewable-based processes”, ChemSusChem. 10:18(2017)3632-3643.DOI:10.1002 / cssc.201700927, "Closed-loopmulti-target optimization for discovery of new emulsion polymerizationrecipes" by C. Houben et al., Org.Process Res. Dev., 19 (2015) 1049-1053, "Automatic discovery and optimization of chemical processes" by C. Houben et al., Curr. Opinion Chem. Engngn.9(2015)1-7, N. Peremezhney et al., “Combining Gaussian processes, mutual information and a genetic algorithm for multi-targeted optimization of expensive-to-evaluate functions”, Engineering Optimisation, 46(2014)1593-1607; N. Peremezhney et al., “Application of dimensionality reduction to visualization of high-throughput data and building of a classification model in formulated consumer product design”, Chem. Res. Proc. Des. 90(2012)2179-2185. However, despite these achievements, controlling production processes and experiments using continuous flow systems remains challenging.

[0004] Furthermore, methods and apparatus for flow chemistry in slugs are well known. For flow chemistry in slugs, typically two liquids are introduced simultaneously into a tubular reactor, or a dispersed phase is introduced into a continuous phase that flows within the tubular reactor and forms a slug. Therefore, flow chemistry in slugs is discontinuous compared to continuous flow chemistry. Flow chemistry in slugs prevents channel blockage because there is no direct contact between the reaction medium and the channel walls. This makes it possible to use flow chemistry in reactions involving solid formation and fouling. Moreover, since the slugs, which can be considered as batch reactors, are transported one after another through the tubular reactor, the reaction time is constant and very precise. In this way, the typical laminar flow of the reaction medium is absent in the small channels, with the middle channels flowing faster than the adjacent channels, resulting in a wider reaction time distribution of the reactants. The reaction volume in the flowing slug can be permanently mixed due to internal vortices generated by the friction between the slugs and the channel walls. Statistics can be applied to the results for each individual slug using rapid inline analytical techniques such as UV-VIS spectroscopy. This method allows for the rapid identification of relevant or irrelevant measurements. For slow analyses such as Raman spectroscopy, many segments are measured over a period of time, and then the individual measurements are accumulated.The flow chemistry in slugs is described as follows: KF. Jensen, “Flow Chemistry-Microreaction Technology Comes of Age”, AIChE, 2017 Vol.63, No.3; BJ. Reizman, KF Jensen, “Simultaneous solvent screening and reaction optimization in microliter slugs”, ChemCommun. 2015; 51(68):13290–13293; M. Movsisyana et al., “Flow Synthesis of Heterocycles”, Advances in Heterocyclic Chemistry, Volume 119, 2016, Elsevier Inc., ISSN 0065-2725, Pages 25-55; L. Shang et al., “Emerging Droplet Microfluidics”, Chem. Rev. 2017, 117, 7964-8040; Anne-Kathrin Liedtke's "Study of a new gas-liquid-solid three-phase contact mode at millimetric scale: catalytic reactors using "slurry Taylor"", Chemical and Process Engineering, Universite Claude Bernard-Lyon I, 2014; and D. Belder et al.'s "On-chip monitoring of chemical syntheses in microdroplets via SERS", Chem. Commun., 2015, 51, 8588. However, to date, the research and development of automation for flow chemistry processes in slugs has been impossible and still requires manual work.

[0005] Problems to be solved

[0006] Therefore, it is desirable to provide methods and apparatus for addressing the aforementioned technical challenges. Specifically, apparatus and methods should be provided for determining a target set of parameters for the flow chemistry setup in a slug, which allows for automation and optimization, thereby reducing the complexity, robustness, and improving control of processes involving flow chemistry in the slug. Summary of the Invention

[0007] This problem is solved by a computer-implemented method for controlling and / or monitoring a production plant, a computer program, and a control system having the features of the independent claims. Advantageous embodiments that may be implemented in isolation or in any combination are listed in the dependent claims.

[0008] In a first aspect of the invention, a computer-implemented method is provided for determining at least one set of target parameters for a flow chemistry setup for flow chemistry in a slug.

[0009] As used herein, the term "computer-implemented" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, a process wholly or partially implemented using data processing means (such as a data processing means including at least one processor). Therefore, the term "computer" generally refers to an apparatus or combination or network of apparatuses having at least one data processing means (such as at least one processor). Additionally, a computer may include one or more further components, such as at least one of a data storage device, an electronic interface, or a human-machine interface.

[0010] As used herein, the term "slug" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, segmented liquid-liquid flow patterns. A slug can be formed by simultaneously introducing at least two liquids into a tubular reactor or by introducing a dispersed phase into a continuous phase flowing within a tubular reactor, also known as the reaction phase and the carrier liquid. For introducing two liquids, a Y-type mixer and / or a T-type mixer can be used. The formation of slugs is generally known to those skilled in the art. A slug can be viewed as a series of small-batch reactors fed through a tubular reactor. Slugs can exhibit well-defined interphase mass transfer zones and flow patterns. Two basic mass transfer mechanisms may emerge: convection within a single liquid slug and diffusion between adjacent slugs, see J. Jovanovic et al., “Liquid–liquid slug flow: Hydrodynamics and pressure drop”, Chemical Engineering Science 66(1):42-54, January 2011, DOI:10.1016 / j.ces.2010.09.040. The term “flow chemistry in slugs” as used herein is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, segmented, and particularly discontinuous flow chemistry, where liquid-liquid flow patterns involve multiple slugs.

[0011] The method of the present invention proposes the use of slug flow as a means of providing true slug or piston flow behavior, which has the advantages of: narrow residence time distribution, a large number of independent experiments in a short time, the potential for statistical analysis of multiple individual slugs, the ability to handle solid particles to a certain extent without clogging the flow channels, and mild mixing of internal volumes.

[0012] As used herein, the term "flow chemistry setup" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, a system configured to perform at least one flow chemistry process (particularly for at least one flow chemistry process involving a slug). A flow chemistry setup may include multiple components. For example, a flow chemistry setup may include one or more of at least one reactor, at least one pump, at least one mixer (such as a Y-mixer or a T-mixer), at least one valve, at least one heating device, at least one pressure regulator, and at least one analytical unit.

[0013] As used herein, the term "parameter set" for flow chemistry setups is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation any specific or custom meaning. The term may specifically refer to, but is not limited to, settable and / or configurable and / or adjustable characteristics and / or properties of a flow chemistry setup. A parameter set may include multiple parameters. A parameter set may include parameters related to the reaction formulation and / or process parameters, particularly control parameters. The parameters of the parameter set of a flow chemistry setup can define the characteristics and / or properties of the components of the flow chemistry setup. The parameter set of a flow chemistry setup can affect one or more of the reaction time, reaction rate, slug formation, and final reaction products.

[0014] As used herein, the term "target parameter set" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, an optimized set of parameters for a flow chemistry setup. The target parameter set may include at least one parameter selected from the group consisting of: the flow rate of at least one pump, e.g., the flow rate of each pump in the flow chemistry setup or the total flow rate; temperature; reaction time; at least one parameter from online analysis of the precipitate, e.g., pH; and the amount of seed particles, e.g., for controlling precipitation during the nucleation process. The reaction time can be adjusted by changing the total flow rate. The target parameter set may include at least one parameter related to the reactor size. However, the reactor size is preferably kept constant and / or unchanged. The target parameter set may include parameters related to the size of the slugs. The size of the slugs can be adjusted by changing the ratio between the reaction phase and the carrier liquid. However, it is preferable to keep the slug size constant such that each slug has the same conditions.

