Scalable physical reactor representation

Abstract

The present invention relates to a method for producing a model reactor configured to reflect the specific defined fluid dynamics of a target reactor of different sizes, the model reactor produced by such a method, and the use of such a model reactor for process up- or down-scaling.

Description

[Technical Field]

[0001]

[0001] The present invention relates to the field of biotechnology production plants for producing useful products, model reactors and their use for scaling up or down of production processes. [Background technology]

[0002] The perfect Downscale Reactor Representation (pDS) attempts to solve the challenges of scale-up in biotechnology manufacturing. Biotechnology process development begins with the use of small laboratory reactors, typically with volumes of about 1 L or less. Processes established in these laboratory reactors are then transferred, step by step, to larger production reactors, typically with volumes up to several cubic meters. The flow physics of small reactors differ significantly from those of larger production reactors, for example, in the distribution of supplied nutrients or the required mixing time. All these parameters depend on the volume and geometry of the reactor used and are physically specific to different reactor sizes. Transferring a tuned process established for a particular reactor size to a different reactor size consumes time and resources, and is therefore expensive. The process must essentially be rebuilt for each reactor size.

[0003]

[0003] U.S. Patent Application No. 2021230532(A1) discloses a system and method for scalable production of therapeutic cells in a bioreactor based on fluid dynamics considerations. This method seeks to optimize conditions for culture without changing the reactor geometry. WO2020205611A1 describes a computational method for modeling a bioreactor by combining mechanistic models of metabolic flux dynamics with flux balance analysis to predict cell culture performance in the reactor. Without a universally valid culture- and product-independent cellular metabolic model, these approaches have limited predictive power.

[0004]

[0004] CN11283644A discloses a computational fluid dynamics (CFD)-based optimization method for a dry anaerobic biogas mixing system. Described is the optimization of CFD modeling by varying different factors within the CFD model itself to obtain a more accurate representation of the CFD modeling of the dry anaerobic biogas mixing system. Furthermore, the use of optimized, high-performance CFD modeling to simplify the computational model and enable rapid evaluation of predefined structures is described. The "optimized model" referred to in CN11283644A relates to the optimization of the CFD modeling process itself, not the internal geometry of the reactor.

[0005]

[0005] Additionally, some work has previously been presented on unstirred vessels and static mixers using CFD as an investigation tool without automated random design. For example, U.S. Patent Application Publication No. 2019 / 0388859(A1) describes a CFD evaluation of a static mixer for a continuous-flow catalytic reactor. The static mixer can be configured for use in a continuous-flow chemical reactor, such as a tubular continuous-flow chemical reactor for heterogeneous catalytic reactions.

[0006]

[0006] CN112100944A describes the use of CFD to visualize multiphase systems and evaluate different predefined reactor geometries. It refers to the visualization of reactor conditions by a simulation tool (CFD), specifically the investigation of flow fields at different scales. The use of particle image velocimetry is shown to validate and optimize the CFD simulation itself for different scales.

[0007]

[0007] Currently used methods for solving process transfer problems are inefficient, time-consuming, and technically complex. For example, adjusting small-scale mixing parameters to a large-scale production reactor can result in suitable mixing characteristics, but on the other hand, it may result in poorer cell growth.

[0008]

[0008] In the past, immediate transfer has been attempted by either taking traditional small-scale culture data and attempting to replicate small-scale conditions at large scale, or by using data from uniformly mixed small-scale cultures to use models to predict behavior in improperly mixed systems. These attempts are inherently likely to fail due to unavoidable scale-dependent flow physics. All of these approaches will have limited success until comprehensive and powerful cell models are available.

[0009]

[0009] An alternative approach is to place cells in an environment that mimics the large-scale production environment. Such an approach uses a two-vessel small-scale system, where one reactor is mixed and fed with nutrients, and the second reactor is not. This can be either a mixed vessel or a tubular reactor. As a complete system, this two-vessel system has the appropriate mixing time of the large-scale vessel, but still does not fully represent the large-scale conditions in terms of nutrient and oxygen distribution, and is technically very complex to implement. A further approach is the use of two vessels in a one-vessel system by dividing the culture volume into two parts with a divider. This achieves the same as a two-reactor system in terms of mixing time, but still does not accurately mimic the large-scale nutrient and oxygen distribution.

