AI-SYSTEM FOR FLOW CHEMISTRY

Abstract

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

Claims

1. A computer implemented method for determining at least one target parameter set for a flow chemistry setup for flow chemistry in slugs, wherein the method is a self-learning method, the method comprising: a) determining at least one process variable by using at least one sensor of a flow chemistry setup; b) training of at least one machine-learning model based on the process variable; c) determining the target parameter set by applying an optimizing algorithm in terms of at least one optimization target on the trained machine-learning model; d) providing the determined target parameter set and/or considering the determined target parameter set for evaluating a flow chemistry setup and/or for evaluating at least one flow chemistry product.

2. The method according to claim 1, wherein the determined target parameter set is used as starting point for a next optimization.

3. The method according to claim 1, wherein steps a) to d) are repeated until the process variable measured by the sensor fits to the previously defined target value within a pre-defined accuracy.

4. The method according to claim 1, wherein the target parameter set comprises at least one parameter selected from the group consisting of: flow rate of at least one pump; temperature; reaction time; at least one parameter from online analytics of an educt; and an amount of seed particles.

5. The method according to claim 1, wherein the process variable is determined by measuring of one or more quantities of slugs flowing through at least one tubular reactor.

6. The method according to claim 1, wherein the process variable comprises at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information, or at least one colorfulness information.

7. The method according to claim 1, wherein the determining of the process variable comprises one or more of: ultraviolet and visible spectroscopy (UV-VIS) spectroscopy, 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), conductometry and pH-determination, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), or automated titration.

8. The method according to claim 1, wherein the sensor comprises one or more of at least one spectrometer, at least one light barrier, at least one chromatograph, a viscometer, at least one titration device, or at least one calorimeter.

9. The method according to claim 1, wherein the method comprises at least one validation step, wherein at least one measurement value of the determined process variable is validated, wherein the validation comprises comparing the measurement value with at least one predefined criterion, wherein step a) is repeated in case the measurement value of the determined process variable is not validated.

10. The method according to claim 1, wherein the method comprises at least one anomaly detection step, wherein at least one algorithm monitors at least one measurement signal of the sensor, wherein the algorithm is configured for determining at least one anomaly, wherein step a) is repeated in case an anomaly is detected.

11. The method according to claim 1, wherein the optimization target is at least one user's specification, wherein the optimization target is a concentration of at least one produced fluid.

12. A computer program for determining at least one target parameter set for a flow chemistry setup for flow chemistry in slugs, configured for causing a computer or a computer network to fully or partially perform the method according to claim 1, when executed on the computer or the computer network, wherein the computer program is configured to perform at least steps a) to d) of the method according to claim 1.

13. A computer-readable storage medium comprising instructions which, when executed by a computer or computer network, cause to carry out at least steps a) to d) of the method according to claim 1.

14. An automated control system for a flow chemistry setup for flow chemistry in slugs comprising: at least one communication interface configured for receiving at least one process variable determined by at least one sensor of at least one flow chemistry setup; at least one machine-learning model configured for training based on the process variable; at least one processing unit configured for determining at least one target parameter set by applying an optimizing algorithm in terms of at least one optimization target on the trained machine-learning model; at least one output device configured for providing the determined target parameter set.

15. The system according to claim 14, wherein the system is configured for performing the method according to claim 1.

Description

SHORT DESCRIPTION OF THE FIGURES

[0101] Further optional features and embodiments will be disclosed in more detail in the subsequent description of embodiments, preferably in conjunction with the dependent claims. Therein, the respective optional features may be realized in an isolated fashion as well as in any arbitrary feasible combination, as the skilled person will realize. The scope of the invention is not restricted by the preferred embodiments. The embodiments are schematically depicted in the Figures. Therein, identical reference numbers in these Figures refer to identical or functionally comparable elements.

[0102] In the Figures:

[0103] FIG. 1 shows an embodiment of a method according to the present invention; and

[0104] FIG. 2 shows an embodiment of a flow chemistry setup and automated control system according to the present invention.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0105] FIG. 1 shows an embodiment of a computer-implemented method for determining at least one target parameter set for a flow chemistry setup 110 for flow chemistry in slugs according to the present invention.

[0106] An embodiment of the flow chemistry setup 110 is shown in FIG. 2. Slugs may be formed by introducing at least two liquids 112, 114 into a tubular reactor 116 at the same time or by introducing a dispersed phase into a continuous phase which flows within the tubular reactor, also denoted reaction-phase and carrier liquid. For introducing the two liquids 112, 114 a T-mixer 118 may be used. Formation of slugs is generally known to the skilled person. Slugs can be seen as small batch reactors which are transported through a tubular reactor one after each other. Slugs may exhibit well-defined interphase mass transfer areas and flow patterns. Two basic mass transfer mechanisms may arise: convection within the individual liquid slugs 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 slugs may comprise segmented, and in particular, non-continuous flow chemistry, with a liquid-liquid flow pattern comprising a plurality of slugs.