[0015] This method is a self-learning method. As used herein, the term "self-learning method" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or tailored meaning. The term can specifically refer to, but is not limited to, the ability of a method to learn through repetition, and particularly to the ability to improve over time in the sense of providing the most appropriate or suitable set of target parameters for the optimization objective. The method can be configured to learn on each repetition. However, embodiments are possible where the method learns after a predefined number of repetitions. For example, multiple experiments can be performed, where the method is trained after each experiment run. The method may include the use of at least one artificial intelligence (AI) system. The method may include the use of at least one machine learning tool, particularly a deep learning architecture. The method can be executed fully automatically. Full automation of the method allows the AI ​​system to find the optimal parameters on its own. Specifically, the method can achieve self-optimization by iteratively setting parameters, thereby achieving a predefined final objective without human interaction. For this purpose, a machine learning model is used. Based on observation, the machine learning model facilitates the configuration of the parameters.

[0016] As outlined above, the flow chemistry in slugs is generally known and described in references such as KF. Jensen, “Flow Chemistry - Microreaction Technology Comes of Age”, AIChE, 2017, Vol. 63, No. 3; BJ. Reizman and KF. Jensen, “Simultaneous solvent screening and reaction optimization in microliter slugs”, Chem Commun. 2015; 51(68): 13290–13293; M. Movsisyana et al., “Flow Synthesis of Heterocycles”, Advances in Heterocyclic Chemistry, Volume 119, 2016, Elsevier Inc., ISSN 0065-2725, Pages 25-55; L. Shang et al., “Emerging Droplet”. "Microfluidics", Chem. Rev. 2017, 117, 7964-8040; Anne-Kathrin Liedtke, "Study of a new gas-liquid-solid three-phase contact mode at millimetric scale: catalytic reactors using "slurry Taylor"", Chemical and Process Engineering. Universite Claude Bernard-Lyon I, 2014; D. Belder et al., "On-chip monitoring of chemical syntheses in microdroplets via SERS", Chem. Commun. 2015, 51, 8588. Known setups are typically automated, allowing pumps and other components to be controlled by a computer. However, these known setups are not configured to learn from data and cannot perform new experiments.

[0017] The method includes the following method steps, which can be performed in a given order. However, a different order is also possible. Furthermore, two or more method steps can be performed simultaneously, either wholly or partially. Additionally, one or more, or even all, method steps can be performed once or repeatedly, such as once or multiple times. Moreover, the method may include other method steps not listed.

[0018] The method includes the following steps:

[0019] a) Determine at least one process variable by using at least one sensor in a flow chemistry setup;

[0020] b) Train at least one machine learning model based on process variables;

[0021] c) Determine the target parameter set by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model;

[0022] d) Provide a defined set of target parameters and / or consider a defined set of target parameters in order to evaluate the flow chemistry setup and / or evaluate at least one flow chemistry product.

[0023] As used herein, the term "process variable" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, specifying at least one quantity of a final reaction product. As used herein, the term "final reaction product," also referred to as producing fluids and flowing chemical products, may refer to the result or output of a flow chemical process, particularly the result or output of a tubular reactor. A tubular reactor may include at least one outlet. The process variable may be determined at the outlet of the tubular reactor. The process variable can be determined by measuring the amount flowing through one or more slugs in the tubular reactor. The process variable may include at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information; and at least one color information.

[0024] As used herein, the term "determine at least one process variable" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, at least one process that generates at least one measurement (particularly at least one or more representative results indicating a process variable). Step a) may include a single measurement of the process variable or multiple consecutive measurements of one or more quantities. The flow chemistry setup includes at least one sensor, specifically multiple sensors. As used herein, the term "sensor" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, a process suitable for performing the above determination of a process variable and / or any element suitable for use in the above determination of a process variable. Thus, the sensor may be specifically suitable for determining a process variable.

[0025] Process variables can be determined using “inline,” “online,” or “on-line” analysis. For inline analysis, the process variable can be measured within the reactor. Advantages of inline analysis include no sample preparation required, no distortion due to sample extraction, the ability to maintain pressure and / or temperature, real-time measurement, and spatially resolved measurements. For continuous online analysis, a bypass can be used to determine the process variable measurement. Advantages of continuous online analysis include no sample preparation required, no distortion due to sample extraction, the ability to maintain pressure and / or temperature, and real-time measurement. For discontinuous online analysis, the process variable can be measured after automated sample extraction. Advantages of discontinuous online analysis include the ability to measure very close to the process and with only a short time delay. For on-line analysis (also known as offline analysis), the process variable can be measured after manual sample extraction. This may require sample preparation but may allow for shorter waiting periods.

[0026] The determination of process variables may include one or more of the following: ultraviolet and visible spectroscopy (UV-VIS), fluorescence spectroscopy, Raman spectroscopy, infrared (IR) spectroscopy, attenuated total reflectance (ATR) infrared (IR) spectroscopy, nuclear magnetic resonance (NMR) spectroscopy, optical detection, fluorescence spectroscopy, mass spectrometry (MS), high-performance liquid chromatography (HPLC), gas chromatography (GC); conductivity and pH measurements, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), and automated titration. For example, a process variable may be determined using UV-VIS spectroscopy, and the process variable may be one or more of intensity, wavelength, peak area, half-width, and half-maximum. For example, a process variable may be determined using fluorescence spectroscopy, and the process variable may be intensity or wavelength. For example, a process variable may be determined using Raman spectroscopy, and the process variable may be intensity. For example, a process variable may be determined using infrared spectroscopy, and the process variable may be intensity. For example, a process variable may be determined using light scattering, and the process variable may be intensity. For example, a process variable may be determined using optical detection, and the process variable may be particles in the stream. For example, a process variable may be determined using conductivity and pH measurements. For example, process variables can be determined using calorimetry, and the process variable can be heat flux. For example, process variables can be determined by viscosity determination, and the process variable can be pressure difference. For example, process variables can be determined using HPLC, and the process variable can be peak area. For example, process variables can be determined using GC, and the process variable can be peak area.

[0027] The sensor may include one or more of at least one spectrometer, at least one light barrier, at least one chromatograph, at least one viscometer, at least one titration device, and at least one calorimeter. For example, the sensor may be or may include at least one light barrier. The light barrier may be configured to determine how much light passes through the final reaction product at the outlet of the tubular reactor. Specifically, the sensor may be configured to determine the intensity or change in intensity of at least one beam of light passing through the final reaction product.