[0010]

[0010] Therefore, there is a need for a method that allows for the immediate transfer of process parameters from small laboratory scale to large production scale without causing any complications to the established bioprocess. Summary of the Invention [Problem to be solved by the invention]

[0011]

[0011] The object of the present invention is to provide a method that allows the smooth transfer of a bioprocess established in a model reactor to a corresponding production reactor. This object is solved by the subject matter of the present invention. [Means for solving the problem]

[0012] According to the present invention, there is provided a method for generating a reactor geometry with specific defined fluid dynamics, comprising the steps of: a. Defining an objective function; b. applying an algorithm to generate a number of at least three reactor geometries with randomized structures on the interior walls of the reactor; c. determining the fluid dynamics properties for the generated reactor geometry with the randomized structure of step b) using automated computational fluid dynamics; d. Ranking all generated reactor geometries according to the objective function of step a); e. Selecting the best-fitting randomized structures, the number of selected structures being less than the number of generated structures; f. applying an automated algorithm to randomly subtract and further add randomized structures to these selected best-fit structures of step e) to generate an additional number of at least three reactor geometries with randomized structures; g. iteratively repeating steps c through f until a target measure of similarity between the objective function values ​​and the properties of the randomly generated reactor geometry is found; h. Optionally, preparing a digital file for 3D printing the generated reactor; A method is provided that includes:

[0013] According to one embodiment of the present invention, the randomized structures are derived from a library of structural elements, which are voxel-based appendages or plates.

[0014] A further embodiment relates to the method described herein, wherein the outer diameter, height, and minimum wall thickness of the target reactor are taken as inputs.

[0014]

[0015] A further embodiment relates to the methods described herein, wherein during the random structure generation in steps c and e, the generation of completely enclosed spaces is avoided.

[0015]

[0016] A further embodiment relates to the method described herein, wherein the generation of structures within the space expected due to assemblies inserted into the reactor after the reactor is fabricated is avoided. The assemblies are, for example, sensors, agitators, additional ports, etc.

[0016]

[0017] A further embodiment relates to the methods described herein, wherein the reactor geometry produced does not comprise overhangs according to the intended direction of 3D printing.

[0017]

[0018] Further embodiments relate to the methods described herein, wherein the reactor geometry produced comprises a minimum wall thickness.

[0019] Further embodiments relate to the methods described herein, wherein the reactor geometry produced comprises the smallest opening diameter of any configuration.

[0018]

[0020] A further embodiment relates to the method described herein, wherein the objective function is a target value derived from an existing reactor.

[0021] A further embodiment relates to the method described herein, wherein a single vessel mixing reactor is generated based on defined fluid dynamics derived from a large-scale target reactor.

[0019]

[0022] A further embodiment relates to the methods described herein, wherein the reactor produced is a small-scale reactor.

[0023] Further embodiments relate to the methods described herein, wherein the computational methodology is selected from computational fluid dynamics, engineering equations, and CFD equivalents derived from deep learning AI.

[0020]

[0024] Further embodiments relate to the methods described herein, wherein the target values ​​are one or more of kLa, mixing time, nutrient dispersion, and the like.

[0025] Further embodiments relate to the methods described herein, wherein the objective function is an optimum in any of the hydraulic properties, such as kLa, mixing time, and nutrient dispersion.

[0021]

[0026] Further embodiments relate to the methods described herein, wherein for the second and any further iterations the objective function defined in step a and / or the method of adding elements in step b are adapted.

[0022]

[0027] One embodiment of the present invention relates to a reactor that reflects the same mixing characteristics as a target reactor of different sizes, and that exhibits a randomized structure in the interior wall.

