Transient prediction in reactive fluid flow simulation

Abstract

Transient prediction in a reaction fluid flow simulation. The invention relates to transient prediction in reaction flow simulation. In one embodiment, a method (500) includes generating an initial simulation of a reaction flow for a first duration by using a simulation technique. The method (500) includes predicting, by a machine learning model, a behavior of the reaction flow during the second duration based on the initial simulation. The second duration is continuous with the first duration. The method (500) includes generating a subsequent simulation for the reaction flow for a subsequent duration. The method (500) includes providing data corresponding to the predicted behavior and the subsequent simulation as input to a machine learning model to predict behavior of the reaction flow for a subsequent time period. Further, the method (500) includes repeatedly generating a subsequent simulation and predicting behavior of the reaction stream for one or more successive durations until the reaction of the reaction stream is completed.

Description

Technical Field

[0001] The present invention relates generally to flow simulation, and more particularly to methods and systems for transient prediction in reacting fluid flow simulation. Background Art

[0002] Reactive fluid flow is a type of fluid flow in which chemical reactions occur at the interface between multiple fluids, within a single fluid, or at a solid-liquid interface. Reactive fluid flow involves multiple chemical reactions with exothermic and endothermic reactions that result in the generation and consumption of multiple species / products. The physics involved in reactive fluid flow is highly complex and transient, and reactive fluid flow is governed by different mechanisms (fluid dynamics, kinetics, interfaces, and structures) that exhibit instability.

[0003] One of the main challenges in designing industrial applications with reactive flows is the design methodology. For example, most industrial equipment (such as mixing tanks, burners, etc.) have complex geometries as well as multiple inlets, fast chemical reactions, and high pressures. Designing such complex systems requires a deep understanding of the physics and underlying mechanisms.

[0004] Laboratory scale testing and numerical simulation modeling are two methods commonly used to gather insights about physics and underlying mechanisms to provide design-appropriate solutions. Laboratory scale testing is convenient for testing a wide variety of parameters and conditions, such as temperature, pressure, and pH, in a controlled environment. However, laboratory scale testing can be expensive and may provide limited macroscopic information about the behavior of complex systems. Many times, it may not be economically feasible to conduct multiple experiments, as required in the product development and optimization stages.

[0005] Compared to laboratory scale testing, numerical simulation models such as computational fluid dynamics (CFD) provide a cheaper alternative. CFD is a very powerful and widely used numerical simulation method for studying industrial flows. With the advancement of chemical kinetics, turbulence and reactive flow modeling, CFD has found widespread use in the simulation of reactive flows. However, CFD simulations (especially for unsteady reactive flows) are highly computationally intensive and time consuming. For example, depending on the scale and nature of the industrial application, in a high performance computing (HPC) environment, simulations may take anywhere from a few hours to several months of run time. Therefore, the involvement of CFD simulations in product development and optimization of industrial reactive flow problems is limited.

[0006] In view of the above, a need exists to provide techniques for performing transient prediction of reacting flows in a time-reduced and cost-effective manner. Summary of the invention

[0007] It is therefore an object of the present invention to provide a method and a computing system for reducing the time required in simulating reactions of a reacting flow.

[0008] The object of the present invention is achieved by the method described herein. In a preferred embodiment of the present invention, the present invention describes a method for optimizing the simulation of a reaction flow. The method includes generating an initial simulation of the reaction flow for a first duration by using a simulation technique. In one example, the simulation technique can be a numerical simulation technique, such as a computational fluid dynamics (CFD) simulation technique.

[0009] Based on the initial simulation, the method includes predicting the behavior of the reaction flow during a second duration by a machine learning model. In a preferred embodiment, the second duration is continuous with the first duration. In one example, the machine learning model can be a recurrent neural network (RNN) model. In addition, the method can include generating a subsequent simulation of the reaction flow for a subsequent duration by using the simulation technique. The subsequent duration is continuous with the second duration.

[0010] In addition, the method includes: providing data corresponding to the predicted behavior and subsequent simulations as input to the machine learning model to predict the behavior of the reaction flow for a subsequent time period; determining whether the simulation of the reaction flow is completed; and when it is determined that the simulation of the reaction flow is not completed, repeatedly generating the subsequent simulations and predicting the behavior of the reaction flow for one or more consecutive durations until a complete simulation of the reaction flow is achieved.

[0011] In a preferred embodiment, predicting the behavior of the reacting flow comprises capturing a plurality of images associated with at least one region of interest during the first duration. In one example, the plurality of images correspond to temperature contours of flow characteristics of the reacting flow. The plurality of images depict different time frames for each predefined duration.

[0012] In a preferred embodiment, generating a subsequent simulation for the subsequent duration includes determining, after completion of the second duration, how the prediction predicted by the machine learning model deviates from actual behavior of the reacting flow.