[0028] Choosing an appropriate separation medium can take into account the specific circumstances of the selected analytical method. For example, in the case of using UV-Vis and scattered light, air, gas, or fluorinated oil can be used. The UV-Vis signals of the reaction and separation media may differ at specific wavelengths or wavelength ranges, making differentiation possible. Scattering at the interface can be used as a trigger for slug detection. For example, in the case of using NMR fluorocarbon oil, a specific NMR tracer may be used. NMR fluorocarbon oil can be immiscible with both hydrophilic and lipophilic components. Specifically, F substituents do not show activity in proton NMR. For example, HCF has a fairly typical chemical shift in the 4–4.5 range. For example, in the case of using IR spectroscopy, air or N2 gas can be used. These materials do not have significant signals at wavenumbers in the typical C=O vibrational mode range, and a signal drop to almost zero can be used as a trigger. For example, in the case of using ATR, IR fluorocarbon oil immiscible with both hydrophilic and lipophilic components can be used, and the reaction medium should wet the ATR crystals while the separation medium should not. For example, in the case of using Raman spectroscopy, fluorocarbon oil, air, gas, or other liquids can be used. The Raman signal of the separation medium should not overshadow the Raman signal of the reactor medium being analyzed. For example, in the case of using HPLC with an automated sampler (online HPLC), air or gas can be used. Air or gas is separated in the HPLC vial, and measurements can be performed as usual. For example, in the case of HPLC using a gas / liquid or liquid / liquid separator, air, gas, or fluorinated oil can be used. The separation medium is separated in flow, and only the contents of the slug are injected into the HPLC. For this purpose, a small amount of the slug contents are measured together. A trigger, as described with regard to UV-VIS, can detect the slug, and if the HPLC is ready to measure the next sample, an automated valve can feed the slug into the HPLC sample loop and stop the flow when the loop is full. Therefore, the size of the slug should be much larger than the volume of the sample loop. In the case of using mass spectrometry, air can be used. This prevents the signal from convolving with the fragment of interest, and the signal reduced to almost zero can be used as a trigger.

[0029] The separation medium between slugs can be selected to ensure compatibility with the analytical setup. For example, scattering and interfaces can interfere with UV-Vis measurements. This method may include algorithmic removal of scattering signals. For example, oil can damage the column. This method may include using bubbles for separation and / or possibly using control valves to direct the correct (partial) slugs into the autosampler sampling loop. For example, interference with desired mass spectrometry and NMR signals can occur. This method may include using fluorinated solvents or bubbles for separation. For example, for ATR-IR, appropriate wettability may need to be considered. When a gaseous medium is used for separation, the reaction medium comes into contact with the channel / tube wall, which can lead to deposition and broadening of the residence time distribution.

[0030] The separation medium can be selected to be compatible with the wet materials in the system (or vice versa) to ensure stable separation and avoid merging.

[0031] The separation medium can be selected such that it generates a detectable signal in the chosen analytical method. In this way, the method can include using an algorithm capable of automatically distinguishing signals of interest from the sluice gate where the reaction occurs from those from the separation medium or interface. Refining the data using an appropriate algorithm can include two steps:

[0032] - Use the signal from the separated medium in the analysis detection to cut out the data of interest from a series of time-resolved measurements and make them available for further analysis;

[0033] - Based on this data, perform machine learning to, for example, determine the next set of parameters for screening and achieve self-optimizing screening.

[0034] It was discovered that it is essential to distinguish between very rapid methods like UV-VIS and slow methods like HPLC. With slow methods, it is possible to accumulate many slugs and wait for the HPLC to be ready to measure the next sample. With rapid methods, each slug can be a separation experiment.

[0035] As used herein, the term "reactor" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, any apparatus in which a chemical reaction takes place. The term "tubular reactor" as used herein is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, the geometry, particularly the shape, of a reactor having a tubular form or composed of tubes. The tube may include a cylindrical surface. The tube may have a length h and a diameter d.

[0036] The sensor can be configured to generate at least one sensor signal. As used herein, the term "sensor signal" (also called a measurement signal) is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation any specific or customary meaning. The term specifically refers to, but is not limited to, a signal generated by the sensor in response to a detected event or a change in its environment (particularly in response to illumination). Specifically, the sensor signal can be or may include at least one electrical signal, such as at least one analog electrical signal and / or at least one digital electrical signal. More specifically, the sensor signal can be or may include at least one voltage signal and / or at least one current signal. Furthermore, a raw sensor signal can be used, or the sensor can be adapted to process or preprocess the sensor signal, such as by filtering, to generate a secondary sensor signal, which can also be used as the sensor signal. For example, preprocessing may include consideration of statistical information about multiple slugs.

[0037] Flow chemistry setups can be used to assess whether they produce valid data. If the data is invalid, the system will automatically repeat the measurement, particularly experiments. For example, the method may include at least one validation step. The validation step may include validating at least one measurement of the identified process variable. As used herein, the term “validation” is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, the process of determining the suitability of a measurement of a process variable, particularly considering accuracy and reliability. Validation may include comparing the measurement to at least one predefined standard. The predefined standard may be an accuracy standard, such as a tolerable measurement error. If the measurement of the identified process variable is not validated, step a) may be repeated. If the measurement is validated, the measurement can be considered a valid data point. If the measurement is validated, the method may proceed to step b).

[0038] Alternatively or concurrently, the method may include at least one anomaly detection step. The term "anomaly" as used herein is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation a specific or customary meaning. The term may specifically refer to, but is not limited to, deviations from the expected sensor signal, particularly outside at least one tolerance interval. At least one algorithm can monitor the sensor's measurement signal. The algorithm may be configured to identify at least one anomaly. The algorithm may be based on artificial intelligence. The algorithm may include at least one machine learning algorithm. The machine learning algorithm can be trained using historical sensor signals, which may include sensor signals without anomalies and sensor signals with anomalies. If an anomaly is detected, step a) may be repeated. If no anomaly is detected, the measurement is considered a valid data point. If no anomaly is detected, the method may proceed to step b).

[0039] The verification and / or anomaly detection steps can be configured to extract clean and high-quality analytical data. Specifically, the method according to the invention proposes the intelligent use of an internal trigger detected by the analytical tools employed in the screening process to extract clean and high-quality analytical data. This allows a specialized algorithm to identify the analytical data of interest from the slug and eliminate signals from the separation medium. To achieve this, the algorithm can analyze the data sequence along the time axis and cut out neat signals from the slug. The advantage of using an internal trigger is that it eliminates the need for additional analytical instruments (the analytical method itself determines when the slug and piston start and end, respectively) and avoids complex synchronization between different measuring devices (one for detection, the other for analysis). This is particularly useful if the flow rate is varied to screen for the effects of residence time. Detection and separation of the slug / piston can prevent interference with the screening by generating noise, reducing the integrated signal strength, and causing damage to the column.

[0040] As used herein, the term "machine learning" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation any specific or customary meaning. The term specifically refers to, but is not limited to, methods of artificial intelligence (AI) using automated model building for machine learning models (particularly predictive models). Training can be performed using at least one machine learning system. As used herein, the term "machine learning system" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation any specific or customary meaning. The term specifically refers to, but is not limited to, systems or units comprising at least one processing unit, such as processors, microprocessors, or computer systems configured for machine learning (particularly for executing logic in a given algorithm). The machine learning system can be configured to implement and / or execute at least one machine learning algorithm, wherein the machine learning algorithm is configured to build at least one machine learning model. The machine learning model can include at least one machine learning architecture and model parameters. The machine learning model can be a Bayesian machine learning model and / or based on neural networks, such as reinforcement neural networks. The machine learning model can include at least one designed experimental method. The machine learning model can be configured to take into account noise of the determined process variables. The noise can be based on different distributions (e.g., Gaussian) and types (e.g., additive and / or multiplicative). This can be handled accordingly by a machine learning model. The machine learning model can be configured to consider constraints on the target parameter set. The parameters of the target parameter set can have constraints, such as upper and / or lower limits or constants for the pump flow rate that the machine learning model might consider.

[0041] As used herein, the term "training" (also known as "learning") is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, the process of building a machine learning model (particularly determining and / or updating the parameters of the machine learning model). The machine learning model may be at least partially data-driven. The machine learning model can learn from valid data points. Training can be performed on sensor data, such as considering determined process variables. Training can be performed on process parameters determined in historical production runs (particularly historical production runs with a known set of parameters for flow chemistry settings). As used herein, the term "at least partially data-driven model" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, the fact that a machine learning model includes both a data-driven model portion and other model portions based on physicochemical laws.