[0028] The reactor may be a miniature reactor that mirrors the same fluid flow characteristics as the target reactor. The reactor exhibits a specifically designed internal wall structure that allows it to mirror, for example, the mixing characteristics of an existing target reactor that may be the same size or different.

[0023]

[0029] Further embodiments relate to the methods described herein, wherein the reactor is produced by the methods described herein.

[0030] Further embodiments relate to the methods described herein, wherein the reactor is 3D printed.

[0024]

[0031] One embodiment of the present invention relates to the use of the reactor described herein.

[0032] A further embodiment relates to the use of the reactor described herein, wherein the reactor is used for the scale-up or scale-down of a bioprocess.

[0025]

[0033] A further embodiment relates to the use of the reactor described herein, wherein the reactor is used to transfer a bioprocess to a large-scale bioreactor.

[0026]

[0034] A further embodiment relates to the use of the reactor described herein, wherein the reactor is used for transferring a bioprocess from large scale to large scale.

[0035] One embodiment of the present invention is a method for scaling up a bioprocess, comprising: a. determining at least one objective function derived from large-scale reactor characteristics; b. Producing a small-scale model reactor according to the method of any one of claims 1 to 17; c. 3D printing a small-scale model reactor; d. setting up the bioprocess in a small-scale model reactor; e. determining process parameters; f. transferring the process parameters to a large-scale reactor; The present invention relates to a method comprising:

[0027]

[0036] Further embodiments relate to the methods described herein, wherein the determined objective function is one or more selected from mixing time, nutrient feed mixing, shear rate, oxygen distribution, nutrient distribution, mass transfer, and the like.

[0028]

[0037] As used herein, the term "objective function" refers to any function that formulates a goal. This objective function can have an optimization goal of minimum, maximum, range, or specific value, and can consist of one or more constraints / equations that can be combined into one equation. The objective function can be as simple as a maximization / minimization goal of one parameter, such as mixing time, or can include any number of aspects in any combination, such as minimum mixing time combined with a specific kLa.

[0029]

[0038] The specific defined fluid dynamics may be determined by evaluation of the fluid dynamics of an existing reactor or by defining optimization parameters. The fluid dynamics of an existing reactor may be determined by computational fluid dynamics (CFD) software.

[0030]

[0039] According to one embodiment of the invention, specific defined fluid dynamics of an existing large scale reactor are determined and used, or specific defined optimization parameters are used, which are either selected from a group of process related parameters such as mixing time, power input, etc., or selected from a group of reactor related parameters such as volume, surface area, etc.

[0031]

[0040] According to one embodiment of the present invention, the specific geometry of the reactor structure is generated based on the results obtained from the CFD software in terms of mixing times or component distribution calculated by the CFD software. According to one example, the mixing times or component distribution aspects of dissolved oxygen and / or nutrients are calculated by the CFD software.

[0032]

[0041] One embodiment of the present invention relates to a model reactor that reflects the same mixing characteristics as an existing reactor, and exhibits a structure that replicates the mixing characteristics of the existing reactor, the model reactor having a predetermined wall structure.

[0033]

[0042] The model reactor may be a small-scale reactor, a medium-scale reactor, or even a large-scale reactor.

[0043] The model reactor may be produced by 3D printing.

[0034]

[0044] One embodiment of the present invention relates to the use of the model reactor described in this invention for the scale-up or scale-down of bioprocesses.

[0045] 1. A method for scaling up a manufacturing process, comprising: a. determining process conditions for a large-scale production reactor; b) transferring the determined process conditions to a small-scale reactor; c. evolving a model reactor structure based on the transitions of step b); d. generating a 3D model of the reactor showing the structure resulting from step c); e. Establishing a manufacturing process for the model reactor of step d); f. Transferring the established process to a large-scale production reactor; A method comprising:

[0035]

[0046] The transition and evolution of the model reactor may be performed by computational fluid dynamics (CFD) software, for example.