[0013] The present invention provides techniques for accelerating the simulation of reactive fluid flows by employing simulation techniques in conjunction with machine learning models. Thus, simulations associated with unsteady fluid flows can be completed in fewer time periods.

[0014] The object of the present invention is also achieved by a computing system for optimizing the simulation of a reaction flow. The computing system includes one or more processing units and a memory coupled to the processing unit (s). The memory includes a module stored in the form of machine-readable instructions executable by the processing unit. The module is configured to perform the method as described above.

[0015] The object of the present invention is also achieved by a computer program product having machine readable instructions stored therein, which, when executed by one or more processing units, cause the one or more processing units to perform the method steps as described above.

[0016] The object of the present invention is also achieved by a computer-readable medium comprising a computer program code which, when executed by a processing unit, causes the processing unit to perform the method steps as described above.

[0017] The object of the present invention is also achieved by a cloud computing system for optimizing simulation of reactive flows. The cloud computing system includes one or more processing units and at least one memory unit coupled to the one or more processing units. The at least one memory unit includes a simulation optimization module stored in the form of machine-readable instructions executable by the processing unit. The simulation optimization module is configured to perform the method as described above.

[0018] Advantageously, when working in tandem, simulation techniques and machine learning models result in significant reductions in simulation run times. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above-mentioned and other features of the present invention will now be described with reference to the accompanying drawings of the present invention.The illustrated embodiments are intended to illustrate, rather than limit, the present invention.

[0020] The invention is further described hereinafter with reference to the illustrative embodiments shown in the accompanying drawings, in which:

[0021] Figure 1 is a block diagram of a network environment for transient prediction in reactive fluid flow simulation according to one embodiment of the present invention;

[0022] Figure 2 is a graphical user interface according to one embodiment of the present invention, which shows an initial simulation of a reacting flow using a numerical simulation technique in a first time interval;

[0023] Figure 3 is a graphical user interface according to one embodiment of the present invention showing a prediction of reactant fluid flow for a second duration;

[0024] Figure 4is a graphical user interface according to one embodiment of the present invention showing a chart depicting a comparison of the runtimes involved in simulation techniques and machine learning models;

[0025] Figure 5 is a process flow diagram illustrating an exemplary method of performing transient prediction in a reactive fluid flow simulation according to one embodiment of the present invention; and

[0026] Figure 6 is a schematic representation of a cloud computing system according to one embodiment. DETAILED DESCRIPTION

[0027] Various embodiments are described with reference to the accompanying drawings, wherein the same reference numerals are used to refer to the drawings, wherein the same reference numerals are used to refer to the same elements throughout the text. In the following description, for the purpose of explanation, numerous specific details are set forth in order to provide a thorough understanding of one or more embodiments. It may be apparent that such embodiments may be practiced without these specific details.

[0028] Figure 1 1 is a block diagram of a network environment 100 for transient prediction in reactive fluid flow simulation according to one embodiment of the present invention. In particular, Figure 1 A computing system 102 is depicted that is capable of providing transient predictions in reactive fluid flow simulations in industrial applications. Examples of computing system 102 may include, but are not limited to, a laptop, a notebook, a desktop computer. Computing system 102 may be a standalone server, or may be a remote server on a cloud computing platform. In a preferred embodiment, computing system 102 may be a cloud-based computing system.

[0029] The computing system 102 is connected to a plurality of user devices 104-1, 104-2, ..., 104-N, collectively referred to as user devices 104 and individually referred to as user devices 104. Examples of user devices 104 may include, but are not limited to, laptops, notebook computers, desktop computers, and tablet computers. The network environment 100 further includes a database 106. The database 106 may be in communication with the computing system 102 for storing training data related to various predictions of reactive fluid flows.

[0030] In a preferred embodiment of the present invention, computing system 102 is communicatively connected to user device 104 via network 108. Network 108 may be a wireless network, a wired network, or a combination thereof. Network 108 may also be an individual network or a collection of many such individual networks that are interconnected with each other and act as a single large network, such as the Internet or an intranet. Network 108 may be implemented as one of different types of networks, such as an intranet, a local area network (LAN), a wide area network (WAN), the Internet, etc.

[0031] The computing system 102 may include processor(s) 110, interface(s) 112, and memory 114. The processor(s) 110 may include a microprocessor, a microcomputer, a microcontroller, a digital signal processor, a central processing unit, a state machine, a logic circuit, and / or any other device that manipulates signals and data based on computer-readable instructions. In addition, the functionality of the various elements shown in the figures, including any functional blocks labeled as "processor(s)", may be provided using dedicated hardware as well as hardware capable of executing computer-readable instructions.

[0032] The interface(s) 112 may allow the computing system 102 to communicate with the computing system 102 via a wired connection (eg, a local area network, LAN) or via a wireless connection (eg, The interface(s) 112 may also enable communication between different logic and hardware components of the computing system 102.