[0042] The determination of the target parameter set in step c) may include at least one optimization step. As used herein, the term "optimization" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation a specific or customized meaning. Specifically, the term may refer to, but is not limited to, the process of selecting the optimal set of parameters from a parameter space of possible parameters with respect to the optimization objective. As used herein, the term "optimization objective" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation a specific or customized meaning. Specifically, the term may refer to, but is not limited to, at least one criterion for performing optimization. The optimization objective may include at least one optimization objective and accuracy and / or precision. The optimization objective may be pre-specified by at least one user of the flow chemistry setup. The optimization objective may be a specification of at least one user. The user can select the optimization objective and the desired accuracy and / or precision. For example, the optimization objective may include at least one value from sensor data. The corresponding concentration of the produced fluid can be determined from the sensor data. The optimization objective may be a measurement determined by a sensor, such as an intensity value with the desired accuracy and / or precision determined by a light barrier.

[0043] Optimization may include the application of machine learning models. Based on the current state of the machine learning model and / or on the determined process variables, the optimization algorithm can determine how optimally to set the parameters of the flow chemistry setup. As used herein, the term "optimization algorithm" may refer to at least one algorithm used to solve at least one optimization problem. Optimization may include solving at least one optimization problem, such as at least one maximization problem or at least one minimization problem. The optimization may include computational steps, such as computing the solution to the optimization problem. The optimization algorithm may be Bayesian optimization, for example, using a Gaussian process as a surrogate model. Other optimization algorithms are also possible. For example, the optimization algorithm may be a reinforcement learning network. The optimization algorithm may be configured to consider noise of the determined process variables. The noise may be based on different distributions (e.g., Gaussian) and types (e.g., additive and / or multiplicative). This may be addressed accordingly in steps b) and / or c). The optimization algorithm may be configured to consider constraints on the objective parameter set. The parameters of the objective parameter set may have constraints, such as upper and / or lower limits or constants for pump flow rates that the optimization algorithm may consider.

[0044] Optimization steps may depend on the machine learning model. However, this may not mean that every decision in every step must depend on the machine learning model. For example, multiple experiments may be performed, with the method trained after each experiment. For example, an average window derived from previous parameter values ​​may also be used. Furthermore, optimization steps may include trade-offs between developing and exploring the underlying space.

[0045] In step c), the optimization algorithm may determine one or more target parameter sets, such as Pareto-front or subsets of Pareto-front. In the case of multiple target parameter sets, step d), in particular repeating one or more method steps a) to d), may be performed for all configurations of the target parameter sets.

[0046] Research projects typically begin with a screening task. In this phase (stadium), the research problem is clearly described, meaning the optimization objective and a list of influencing parameters are defined. The parameter space is usually large, necessitating a large number of experiments. The method according to the invention can automate this screening task with minimal raw material consumption and manual labor. Specifically, as will be outlined in detail below, the instructions for executing the method according to the invention can be implemented as a computer program (particularly software), such that when the program is executed by a computer or computer network, the instructions cause the computer or computer network to execute the method according to the invention. Flow chemistry settings can be fully automated, allowing all relevant parameters (particularly formulation and process parameters) to be controlled within a given range using the method according to the invention. Machine learning models can learn from effective experimental data points, and optimization steps can provide the parameters required to achieve a given optimization objective for new experimental planning.

[0047] The method proposed in this invention combines process control settings with machine learning and / or artificial intelligence to achieve self-optimization screening, which has the following advantages: high-throughput execution of experiments, automatic recording and analysis of data to avoid human error and reduce repetitive manual work, and rapid screening of those domains in the parameter space containing valuable information by dynamically setting parameters according to a customized machine learning algorithm.

[0048] Step d) includes providing the determined set of target parameters and / or considering the determined set of target parameters to evaluate the flow chemistry setup and / or evaluate at least one flow chemistry product.

[0049] As used herein, the term "provide target parameter set" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation its specific or customized meaning. The term may specifically refer to, but is not limited to, presenting and / or displaying and / or communicating the target parameter set to a user. The provision of the determined target parameter set can be performed using at least one output device. As used herein, the term "output device" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation its specific or customized meaning. The term may specifically refer to, but is not limited to, at least one interface configured to, for example, provide the target parameter set to at least one user. The output device may include at least one display device.

[0050] As used herein, the term "considering a set of target parameters for evaluating a flow chemistry setup" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation its specific or custom-defined meaning. The term may specifically refer to, but is not limited to, setting parameters of a flow chemistry setup according to a defined set of target parameters. Similarly, as used herein, the term "considering a defined set of target parameters for evaluating at least one flow chemistry product" is a broad term and should be given a general and conventional meaning by those skilled in the art, without limitation its specific or custom-defined meaning. The term may specifically refer to, but is not limited to, adapting and / or adjusting the formulation of a product produced using a flow chemistry setup, particularly the raw material and / or raw material concentrations.

[0051] As outlined above, one or more, or even all, of the method steps can be repeated, such as once or multiple times. This allows for self-learning and / or self-optimization. Specifically, the determined set of target parameters can be used as the starting point for the next optimization. As used herein, the term "next optimization" may refer to repeating method steps a) through d), where in step a) the process variables of the flow chemistry setting are determined, having parameters set as target parameters determined in the previous round of the method. Steps a) through d) can be repeated until the process variables measured by the sensors meet the previously defined target values ​​within a predefined accuracy.

[0052] This invention allows for accelerated research processes, particularly for screening purposes. Rapid development of new materials is possible. Users can focus on other tasks involving manual operations. Flow chemistry within the slug allows for resource-efficient research. Only small amounts of raw materials are required for each experiment. Using a slug allows for simplified optimization, especially considering the reduced noise in the input data for the algorithms used in steps b) and c). The slug can provide explicit constraints on measurements. The absence of residue within the slug may result in enhanced input data for the algorithms in steps b) and c). Therefore, enhanced results can be achieved in continuous flow chemistry comparisons.

[0053] Slugs are generally not used in known methods and apparatus because their use increases workload. Slug interference analysis includes scattered light signals from UV-VIS. Suitable materials need to be selected to prevent detector contamination. Algorithms for slug detection are required, and extracting clean, high-quality data is essential. Furthermore, statistical analysis of multiple single measurements of the slug is necessary. This invention describes a method capable of generating high-quality data by combining the following aspects:

[0054] - Using slug flow as a means of providing real slug or piston flow behavior has the advantages of: narrow residence time distribution, a large number of independent experiments in a short time, the potential for statistical analysis of multiple individual slugs, the ability to handle solid particles to some extent without clogging the flow channels, and mild mixing of internal volumes.

[0055] - Combining process control settings with machine learning and / or artificial intelligence to achieve self-optimizing screening has the advantages of: performing experiments with high throughput, automatically recording and analyzing data to avoid human error and reduce repetitive manual work, and quickly screening those parameter space domains containing valuable information by dynamically setting parameters according to customized machine learning algorithms.