[0047] According to one embodiment of the present invention, the determined process conditions are one or more selected from mixing time, mixing of nutrient feed, shear rate, mass transfer. [Brief explanation of the drawings]

[0036] [Figure 1A]

[0048] FIG. 1A: A diagram showing the stepwise process scale-up from laboratory to large scale that is conventionally used. [Figure 1B]

[0049] FIG. 1B: Scale-up scheme for pDS reactor technology according to the present invention. [Figure 2]

[0050] FIG. 1 illustrates the difference between traditional digital model-based scaling-up and the generation of specialized physical small-scale reactor representations and their use for process scaling-up. [Figure 3A]

[0051] Figure 1 shows the detailed decision parameters and iterative approach to generate the next reactor design using random structure generation / subtraction / selection. At each generation (iteration) of the process, the reactor is expected to increasingly meet the desired target value (objective function). [Figure 3B]

[0052] FIG. 1 illustrates a selected generation map showing the interrelationships of different generations and their respective parent and daughter structures. [Figure 4]

[0053] FIG. 1 shows the mixing times for each generation, starting from generation 1 up to generation 5, to reach a mixing time of at least 40 seconds to represent large-scale mixing. [Figure 5A]

[0054] FIG. 1 shows the addition of tracer from the top into a small-scale bioreactor vessel (1 L) and the resulting dispersion of tracer after 8 seconds. [Figure 5B] FIG. 1 shows the addition of tracer from the top into a small-scale bioreactor vessel (1 L) and the resulting dispersion of tracer after 21 seconds. [Figure 6]

[0055] FIG. 10 shows mixing curves for different reactor geometries giving different mixing times and different mixing behavior compared to exponential washout behavior. [Figure 7]

[0056] FIG. 1 illustrates a workflow starting with a large-scale target reactor, followed by determining the fluid dynamic properties with CFD software, leading to a final small-scale laboratory model reactor exhibiting the same fluid dynamic properties as the large-scale target reactor. [Figure 8]

[0057] FIG. 1 illustrates a method for large-scale to large-scale process transfer. DETAILED DESCRIPTION OF THE INVENTION

[0037]

[0058] The present invention provides a method for the smooth transfer of a bioprocess established in one reactor to a target reactor, where the target reactor is of a different size.

[0059] One embodiment of the present invention relates to a method for designing a small-scale reactor with a specific mixing time, or the largest possible mixing time. The most significant difference between a small-scale reactor and a large-scale reactor is the mixing time, which is significantly greater in a large-scale reactor. This basic, unavoidable hydraulic principle is one of the main reasons why the same biological process performed in a small-scale reactor will perform differently compared to a large-scale reactor, and vice versa. For example, a process can be initially developed at a small scale that inherently exhibits the performance of a large-scale reactor by providing a small-scale reactor with the same length of mixing time as the large-scale reactor.

[0038]

[0060] One issue that is often overlooked is the bulk fluid flow of the system and how this varies across scale. Fluid displacement and flow velocity are not linear across scale, and therefore the mixing characteristics of the liquid motion do not vary solely based on volume, tip speed, and diameter; they also depend on the spatial distribution of forces within the reactor and the formation or lack of small and large vortices. From a practical perspective, designs should degrade the local mixing performance of smaller-volume systems to match the bulk mixing and mass transfer performance of larger systems. A good design will take into account the effects of straighteners or flow disruptors.

[0039]

[0061] Typical process development for biotechnology involves first determining optimal parameters at a small scale and then transferring the optimized process to a large scale. Because mixing times in large reactors are significantly longer than in small-scale reactors and changes in process performance are expected during scale-up, processes are currently scaled up stepwise by increasing reactor size (see Figure 1A) to reduce costs and risks if process performance decreases at a large scale. This process is time-consuming and expensive, and because developing the process at a production scale originally is prohibitively expensive, the process cannot achieve optimal process parameters for large-scale production. To reduce costs and risks, pDS is used for scale-up (see Figure 1B).