[0107] The flow chemistry setup 110 may comprise a plurality of components. For example, the flow chemistry setup 110 may comprise one or more of the at least one reactor 116, at least one mixer such as the T-mixer 118, and further components not shown in FIG. 2 such as at least one pump, at least one valve, at least one heating device, at least one pressure regulator, at least one analytics unit.

[0108] Back to FIG. 1, the method comprises the following method steps which, specifically, may be performed in the given order. Still, a different order is also possible. It is further possible to perform two or more of the method steps fully or partially simultaneously. Further, one or more or even all of the method steps may be performed once or may be performed repeatedly, such as repeated once or several times. Further, the method may comprise additional method steps which are not listed.

[0109] The method comprises the following steps: [0110] a) (denoted with reference number 120) determining at least one process variable by using at least one sensor 122 of the flow chemistry setup 110; [0111] b) (denoted with reference number 124) training of at least one machine-learning model 126 based on the process variable; [0112] c) (denoted with reference number 128) determining the target parameter set by applying an optimizing algorithm in terms of at least one optimization target on the trained machine-learning model 126; [0113] d) providing (denoted with reference number 130) the determined target parameter set and/or (denoted with reference number 131) considering the determined target parameter set for evaluating a flow chemistry setup and/or for evaluating at least one flow chemistry product.

[0114] The parameter set for the flow chemistry setup 110 may comprise settable and/or configurable and/or adjustable characteristics and/or properties of the flow chemistry setup 110. The parameter set may comprise a plurality of parameters. The parameter set may comprise parameters relating to recipe for the reaction and/or process parameters, in particular control parameters. The parameters of the parameter set of the flow chemistry setup 110 may define characteristics and/or properties of the components of the flow chemistry setup 110. The parameter set of the flow chemistry setup 110 may influence one or more of reaction time, reaction rate, slug formation and a final reaction product.

[0115] The target parameter set may be an optimized parameter set for the flow chemistry set up 110. The target parameter set may comprise at least one parameter selected from the group consisting of: flow rate of at least one pump, e.g. flow rates for each pump of the flow chemistry setup or a total flow rate; temperature; reaction time; at least one parameter from online analytics of an educt, e.g. pH value; an amount of seed particles, e.g. for the case of precipitations to control nucleation processes. The reaction time can be adjusted by changing the total flow rate. The target parameter set may comprise at least one parameter relating to a reactor size. Preferably, however, the size of the reactor may be kept constant and/or unchanged. The target parameter set may comprise a parameter relating to the size of the slugs. The size of the slugs can be adjusted by changing a ratio between a reaction-phase and a carrier liquid. Preferably, however, the size of the slugs may be kept constant such that every slug has identical conditions.

[0116] The method is a self-learning method. The method may comprise using at least one artificial intelligence (AI-) system. The method may comprise using at least one machine-learning tool, in particular a deep learning architecture. The method may be performed completely automatic. The complete automatization of the method may allow the AI-system to find the optimal parameters on its own. Specifically, the method may be self-optimizing by setting its parameters iteratively to fulfill a pre-defined final goal without human interaction. To this end, a machine learning model is used. Based on observations the machine learning model 126 facilitates the configuration of the parameters.

[0117] The process variable may be at least one quantity specifying the final reaction product. The final reaction product may be an outcome or output of the flow chemistry process, in particular to an outcome or output of the tubular reactor 116. As shown in FIG. 2, the tubular reactor 116 may comprise at least one outlet 132. The process variable may be determined at the outlet 132 of the tubular reactor 116. The process variable may be determined by measuring of one or more quantities of slugs flowing through the tubular reactor. The process variable may comprise at least one spectral information; at least one intensity information; at least one brightness information; at least one turbidity information, at least one colorfulness information.

[0118] The determining at least one process variable may comprise at least one process of generating at least one measurement value, in particular at least one representative result or a plurality of representative results indicating the process variable. Step a) may comprise one measurement of the process variable or multiple successive measurements of one or more quantities. The flow chemistry setup 110 comprises the at least one sensor 122, specifically a plurality of sensors. The determining of the process variable may comprise one or more of: ultraviolet and visible spectroscopy (UV-VIS) spectroscopy, 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); conductometry and pH-determination, calorimetry, viscosity determination, powder X-ray diffraction (PXRD), and automated titration. The sensor 122 may comprise one or more of at least one spectrometer, at least one light barrier at least one chromatograph, a viscometer, at least one titration device, at least one calorimeter. For example, the sensor 122 may be or may comprise at least one light barrier. The light barrier may be configured for determining at the outlet 132 of the tubular reactor 116 how much light passes through the final reaction product. Specifically, the sensor 122 may be configured for determining intensity of or change of intensity of at least one light beam having passed the final reaction product.