[0056] - The sensitive use of the internal trigger, detected by the analytical tools employed in the screening process, extracts clean and high-quality analytical data. This allows a dedicated algorithm to identify the analytical data of interest from the slug and eliminate those signals originating from the separation medium. To achieve this, the algorithm can analyze the data sequence along the time axis and cut out clean signals from the slug. The advantage of using the internal trigger is that it eliminates the need for additional analytical instruments (the analytical method itself determines when the slug and piston start and end, respectively) and the need for complex synchronization between different measuring devices (one for detection, the other for analysis). This is particularly useful when varying the flow rate to screen for the effects of residence time. Detection and separation of the slug / piston can prevent interference from the slug / piston on the screening process by generating noise, reducing the integrated signal strength, and damaging the column.

[0057] The selection of an appropriate separation medium should take into account the specific circumstances of the analytical method.

[0058] In a further aspect of the invention, a computer program is provided for determining at least one set of target parameters for a flow chemistry setup for flow chemistry in a slug. The computer program includes instructions that, when executed by a computer or computer network, cause the computer or computer network to perform, wholly or partially, the method according to the invention in one or more embodiments disclosed herein. The computer program is configured to perform at least steps a) to d) of the method according to the invention. For possible definitions of most terms as used herein, reference may be made to the description of the computer-implemented method further described above or below.

[0059] Specifically, computer programs can be stored on computer-readable data carriers and / or computer-readable storage media. As used herein, the terms "computer-readable data carrier" and "computer-readable storage media" specifically refer to non-transitory data storage components, such as hardware storage media having computer-executable instructions stored thereon. Computer-readable data carriers or storage media can specifically be or include storage media such as random access memory (RAM) and / or read-only memory (ROM). For example, computer programs and / or machine learning models and / or training data can be stored using at least one database (such as a server or cloud server). For example, computer programs and / or machine learning models and / or training data can be stored in a laboratory information management system.

[0060] Computer program products having program code components are further disclosed and proposed herein so that, when the program is executed on a computer or computer network, it performs the methods of the present invention according to one or more embodiments disclosed herein. Specifically, the program code components may be stored on a computer-readable data carrier and / or a computer-readable storage medium.

[0061] This document further discloses and proposes a data carrier having a data structure stored thereon, which, after being loaded into a computer or computer network, such as into the working memory or main memory of the computer or computer network, can perform the methods of the invention according to one or more embodiments disclosed herein.

[0062] This document further discloses and proposes a computer program product having program code components stored on a machine-readable medium, so that when the program is executed on a computer or computer network, it performs the method of the invention according to one or more embodiments disclosed herein. As used herein, a computer program product refers to a program that is a tradable product. The product can generally be in any format (such as paper format) or on a computer-readable data medium. Specifically, the computer program product can be distributed via a data network.

[0063] In a further aspect of the invention, an automatic control system for flow chemistry settings in a slug is provided. The automatic control system includes:

[0064] - At least one communication interface configured to receive at least one process variable determined by at least one sensor of at least one flow chemistry setup;

[0065] - At least one machine learning model, which is configured for training based on process variables;

[0066] - At least one processing unit configured to determine at least one set of target parameters by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model;

[0067] - At least one output device configured to provide the determined target parameter set.

[0068] An automatic control system may be configured to automatically control a flow chemistry setup. The automatic control system may be configured to perform the method according to the invention. For possible definitions of most terms used herein, reference may be made to the description of the computer-implemented method, which is further described in detail above or below. The control system may be part of the flow chemistry setup or may be embodied separately from the flow chemistry setup.

[0069] As used herein, the term "communication interface" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or customary meaning. The term may specifically refer to, but is not limited to, items or elements forming the boundaries configured for transmitting information. In particular, a communication interface may be configured to transmit information from a computing device (e.g., a computer), such as sending or outputting information (e.g., to another device). Alternatively or additionally, a communication interface may be configured to transmit information to a computing device, such as a computer, for receiving information. A communication interface may specifically provide components for transmitting or exchanging information. In particular, a communication interface may provide data transmission connectivity, such as Bluetooth, NFC, inductive coupling, etc. As an example, a communication interface may be or may include at least one port, including one or more network or internet ports, USB ports, and disk drives. A communication interface may be at least one web interface.

[0070] As used herein, the term "processing unit" is a broad term and should be given its general and conventional meaning by those skilled in the art, without limitation its specific or custom meaning. The term may specifically refer to, but is not limited to, any means suitable for performing optimization, preferably by using at least one data processing means, and more preferably by using at least one processor and / or at least one application-specific integrated circuit (ASIC). Thus, by way of example, a processing unit may include one or more programmable means, such as one or more computers, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other means configured to perform optimization. Thus, by way of example, at least one processing unit may include at least one data processing means having software code, comprising a plurality of computer commands, stored thereon. A processing unit may provide one or more hardware elements for performing one or more specified operations, and / or may provide one or more processors having software running thereon for performing one or more specified operations.

[0071] Flow chemistry setup and control systems can be modular, such as kits. The system can have replaceable and interchangeable hardware and software components, such as those employing different types of pumps and / or different optimization algorithms.

[0072] As used herein, the terms “have,” “contain,” or “include,” or any of their arbitrary grammatical variations, are used in a non-exclusive manner. Thus, these terms can refer either to a situation where no further features exist in the entity described in the context, other than those introduced by these terms, or to a situation where one or more further features exist. For example, the statements “A has B,” “A contains B,” and “A includes B” can refer either to a situation where no other elements exist in A besides B (i.e., A consists solely and entirely of B), or to a situation where entity A contains one or more further elements besides B, such as element C, element D, or even further elements.

[0073] Furthermore, it should be noted that expressions such as "at least one," "one or more," or similar statements indicating that a feature or element may appear once or more are generally used only once when the corresponding feature or element is introduced. In most cases, the expressions "at least one" or "one or more" will not be repeated when referring to the corresponding feature or element, even though there may be a fact that the corresponding feature or element may appear once or more.

[0074] Furthermore, as used herein, the terms “preferredly,” “more preferably,” “particularly,” “more particularly,” “specifically,” “more specifically,” or similar terms are used in combination with optional features without limiting the possibility of substitution. Therefore, features introduced by these terms are optional features and are not intended to limit the scope of the claims in any way. As those skilled in the art will recognize, the invention can be carried out by using alternative features. Similarly, features introduced by phrases such as “in embodiments of the invention” are intended to be optional features, with no limitation on alternative embodiments of the invention, no limitation on the scope of the invention, and no limitation on the possibility of combining features introduced in this way with other optional or non-optional features of the invention.

[0075] The overview does not exclude other possible embodiments, and the following embodiments are conceivable:

[0076] Example 1: A computer-implemented method for determining at least one set of target parameters for a flow chemistry setup in a slug, wherein the method is a self-learning method, and the method includes the following steps:

[0077] a) Determine at least one process variable by using at least one sensor in a flow chemistry setup;

[0078] b) Train at least one machine learning model based on process variables;

[0079] c) Determine the target parameter set by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model;

[0080] d) Provide a defined set of target parameters and / or consider a defined set of target parameters for evaluating flow chemistry setups and / or evaluating at least one flow chemistry product.

[0081] Example 2: According to the method described in the foregoing examples, the determined target parameter set is used as the starting point for the next optimization.

[0082] Example 3: The method according to any one of the foregoing examples, wherein steps a) to d) are repeated until the process variable measured by the sensor meets the previously defined target value within a predefined accuracy.

[0083] Example 4: The method according to any one of the foregoing examples, wherein the optimization algorithm is Bayesian optimization and / or at least one reinforcement learning network.

[0084] Example 5: The method according to any one of the foregoing examples, wherein the machine learning model is configured to take into account noise of the determined process variables.