[0040]

[0062] In addition to saving costs and time, efficient optimization and scale-up of biotechnology processes is crucial for achieving several Sustainable Development Goals. This is especially true when considering the impact of microbial biotechnology in the agri-food, environmental, biopharmaceutical, and chemical industries. Many of these applications require scale-up after proof-of-concept. However, the behavior of such biopharmaceutical culture systems remains unpredictable when transferred from small laboratory scale to industrial conditions. To that end, a complete scale-up or downscaling computational framework is essential. While model-based scale-up already exists, as described herein, these currently available models are product- and host-specific. This means that for each process, data from large-scale production must be generated to recalibrate the model, severely limiting their usefulness in terms of time and cost savings (Figure 2). The present invention provides direct scale-up from laboratory to production scale. The inventive method is a host- and product-independent scale-up or downscaling method. The present invention also provides a model reactor with large mixing times.

[0041]

[0063] Scaling up fermentation processes in the biotechnology industry is a crucial step in commercializing innovative bioprocesses and bioproducts, and life sciences will be a keystone of the future circular economy. This includes carbon capture technologies, bioplastics, and novel foods, as well as biopharmaceuticals such as cancer treatments, vaccines, and gene therapies. As noted above, the scale-up step is notorious for performance losses and delays in development. While downscaling approaches have been proposed as a sensible way to minimize these issues, current technologies are either technically complex (two-vessel systems) or highly limited by human design instead of automated design. The presented approach consists of interrelated activities to overcome current shortfalls. Detailed analysis of conditions in large-scale target reactors Computational representation of fluid dynamics properties Transfer of this hydrodynamic characterization to a laboratory-scale model reactor using automated structure generation of a representative small-scale bioreactor (see Figure 2) Production of such small-scale model reactors, e.g., by 3D printing, showing their complex internal geometries Development of a regulated process at laboratory scale under conditions representative of the large-scale target reactor Finally, transfer of successful findings directly to the large-scale target reactor without any required intermediate steps (Figure 1B).

[0064] The present invention provides a method for avoiding the time-consuming, resource-demanding, and costly scale-up of currently applied biotechnology manufacturing processes, which are first established in small laboratory reactors and then transferred in several subsequent volume-up steps to larger reactors until a final production volume of up to several cubic meters is reached.

[0042]

[0065] The target reactor may be a large-scale production reactor for use in a plant for the production of various desired biotechnology products. Large-scale production reactors may have volumes of about 100 L, 500 L, 1000 L, or up to several cubic meters.

[0043]

[0066] The model reactor may be a small-scale laboratory reactor. The small-scale model reactor may have a volume of 0.1 L, 0.5 L, 1.0 L, 1.5 L, 2.0 L, 5.0, or up to 10.0 L.

[0044]

[0067] In particular, flow physics vary significantly in reactors of different volumes and geometries. For example, nutrient distribution and mixing times depend primarily on the volume and geometry of the reactor used. Each time an existing, well-established process is transferred to a different sized reactor, significant time and resources are required to rebuild the process with the aim of achieving the same results in the target reactor, an aim that is not always met.

[0045]

[0068] One embodiment targets mixing times representative of large scale reactors, e.g., 20 m 3 Reactors typically have a mixing time of approximately 40-50 seconds. This mixing time of approximately 40-50 seconds should be reflected in the generated small-scale model reactor. The overall decision scheme for random structure generation to produce the specific hydraulic behavior of interest in the small-scale model reactor is shown in Figure 3. The structure generation and attrition loop is run as long as needed to produce the specific end goal. In this case, the algorithm automatically generates the next reactor geometry based on the previous iteration and automatically evaluates reactor geometries via CFD (determining their mixing time) until a reactor structure exhibiting the desired mixing time of approximately 40 seconds is generated.