[0119] The sensor 122 may be configured for generating at least one sensor signal. The sensor signal may be or may comprise at least one electrical signal, such as at least one analogue electrical signal and/or at least one digital electrical signal. More specifically, the sensor signal may be or may comprise at least one voltage signal and/or at least one current signal. Further, either raw sensor signals may be used, or the sensor 122 may be adapted to process or preprocess the sensor signal, thereby generating secondary sensor signals, which may also be used as sensor signals, such as preprocessing by filtering or the like. For example, the preprocessing may comprise considering statistics over several slugs, denoted with reference number 134.

[0120] The flow chemistry setup 110 can assess if the flow chemistry setup 110 produces valid data. If the data is not valid the system will repeat the measurement, in particular the experiment, automatically. For example, the method may comprise at least one validation step 136. The validation step 136 may comprise validating at least one measurement value of the determined process variable. The validation may comprise determining suitability of the measurement of the process variable, in particular in view of accuracy and reliability. The validation may comprise comparing the measurement value with at least one predefined criterion. The predefined criterion may be an accuracy criterion such as tolerable measurement error. Step a) may be repeated in case the measurement value of the determined process variable is not validated, denoted with “X” and reference number 138 in FIG. 1. If the measurement value is validated the measurement value may be considered as valid data point, denoted with a hook in FIG. 1. If the measurement value is validated the method may proceed with step b), denoted with reference number 140.

[0121] Additionally or alternatively, the method may comprise at least one anomaly detection step. At least one algorithm may monitor the measurement signal of the sensor 122. The algorithm may be configured for determining at least one anomaly. The algorithm may be based on artificial intelligence. The algorithm may comprise at least one machine-learning algorithm. The machine-learning algorithm may be trained using historic sensor signals, wherein historic the sensor signals may comprise sensor signals having no anomaly and sensor signals having an anomaly. Step a) may be repeated in case an anomaly is detected. If no anomaly is detected the measurement is considered as valid data point. If no anomaly is detected the method may proceed with step b).

[0122] In step b) 124 the machine-learning model 126 is trained. The machine-learning model 126 may comprise at least one machine-learning architecture and model parameters. The machine-learning model may be a Bayesian machine-learning model and/or may be based on neural networks such as a reinforcement neural network. The machine-learning model may comprise at least one design of experiments method. The machine-learning model 126 may be configured for considering noise of the determined process variable. The noise can underlie 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 for considering constraints for the target parameter set. Parameters of the target parameter set can have constraints, e.g., upper and/or lower bounds or constant sum of flow rates of the pumps which might be considered by the machine-learning model 126.

[0123] The training may comprise building the machine-learning model 126, in particular determining and/or updating parameters of the machine-learning model 126. The machine-learning model 126 may be at least partially data-driven. The machine-learning model 126 may learn from the valid data points. The training may be performed on sensor data, such as considering the determined process variable. The training may be performed on process parameters determined in historical production runs, in particular historical production runs, having a known parameter set for the flow chemistry setup. The machine-learning model 126 may comprise data-driven model parts and other model parts based on physico-chemical laws.

[0124] The determining of the target parameter set in step c) 128 may comprise at least one optimization step 128. The optimization may comprise selecting of a best parameter set with regard to the optimization target from a parameter space of possible parameters. The optimization target may comprise at least one criterion under which the optimization is performed. The optimization target may comprise at least one optimization goal and accuracy and/or precision. The optimization target may be pre-specified such as by at least one user of the flow chemistry setup 110.

[0125] The optimization target may be at least one user's specification. The user may select the optimization goal and a desired accuracy and/or precision, e.g. via at least one interface. For example, the optimization target may comprise at least one value of sensor data. A corresponding concentration of the produced fluid can be determined from the sensor data. The optimization goal may be a measurement value determined with the sensor 122, such as an intensity value determined with the light barrier, with a desired accuracy and/or precision.

[0126] The optimization may comprise application of the machine-learning model 126. Based on the current state of the machine learning model 126 and/or based on the determined process variable, the optimization algorithm can decide how to set best the parameters of the flow chemistry setup. The optimization algorithm may be or may comprise at least one algorithm for solving at least one optimization problem. The optimization may comprise solving at least one optimization problem such as at least one maximization problem or at least one minimization problem. The optimization may comprise a computational step such as computing the solution of the optimization problem. The optimization algorithm may be a Bayesian optimization, for example with a Gaussian process as surrogate model. For example, the optimization algorithm may be a reinforcement learning network. Other optimization algorithms may be possible, too. The optimization algorithm may be configured for considering noise of the determined process variable. The noise can underlie different distributions, e.g. Gaussian, and types, e.g. additive and/or multiplicative. This can be handled accordingly in step b) and/or in step c). The optimization algorithm may be configured for considering constraints for the target parameter set. Parameters of the target parameter set can have constraints, e.g., upper and/or lower bounds or constant sum of flow rates of the pumps which might be considered by the optimization algorithm.