[0085] Example 6: The method according to any one of the foregoing examples, wherein the machine learning model is configured to take into account the constraints of the target parameter set.

[0086] Example 7: The method according to any one of the preceding embodiments, wherein the target parameter set includes at least one parameter selected from the group consisting of: the flow rate of at least one pump; temperature; reaction time; at least one parameter from online analysis of the precipitate; and the amount of seed particles.

[0087] Example 8: The method according to any one of the foregoing examples, wherein the process variables are determined by measuring one or more quantities flowing through a slug in at least one tubular reactor.

[0088] Example 9: The method according to any one of the foregoing examples, wherein the process variables include at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information; and at least one color information.

[0089] Example 10: The method according to any one of the foregoing examples, wherein the determination of process variables includes one or more of the following: ultraviolet-visible spectroscopy (UV-VIS), Raman spectroscopy, infrared (IR) spectroscopy, nuclear magnetic resonance (NMR) spectroscopy, optical detection, fluorescence spectroscopy, mass spectrometry (MS), high performance liquid chromatography (HPLC), gas chromatography (GC), conductivity and pH determination, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), and automated titration.

[0090] Example 11: The method according to any one of the foregoing embodiments, wherein the sensor includes one or more of the following: at least one spectrometer, at least one light barrier, at least one chromatograph, at least one viscometer, at least one titration device, and at least one calorimeter.

[0091] Example 12: The method according to any one of the foregoing embodiments, wherein the method includes at least one verification step, wherein at least one measurement of the determined process variable is verified, wherein the verification includes comparing the measurement with at least one predefined standard, wherein step a) is repeated if the measurement of the determined process variable is not verified.

[0092] Example 13: The method according to any one of the foregoing embodiments, wherein the method includes at least one anomaly detection step, wherein at least one algorithm monitors at least one measurement signal of a sensor, wherein the algorithm is configured to determine at least one anomaly, wherein step a) is repeated if an anomaly is detected.

[0093] Example 14: The method according to any one of the foregoing examples, wherein the optimization objective is the specification of at least one user, and wherein the optimization objective is the concentration of at least one production fluid.

[0094] Example 15: A computer program for determining at least one set of target parameters for a flow chemistry setup in a slug, configured to, when executed on a computer or computer network, cause the computer or computer network to perform, fully or partially, the method according to any one of the foregoing embodiments, wherein the computer program is configured to perform at least steps a) to d) of the method according to any one of the foregoing embodiments.

[0095] Example 16: A computer-readable storage medium including instructions that, when executed by a computer or computer network, cause the execution of at least steps a) to d) of the method according to any of the foregoing embodiments relating to the method.

[0096] Example 17: An automatic control system for flow chemistry settings in a slug, comprising:

[0097] - At least one communication interface configured to receive at least one process variable determined by at least one sensor of at least one flow chemistry setup;

[0098] - At least one machine learning model, which is configured to be trained based on process variables;

[0099] - At least one processing unit configured to determine at least one set of target parameters by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model;

[0100] - At least one output device configured to provide a defined target parameter set.

[0101] Example 18: A system according to the foregoing embodiments, wherein the system is configured to perform a method according to any one of the foregoing embodiments involving methods. Attached Figure Description

[0102] Further optional features and embodiments, preferably in conjunction with the dependent claims, will be disclosed in more detail in the subsequent description of the embodiments. As those skilled in the art will understand, the corresponding optional features may be implemented in isolation or in any feasible combination. The scope of the invention is not limited to the preferred embodiments. Embodiments are schematically illustrated in the figures. In these figures, the same reference numerals refer to the same or functionally comparable elements.

[0103] In the diagram:

[0104] Figure 1 An embodiment of the method according to the invention is shown; and

[0105] Figure 2 An embodiment of the flow chemistry setup and automatic control system according to the present invention is shown. Detailed Implementation

[0106] Figure 1 An embodiment of a computer-implemented method for determining at least one set of target parameters for a flow chemistry setup 110 in a slug according to the invention is shown.

[0107] An embodiment of the flow chemistry setup 110 is in Figure 2As shown in the figure. A slug can be formed by simultaneously introducing at least two liquids 112, 114 into a tubular reactor 116, or by introducing a dispersed phase into a continuous phase (also referred to as the reaction phase and carrier liquid) flowing within the tubular reactor. A T-mixer 118 can be used to introduce the two liquids 112, 114. Slug formation is generally known to those skilled in the art. A slug can be viewed as a series of small-batch reactors fed through the tubular reactor. Slugs can exhibit well-defined interphase mass transfer zones and flow patterns. Two basic mass transfer mechanisms may occur: convection within a single liquid slug and diffusion between adjacent slugs, see J. Jovanovic et al., “Liquid–liquid slug flow: Hydrodynamics and pressure drop”, Chemical Engineering Science 66(1):42-54, January 2011, DOI:10.1016 / j.ces.2010.09.040. The flow chemistry in a slug can include segmented, and particularly discontinuous, flow chemistry, where the liquid-liquid flow pattern involves multiple slugs.

[0108] The flow chemistry setup 110 may include multiple components. For example, the flow chemistry setup 110 may include one or more of the following: at least one reactor 116, at least one mixer (such as a T-mixer 118), and Figure 2 Other components not shown include at least one pump, at least one valve, at least one heating device, at least one pressure regulator, and at least one analysis unit.

[0109] Back Figure 1 The method includes the following method steps, which may be performed in a given order. However, a different order is also possible. Two or more steps of the method may be performed simultaneously, either entirely or partially. Furthermore, one or more, or even all, of the method steps may be performed once, or may be performed repeatedly, such as once or multiple times. Additionally, the method may include other method steps not listed.

[0110] The method includes the following steps:

[0111] a) (indicated by reference numeral 120) at least one process variable is determined by using at least one sensor 122 of the flow chemistry setup 110;

[0112] b) (indicated by reference numeral 124 in the figure) train at least one machine learning model 126 based on process variables;

[0113] c) (indicated by reference numeral 128) The objective parameter set is determined by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model 126;

[0114] d) (indicated by reference numeral 130) provide a defined set of target parameters and / or (indicated by reference numeral 131) consider the defined set of target parameters for evaluating flow chemistry setups and / or evaluating at least one flow chemistry product.

[0115] The parameter set for the flow chemistry setup 110 may include settable and / or configurable and / or adjustable characteristics and / or properties of the flow chemistry setup 110. The parameter set may include multiple parameters. This parameter set may include parameters related to reaction formulation and / or process parameters, particularly control parameters. The parameters of the parameter set for the flow chemistry setup 110 may define the properties and / or characteristics of the components of the flow chemistry setup 110. The parameter set for the flow chemistry setup 110 may affect one or more of the following: reaction time, reaction rate, slug formation, and final reaction products.

[0116] The target parameter set may be an optimized parameter set for the flow chemistry setup 110. The target parameter set may include at least one parameter selected from the group consisting of: the flow rate of at least one pump, e.g., the flow rate of each pump in the flow chemistry setup or the total flow rate; temperature; reaction time; at least one parameter from online analysis of the precipitate, e.g., pH; and the amount of seed particles, e.g., for controlling precipitation during the nucleation process. The reaction time can be adjusted by changing the total flow rate. The target parameter set may include at least one parameter related to the reactor size. However, the reactor size is preferably kept constant and / or unchanged. The target parameter set may include parameters related to the slug size. The slug size can be adjusted by changing the ratio between the reaction phase and the carrier liquid. However, it is preferable to keep the slug size constant so that each slug has the same conditions.