[0046]

[0069] Using this methodology, mixing times will necessarily increase with each generation if appropriate parameters are selected for the amount and type of structures created and removed in each iteration. According to an exemplary embodiment of the present invention, a 1 L bioreactor was designed using the described methodology with an end goal of at least 40 seconds of mixing time. Bioreactors of 1 L size typically have mixing times of only 2-5 seconds, which is too fast to represent the 40+ seconds of mixing times of larger-scale reactors. Figure 4 shows the mixing times for each generation of a randomly generated reactor using the described methodology, demonstrating the successful increase in mixing time with each iteration. These progressively longer mixing times are achieved by adding and subtracting random structures to the 3D structure of the bioreactor, and for each generation, the best-performing structure according to the objective function is taken as the basis for adding / subtracting structures for the next generation.

[0047]

[0070] Because adding structure to the internal walls impairs efficient mixing in small-scale reactors, the increased mixing time depends on the addition of structure to the internal walls, which prevents mixing of any added materials, such as nutrients. The nature and design of the structures for the internal walls in this embodiment are chosen to be purely random, and the mixing behavior can be easily visualized by adding scalars to the CFD simulation from the top of the reactor to simulate the addition of nutrients via feed lines at the top of the reactor. Figures 5A and 5B show such tracer distribution within one of the reactor structures generated in this embodiment, which exhibited a mixing time of over 40 seconds. This behavior is not possible in conventional bioreactors, as the materials within the reactor would be uniformly mixed within such small-scale reactors after 2-5 seconds. Only the iterative random structure approach allows this type of tracer distribution to represent large-scale reactors.

[0048]

[0071] The resulting standard deviation of the tracers over time in the bioreactor also allows visualization of the differences from traditional single-plate-separated one-vessel or two-vessel systems. For two-vessel systems, as well as for bioreactors separated into two sections by a perforated plate, the resulting curves will always resemble an exponential washout curve. This exponential washout curve is not seen in larger bioreactors because they consist of different zones that mix differently with the main bulk of the bioreactor liquid, resulting in more complex nutrient mixing over time (Figure 6). The mixing curve depicted by the standard deviation of the tracers over time shown attempts to shift the behavior from a subexponential washout behavior (lower curve) to a more complex mixing behavior that is potentially more suitable as a scale-down bioreactor for process development and is more representative of how the mixing curve will look in a large-scale bioreactor. The presented methodology can either use mixing time as the objective function or use the resulting mixing curve presented in Figure 6 as the objective function to produce a physical small-scale reactor that more accurately represents the conditions in a large-scale bioreactor.

[0049]

[0072] In one embodiment of the present invention, the reactor was numerically evaluated using M-Star CFD, although this evaluation could, of course, be performed with any other CFD modeling software, such as Siemens StarCCM+ or open-source solutions like OpenFoam. The basic idea of ​​CFD is to solve the equations of fluid motion (i.e., the Navier-Stokes equations) within the reactor. These equations describe the conservation of mass and momentum balance in the fluid. Momentum balance, in particular, is nonlinear in nature, which prevents analytical solutions for virtually all real flows. This necessitates discretizing the mathematical equations and solving them numerically. As the scale increases, flows become more turbulent, and the characteristics of turbulence span a wide range of scales, both spatially and temporally, leading to significant differences in mixing between small- and large-scale reactors.

[0050]

[0073] The histograms of well-mixed / poorly mixed regions, as well as all other values ​​or columns calculated by CFD, can be used as objective functions in the presented embodiment to determine whether large-scale and small-scale reactors exhibit the same flow characteristics. According to the present invention, any target property ranking may be used, provided the target property can be calculated by appropriate software, such as CFD. Even combinations of ranked properties and properties can be used, such as targeting a mixing time of more than 40 seconds and simultaneously a specific kLa gas transfer coefficient.

[0051]

[0074] The reactor geometry, both small and large scale, may be chosen arbitrarily as long as it can be modeled in 3D. The geometry may be a simple cylindrical reactor, with a known volume, agitator, optionally including feed piping, and any other configuration, size, or structure required for use of the reactor.

[0052]

[0075] For one embodiment according to the invention, a simple cylindrical reactor with a volume of 1 L was used as a model reactor, although any other reactor geometry is possible.