[0127] The optimization step 128 may be dependent on the machine-learning model. However, this may not imply that every decision in every step must depend on the machine-learning model 126. For example, a plurality of experiments may be executed, wherein after running of the experiments the method is trained. For example, an average window from previous values of parameters can be used in addition. Moreover, the optimization step 128 may comprise a trade-off between exploitation and exploration of the underlying space.

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

[0129] Usually research projects begin with a screening task. In this stadium, the research question is clearly described which means that the optimization target is defined and a list of influencing parameters is defined. Usually the parameter space is large, which means that a lot of experiments would need to be performed. The method according to the present invention can handle this screening task automatically with a small consumption of ingredients and manual work. Specifically, as will be outlined in detail below, instructions to execute the method according to the present invention may be implemented as computer program, in particular software, such that when the program is executed by a computer or computer network, the instructions cause the computer or computer network to carry out the method according to the present invention. The flow chemistry setup 110 may be completely automatized, in a way that all relevant parameters, in particular recipe and process parameters, may be controlled by the method according to the present invention in a given range. The machine-learning model 126 may learn from the valid experimental data points and the optimization step may plan parameters for new experiments, which are necessary to achieve a given optimization target.

[0130] Step d) comprises providing 130 the determined target parameter set and/or (denoted with reference number 131) considering the determined target parameter set for evaluating a flow chemistry setup and/or for evaluating at least one flow chemistry product.

[0131] The providing 130 may comprise presenting and/or displaying and/or communicating the target parameter set, e.g. to a user. The providing 130 of the determined target parameter set may be performed using at least one output device 138. The output device 138 may comprise at least one display device.

[0132] The considering 131 of the target parameter set for evaluating a flow chemistry setup 110 may comprise setting of parameters of the flow chemistry setup in accordance with the determined target parameter set, in particular preparation of a new experiment.

[0133] As outlined above, one or more or even all of the method steps may be performed repeatedly, such as repeated once or several times. This may allow for self-learning and/or self-optimizing. Specifically, the determined target parameter set may be used as starting point for a next optimization 142. The next optimization 142 may comprise repeating method steps a) to d), wherein in step a) the process variable is determined for a flow chemistry setup 110 having parameters set to the target parameters determined in the previous round of the method. Steps a) to d) may be repeated until the process variable measured by the sensor 122 fits to the previously defined target value within a pre-defined accuracy.

[0134] In FIG. 1, moreover an embodiment of an automated control system 144 for the flow chemistry setup 110 for flow chemistry in slugs is depicted. The automated control system 144 may be configured for automatically controlling the flow chemistry setup 110. The automated control system 144 comprises [0135] at least one communication interface 146 configured for receiving at least one process variable determined by the sensor 122; [0136] the at least one machine-learning model 126 configured for training based on the process variable; [0137] at least one processing unit 148 configured for determining at least one target parameter set by applying an optimizing algorithm in terms of at least one optimization target on the trained machine-learning model 126; [0138] the at least one output device 138 configured for providing the determined target parameter set.

[0139] The present invention may allow to speed up of research process, especially for screening purpose. Fast development of new materials may be possible. Users can concentrate on other tasks involving manual action. Flow chemistry in slugs may allow for a resource efficient research. For each experiment only a small amount of ingredients may be necessary. Using slugs may allow for simplifying optimization, in particular in view of less noise in input data used for the algorithms in steps b) and c). The slugs may provide clear limits for a measurement. No residues may be present within the slugs resulting in enhanced input data for the algorithms in steps b) and c). Thus, in comparison of continuous flow chemistry enhanced results can be achieved.

LIST OF REFERENCE NUMBERS

[0140] 110 flow chemistry setup [0141] 112 liquid [0142] 114 liquid [0143] 116 reactor [0144] 118 T-mixer [0145] 120 determining at least one process variable [0146] 122 sensor [0147] 124 training [0148] 126 machine-learning model [0149] 128 determining the target parameter set [0150] 130 providing [0151] 131 considering [0152] 132 outlet [0153] 134 preprocessing [0154] 136 validation step [0155] 138 not validated, [0156] 140 validated, [0157] 142 next optimization [0158] 144 automated control system [0159] 146 communication interface [0160] 148 processing unit