[0117] This method is a self-learning approach. It may include the use of at least one artificial intelligence (AI) system. It may include the use of at least one machine learning tool, particularly a deep learning architecture. The method can be executed fully automatically. This full automation allows the AI ​​system to find optimal parameters on its own. Specifically, the method can achieve a predefined final goal without human interaction by iteratively setting its parameters to optimize itself. For this purpose, a machine learning model is used. Based on observations, the machine learning model 126 facilitates the configuration of the parameters.

[0118] Process variables can be at least one quantity specifying the final reaction product. The final reaction product can be a product or output of a flow chemistry process, particularly a product or output of tubular reactor 116. For example... Figure 2As shown, the tubular reactor 116 may include at least one outlet 132. Process variables may be determined at the outlet 132 of the tubular reactor 116. The process variables may be determined by measuring the amount flowing through one or more slugs in the tubular reactor. The process variables may include at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information; and at least one color information.

[0119] Determining at least one process variable may include at least one process that generates at least one measurement, particularly indicating at least one or more representative results of the process variable. Step a) may include a single measurement of the process variable or multiple consecutive measurements of one or more quantities. The flow chemistry setup 110 includes at least one sensor 122, specifically multiple sensors. The determination of the process variable may include one or more of the following: ultraviolet-visible spectroscopy (UV-VIS), Raman spectroscopy, infrared (IR) spectroscopy, nuclear magnetic resonance (NMR) spectroscopy, optical detection, fluorescence spectroscopy, mass spectrometry (MS), high-performance liquid chromatography (HPLC), gas chromatography (GC), conductivity and pH determination, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), and automated titration. Sensor 122 may include one or more of at least one spectrometer, at least one light barrier, at least one chromatograph, a viscometer, at least one titration device, and at least one calorimeter. For example, sensor 122 may be or may include at least one light barrier. The light barrier may be configured to determine how much light passes through the final reaction products at the outlet 132 of the tubular reactor 116. Specifically, sensor 122 may be configured to determine the intensity or intensity change of at least one beam of light passing through the final reaction product.

[0120] Sensor 122 can be configured to generate at least one sensor signal. The sensor signal can be or may include at least one electrical signal, such as at least one analog electrical signal and / or at least one digital electrical signal. More specifically, the sensor signal can be or may include at least one voltage signal and / or at least one current signal. Furthermore, either the original sensor signal can be used, or sensor 122 can be adapted to process or preprocess the sensor signal, such as by filtering, to generate a secondary sensor signal, which can also be used as the sensor signal. For example, preprocessing may include considering statistical data from several sluice gates, indicated by reference numeral 134.

[0121] The flow chemistry setup 110 can be used to assess whether it generates valid data. If the data is invalid, the system will automatically repeat the measurement, particularly the experiment. For example, the method may include at least one validation step 136. Validation step 136 may include validating at least one measurement of the identified process variable. Validation may include determining the suitability of the process variable measurement, particularly considering accuracy and reliability. Validation may include comparing the measurement to at least one predefined standard. The predefined standard may be an accuracy standard, such as a tolerable measurement error. If the measurement of the identified process variable has not been validated, step a) may be repeated. Figure 1 In this context, "X" and the reference numeral 138 are used to indicate the measurement. If the measurement is verified, it can be considered a valid data point. Figure 1 The hook in the figure represents the measurement. If the measurement is verified, the method can proceed to step b), indicated by reference numeral 140.

[0122] Alternatively or concurrently, the method may include at least one anomaly detection step. At least one algorithm can monitor the measurement signals of sensor 122. The algorithm can be configured to determine at least one anomaly. The algorithm may be based on artificial intelligence. The algorithm may include at least one machine learning algorithm. The machine learning algorithm can be trained using historical sensor signals, which may include sensor signals without anomalies and sensor signals with anomalies. If an anomaly is detected, step a) can be repeated. If no anomaly is detected, the measurement is considered a valid data point. If no anomaly is detected, the method may continue to step b).

[0123] In step b) 124, a machine learning model 126 is trained. The machine learning model 126 may include at least one machine learning architecture and model parameters. The machine learning model may be a Bayesian machine learning model and / or based on neural networks, such as reinforcement neural networks. The machine learning model may include the design of at least one experimental method. The machine learning model 126 may be configured to consider noise of a defined process variable. The noise may be based on different distributions (e.g., Gaussian) and types (e.g., additive and / or multiplicative). This can be handled accordingly by the machine learning model. The machine learning model 126 may be configured to consider constraints on a target parameter set. The parameters of the target parameter set may have constraints, such as upper and / or lower limits or constants for the pump flow rate that the machine learning model 126 may consider.

[0124] Training may include building a machine learning model 126, and specifically determining and / or updating the parameters 126 of the machine learning model. The machine learning model 126 may be at least partially data-driven. The machine learning model 126 can learn from valid data points. Training may be performed on sensor data, such as considering determined process variables. Training may be performed on process parameters determined in historical production runs (particularly historical production runs with a known set of parameters for flow chemistry settings). The machine learning model 126 may include data-driven model parts and other model parts based on physicochemical laws.

[0125] The determination of the target parameter set in step c) 128 may include at least one optimization step 128. This optimization may include selecting an optimal set of parameters from a parameter space of possible parameters with respect to the optimization objective. The optimization objective may include at least one criterion upon which optimization is performed. The optimization objective may include at least one optimization objective and accuracy and / or precision. The optimization objective may be pre-specified by at least one user of the flow chemistry setup 110, for example. The optimization objective may be a specification of at least one user. The user may select the optimization objective and the desired accuracy and / or precision, for example, via at least one interface. For example, the optimization objective may include at least one value from sensor data. The corresponding concentration of the produced fluid can be determined from the sensor data. The optimization objective may be a measurement determined by sensor 122 with the desired accuracy and / or precision, such as an intensity value determined using a light barrier.

[0126] Optimization may include applying machine learning model 126. Based on the current state of machine learning model 126 and / or based on determined process variables, the optimization algorithm may determine how to optimally set the parameters of the flow chemistry setup. The optimization algorithm may be or may include at least one algorithm for solving at least one optimization problem. Optimization may include solving at least one optimization problem, such as at least one maximization problem or at least one minimization problem. The optimization may include computational steps, such as computing the solution to the optimization problem. The optimization algorithm may be Bayesian optimization, for example, using a Gaussian process as a surrogate model. For example, the optimization algorithm may be a reinforcement learning network. Other optimization algorithms may also be implemented. The optimization algorithm may be configured to consider noise of determined process variables. The noise may be based on different distributions (e.g., Gaussian) and types (e.g., additive and / or multiplicative). This may be addressed accordingly in steps b) and / or c). The optimization algorithm may be configured to consider constraints on a target parameter set. The parameters of the target parameter set may have constraints, such as upper and / or lower limits of pump flow rates or constants that may be considered by the optimization algorithm.

[0127] Optimization step 128 may depend on the machine learning model. However, this may not mean that every decision in each step must depend on the machine learning model 126. For example, multiple experiments may be performed, where the method is trained after the experiments are run. For example, an average window from previous parameter values ​​may also be used. Furthermore, optimization step 128 may include trade-offs between developing and exploring the underlying space.

[0128] In step c)128, the optimization algorithm may determine one or more target parameter sets, such as Pareto-front or subsets of Pareto-front. In the case of multiple target parameter sets, step d), in particular repeating one or more method steps a) to d), may be performed for all configurations of the target parameter sets.