[0076] The reactor provided by the present invention is a physical, scaled-down model of large-scale hydraulic behavior and is therefore suitable for host and product independence. The lack of a sufficiently detailed model of cellular behavior, which is an obstacle to the predictive power of digital models for conventional approaches (top panel of Figure 2), does not apply to the present invention because no cellular modeling is performed. The cells used are transferred into a physical environment representative of a large-scale production reactor, allowing direct scaling-up of process parameters developed on a small scale to large-scale production without loss of performance (right side of Figure 1).

[0053]

[0077] Finally, after digitally generating the geometry of the small-scale reactor, which represents the mixing behavior of the large-scale reactor, a small model reactor can be 3D printed, since the method according to the present invention already provides a digital file for 3D printing. Furthermore, 3D printing is the only manufacturing technology that allows for the accurate and rapid production of the complex geometries resulting from the described methodology. The production of the model reactor can be easily achieved by 3D printing. Since the algorithm can reliably prepare the reactor for 3D printing, the physical scaled-down reactor can be realized by any 3D printing technology. Suitable 3D printing processes are, for example, laser sintering, FDM, DLP, LCA, or any other methodology. Suitable 3D printing materials are, for example, thermoplastics such as PLA or ABS, as well as resins, medical resins, glass and ceramic resins, and any sintered material, from plastic to metal, without any limitation. The interconnection between the methodology for generating the reactor geometry and its subsequent manufacturing is shown in Figure 7.

[0054]

[0078] Mixing properties are evaluated on a large scale, and the small scale is iteratively developed to achieve the same flow properties by randomly varying its structure and selecting the best performer for the next iteration.

[0055]

[0079] Different methods can be used to select model reactors for each generation (e.g., using the best x-ranking and additionally using some outliers). Additionally, different mutation methods can be used, potentially only for specific generations (e.g., using larger changes in reactor structure in the first iteration and smaller changes in subsequent iterations).

[0056]

[0080] The process of getting from a model reactor to a large-scale production reactor is an evolutionary approach (random modification and selection of the best performer), where the same pre-specified objective function is used.

[0057]

[0081] The optimization loop mainly includes the following steps: 1. Steps for selecting a suitable model reactor for the loop -If this is the first loop, the provided model reactor will be used - Otherwise, the reactor with the best rating (or a number of best performing reactors) is selected. 2. A step in which the model reactor is crossed or randomly mutated with other high-performing specimens from the previous generation to create several reactor variants (approximately 100 specimens, this number can be chosen freely). 3. Step 3: All mutated reactors are run in CFD software 4. A rating based on the CFD results is calculated, taking into account the main parameters of the large-scale production reactor. 5. This loop is repeated as long as required to get the calculated CFD results close enough to the pre-specified key parameters, which may take up to 50 loops (this number can vary depending on the individual simulation).

[0058]

[0082] The model mutated reactor is generated based on the following steps: a. Creating random structures until a target volume is reached b. Removing random structures until a target volume is reached c. Creating a weighted random structure with seed points to achieve greater structure and more variation with each generation. d. Creating structures from predefined elements (e.g., disks on walls, bowls, etc.) e. Varying the size and position of the structure of the predefined elements f. Using a predefined element as the origin of the random structure

[0083] Model reactors may be generated using a mixed approach, for example, by creating a predefined structure, followed by a weighted random structure approach, and finally mutating the predefined structure accordingly.

[0059]

[0084] Finally, each algorithm is used to generate a 3D printable file of the model reactor, whose quality is checked for having a verification structure, e.g., no holes in the model reactor walls, and a printable structure that does not require any supports.

[0060]

[0085] The optimization loop is completed when the CFD key parameters are close enough to the predefined parameters of the large-scale production reactor.

[0086] Another embodiment of the present invention uses the small-scale representation reactor described herein for process transfer of a large-scale process operating in one facility to another facility, where the processes in both facilities use large-scale reactors.