[0129] Research projects typically begin with a screening task. This phase clearly describes the research problem, meaning the optimization objective and a list of influencing parameters are defined. The parameter space is usually large, necessitating numerous experiments. According to the method of the invention, this screening task can be automated with minimal raw material consumption and manual labor. Specifically, as will be outlined in detail below, the instructions for executing the method according to the invention can be implemented as a computer program (particularly software), such that when executed by a computer or computer network, the instructions cause the computer or computer network to perform the method according to the invention. The flow chemistry setup 110 can be fully automated, allowing all relevant parameters (particularly formulation and process parameters) to be controlled within a given range using the method according to the invention. The machine learning model 126 can learn from effective experimental data points, and the optimization steps can provide the parameters required to achieve a given optimization objective for new experimental planning.

[0130] Step d) includes providing a determined set of target parameters 130 and / or (indicated by reference numeral 131) considering the determined set of target parameters for evaluating flow chemistry settings and / or evaluating at least one flow chemistry product.

[0131] The provision 130 may include, for example, presenting and / or displaying and / or transmitting a target parameter set to a user. The target parameter set determined by the provision 130 may be executed using at least one output device 138. The output device 138 may include at least one display device.

[0132] Considering 131 the target parameter set used to evaluate the flow chemistry setup 110 may include setting the parameters of the flow chemistry setup based on the determined target parameter set (especially for the preparation of new experiments).

[0133] As outlined above, one or more, or even all, of the method steps can be repeated, such as once or multiple times. This allows for self-learning and / or self-optimization. Specifically, the determined set of target parameters can be used as the starting point for the next optimization 142. The next optimization 142 may include repeating method steps a) through d), wherein in step a) the process variables of the flow chemistry setup 110 are determined, having parameters set as target parameters determined in the previous round of the method. Steps a) through d) can be repeated until the process variables measured by sensor 122 meet the previously defined target values ​​within a predefined accuracy.

[0134] In addition, Figure 1 An embodiment of an automatic control system 144 for a flow chemistry setup 110 for use in a slug is also described. The automatic control system 144 can be configured to automatically control the flow chemistry setup 110. The automatic control system 144 includes:

[0135] - At least one communication interface 146, configured to receive at least one process variable determined by sensor 122;

[0136] - At least one machine learning model 126, which is configured for training based on process variables;

[0137] - At least one processing unit 148 is configured to determine at least one set of target parameters by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model 126;

[0138] - At least one output device 138 is configured to provide a defined set of target parameters.

[0139] This invention allows for accelerated research processes, particularly for screening purposes. Rapid development of new materials is possible. Users can focus on other tasks involving manual operations. Flow chemistry within the slug allows for resource-efficient research. Only small amounts of raw materials are required for each experiment. Using a slug allows for simplified optimization, especially considering the reduced noise in the input data for the algorithms used in steps b) and c). The slug can provide explicit constraints on measurements. The absence of residue within the slug may result in enhanced input data for the algorithms in steps b) and c). Therefore, enhanced results can be achieved in continuous flow chemistry comparisons.

[0140] List of reference numerals

[0141] 110 Flow Chemistry Setup

[0142] 112 Liquid

[0143] 114 Liquid

[0144] 116 Reactor

[0145] 118 T mixer

[0146] 120. Determine at least one process variable.

[0147] 122 Sensors

[0148] 124 Training

[0149] 126 Machine Learning Models

[0150] 128 Determine the target parameter set

[0151] 130 provides

[0152] 131 Consideration

[0153] 132 Exports

[0154] 134 Preprocessing

[0155] 136 Verification Steps

[0156] 138 Unverified

[0157] 140 Verification

[0158] 142 Next optimization

[0159] 144 Automatic Control System

[0160] 146 Communication Interface

[0161] 148 processing units

Claims

1. A computer-implemented method for determining at least one set of target parameters for a flow chemistry setup (110) in a slug, wherein the method is a self-learning method, the method comprising the steps of: a) Determine at least one process variable by using at least one sensor (122) of the flow chemistry setup (110), wherein the process variable is determined by measuring one or more quantities of a slug flowing through at least one tubular reactor (116); b) Train at least one machine learning model based on the process variables (126); c) Determine the target parameter set by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model (126); d) Provide a defined set of target parameters and / or consider the defined set of target parameters for evaluating the flow chemistry setup (110) and / or evaluating at least one flow chemistry product.

2. The method of claim 1, wherein, The determined set of target parameters is used as the starting point for the next optimization.

3. The method according to claim 1 or 2, wherein, Repeat steps a) to d) until the process variable measured by the sensor (122) meets the previously defined target value within a predefined accuracy.

4. The method according to claim 1 or 2, wherein, The target parameter set includes at least one parameter selected from the group consisting of: the flow rate of at least one pump; temperature; reaction time; at least one parameter from online analysis of the precipitate; and the amount of seed particles.

5. The method according to claim 1 or 2, wherein, The process variables include one or more of the following: at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information; at least one color information.

6. The method according to claim 1 or 2, wherein, The process variables to be determined include one or more of the following: ultraviolet-visible spectroscopy (UV-VIS), Raman spectroscopy, infrared (IR) spectroscopy, nuclear magnetic resonance (NMR) spectroscopy, optical detection, fluorescence spectroscopy, mass spectrometry (MS), high performance liquid chromatography (HPLC), gas chromatography (GC), conductivity and pH determination, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), and automated titration.

7. The method according to claim 1 or 2, wherein, The sensor (122) includes one or more of the following: at least one spectrometer, at least one light barrier, at least one chromatograph, at least one viscometer, at least one titration device, and at least one calorimeter.

8. The method according to claim 1 or 2, wherein, The method includes at least one verification step (136), wherein at least one measurement of the determined process variable is verified, wherein the verification includes comparing the measurement with at least one predefined standard, wherein step a) is repeated if the measurement of the determined process variable is not verified.

9. The method according to claim 1 or 2, wherein, The method includes at least one anomaly detection step, wherein at least one algorithm monitors at least one measurement signal of the sensor (122), wherein the algorithm is configured to determine at least one anomaly, wherein step a) is repeated if an anomaly is detected.

10. The method according to claim 1 or 2, wherein, The optimization objective is at least one user specification.

11. The method according to claim 10, wherein, The optimization objective is the concentration of at least one production fluid.

12. A computer program for determining at least one set of target parameters for a flow chemistry setup in a slug, configured to, when executed on a computer or computer network, cause the computer or computer network to perform, fully or partially, the method according to any one of claims 1 to 11, wherein the computer program is configured to perform at least steps a) to d) of the method according to any one of claims 1 to 11.

13. A computer-readable storage medium comprising instructions, said instructions, when executed by a computer or computer network, causing at least steps a) to d) of the method according to any one of claims 1 to 11.

14. An automatic control system (144) for a flow chemistry setup (110) for flow chemistry in a slug, comprising: - At least one communication interface (146) configured to receive at least one process variable determined by at least one sensor (122) of at least one flow chemistry setup (110), wherein the process variable is determined by measuring one or more quantities of a slug flowing through at least one tubular reactor (116); - At least one machine learning model (126) is configured to be trained based on the process variables; - At least one processing unit (148) configured to determine at least one set of target parameters by applying an optimization algorithm based on at least one optimization objective to a trained machine learning model (126); - At least one output device (138) is configured to provide a defined set of target parameters.

15. The system (144) according to claim 14, wherein, The system (144) is configured to perform the method according to any one of claims 1 to 11.

Images