Claims

1. 1. A method for generating reactor geometries with specific defined fluid dynamics, comprising: a. defining an objective function; b. applying an algorithm to generate a number of at least three reactor geometries with randomized structures on the interior walls of the reactor; c. determining the fluid dynamics properties for the generated reactor geometry with randomized structure of step b) using automated computational fluid dynamics; d. Ranking all generated reactor geometries according to the objective function of step a); e. Selecting the best-fitting randomized structures, wherein the number of selected structures is less than the number of generated structures; f. applying an automated algorithm to randomly subtract and further add randomized structures to these selected best-fit structures of step e) to generate several additional at least three reactor geometries with randomized structures; g. Iteratively repeating steps c through f until a target measure of similarity between the objective function values ​​and the properties of the randomly generated reactor geometry is found; h. Optionally, preparing a digital file for 3D printing said generated reactor; A method comprising:

2. The method of claim 1 , wherein the randomized structure is derived from a library of structural elements.

3. The method of claim 2 , wherein the structural elements are voxel-based appendages or plates.

4. 4. The method of any one of claims 1 to 3, wherein the outer diameter, height, and minimum wall thickness of the target reactor are taken as inputs.

5. 5. The method according to claim 1, wherein during the random structure generation in steps c and e, the generation of completely enclosed spaces is avoided.

6. 5. The method according to claim 1, wherein the formation of structures in the space expected due to assemblies inserted into the reactor after the reactor has been manufactured is avoided.

7. The method of claim 6, wherein the assembly is a sensor, an agitator, an addition port, or the like.

8. 8. The method of claim 1, wherein the generated reactor geometry does not comprise overhangs according to the intended direction of the 3D printing.

9. 9. The method of claim 1, wherein the generated reactor geometry comprises a minimum wall thickness.

10. 10. The method of any one of claims 1 to 9, wherein the generated reactor geometry comprises a minimum opening diameter of any configuration.

11. 11. The method of claim 1, wherein the objective function is a target value derived from an existing reactor.

12. 12. The method of any one of claims 1 to 11, wherein a single-vessel mixing reactor is generated based on the defined fluid dynamics derived from a large-scale target reactor.

13. 12. The method of any one of claims 1 to 11, wherein the generated reactor is a small-scale reactor.

14. 14. The method of any one of claims 1 to 13, wherein the computational methodology is selected from CFD equivalents derived from computational fluid dynamics, engineering equations, and deep learning AI.

15. 15. The method of claim 11 or 14, wherein the target values ​​are one or more of kLa, mixing time, nutrient dispersion, and the like.

16. 16. The method of any one of claims 1 to 15, wherein the objective function is an optimum in any of the hydraulic properties, such as kLa, mixing time, and nutrient dispersion.

17. 17. A method according to any one of claims 1 to 16, wherein for the second and any further iterations the objective function defined in step a and / or the method of adding elements in step b are adapted.

18. A reactor that reflects the same mixing characteristics as a target reactor of different size, but exhibits a randomized structure on the inner wall.

19. 19. The reactor of claim 18 produced by the method of any one of claims 1 to 17.

20. 20. The reactor of claim 18 or 19, which is 3D printed.

21. 21. Use of a reactor according to any one of claims 18 to 20, wherein the reactor is used for the scale-up or scale-down of a bioprocess.

22. 22. The use according to claim 21, wherein the reactor is used for transferring a bioprocess to a large-scale bioreactor.

23. 21. The use of claim 20, wherein the reactor is used for transferring a bioprocess from large scale to large scale.

24. 1. A method for the scale-up of a bioprocess, comprising: a. determining at least one objective function derived from large-scale reactor characteristics; b. Producing a small-scale model reactor according to the method of any one of claims 1 to 17; c. manufacturing the small-scale model reactor by 3D printing; d. Setting up the bioprocess in the small-scale model reactor; e. Determining process parameters; f. transferring the process parameters to the large-scale reactor; A method comprising:

25. 25. The method of claim 24, wherein the determined objective function is one or more selected from mixing time, nutrient feed mixing, shear rate, oxygen distribution, nutrient distribution, mass transfer, and the like.

Images