[0033] The memory(s) 114 may be a computer-readable medium, examples of which include volatile memory (e.g., RAM) and / or non-volatile memory (e.g., erasable programmable read-only memory, i.e., EPROM, flash memory, etc.). The memory(s) 114 may be an external memory or an internal memory, such as a flash drive, a compact disk drive, an external hard drive, etc. The memory(s) 114 may further include data that may be utilized or generated during operation of the computing system 102.

[0034] The computing system 102 may further include (multiple) engines 116 and data 118. The (multiple) engines 116 include simulation engines 120, prediction engines 122, report generation engines 124, and (multiple) other engines 126. The (multiple) other engines 126 may further implement functionality that supplements the functions performed by the computing system 102 or any one of the (multiple) engines 116. On the other hand, the data 118 includes data stored or generated as a result of the functions implemented by the computing system 102 or any one of the (multiple) engines 116. It can be further noted that the information stored in and available in the data 118 can be utilized by the (multiple) engines 116 to perform various functions through the computing system 102. In an example, the data 118 can include simulation data 128, prediction data 130, abnormal data 132, and other data 134. It can be noted that such examples are merely indicative. The present method may be applicable to other examples without departing from the scope of the present subject matter.

[0035] The present invention facilitates reducing the time involved in simulating reactive fluid flows, particularly in industrial applications.The simulation of reactive fluid flows is accelerated by employing both numerical simulation techniques and machine learning models.

[0036] In operation, simulation engine 120 can obtain data associated with two reaction streams. Based on this data, simulation engine 120 can generate the initial simulation of reaction stream. In a preferred embodiment, simulation engine 120 can adopt numerical simulation technology (such as computational fluid dynamics (CFD) simulation) to generate the initial simulation of the behavior of reaction stream. The behavior of reaction stream can indicate flame and flow trajectory, flame dispersion, ignition time scale, flow field dynamics, etc. In CFD simulation, fluid flow and its associated physical properties (such as speed, pressure, viscosity, density and temperature) are calculated based on defined operating conditions. These quantities are calculated simultaneously to obtain accurate physical solutions. CFD simulation uses mathematical models and numerical methods to predict desired flow physics. For example, CFD simulation technology can be based on Navier-Stokes (NS) equations. Simulation engine 120 can store simulation as simulation data 128.

[0037] In a preferred embodiment, the simulation engine 120 may perform an initial simulation for a first duration. In one example, the first duration may be predefined. For example, the first duration may be for 50 milliseconds (ms), etc. In the case of 50 ms, the simulation engine 120 may simulate the reaction flow for t=1 ms to t=50 ms.

[0038] In addition, during the initial simulation, multiple images associated with at least one region of interest may be captured. In one embodiment, during the initial simulation, the computing system 102 may capture multiple images using a high-speed camera. Each of the multiple images may correspond to the behavior of each instance of the reaction flow at the first duration. With reference to the earlier example, if the first time period includes 50ms, the high-speed camera may capture 500 frames, so that each image is captured every 0.1ms. In addition, multiple images may be captured at the region of interest. The region of interest may refer to a specific plane of interest along which the simulation is to be observed.

[0039] In a preferred embodiment, the prediction engine 122 may provide an initial simulation obtained from a CFD technique as an input to the machine learning model. Based on the initial simulation, the machine learning model may predict the behavior of the reaction flow during the second duration. In one example, the prediction engine 122 may employ a ConvLSTM model to predict the behavior of the reaction flow during the second duration. The ConvLSTM model is a hybrid of a convolutional neural network (CNN) and a long short-term memory (LSTM) network.

[0040] In one example, before predicting the behavior of a reactive stream, the prediction engine 122 may train a machine learning model based on a simulation generated using a simulation technique. The ConvLSTM network consists of multiple ConvLSTM layers followed by batch normalization layers sandwiched together in varying numbers. Each layer has different parameters inside, followed by a 2D convolutional layer. The input data set is divided into 80% training and 20% validation. Once predicted, the accuracy of the prediction can be verified by visually comparing the prediction with a simulation performed for the same duration. For the next step, the predictions of the ConvLSTM model that match the simulation are considered, and the predictions in which deviations from the simulation are observed are discarded. The trained machine learning model may be stored in a memory, such as memory 114. The prediction engine 122 may store information about any deviations or anomalies as anomaly data 132.

[0041] In one example, the prediction engine 122 may feed a plurality of images captured during a first duration as input to a machine learning model to predict the behavior of the reactive stream for a second duration. The second duration is continuous with the first duration. Thus, if the first duration is 50 ms, the second duration may start at 51 ms. In one embodiment, the first duration and the second duration may be equal. Next is a simulation of the behavior of the reactive stream that predicts the behavior of the reactive stream resulting in a reduction in the time required to perform simulations for large-scale applications.

[0042] In a preferred embodiment, the ConvLSTM model can predict the subsequent behavior of the reaction flow based on the input provided by the prediction engine 122. The predicted behavior is then compared with the simulation performed for the second duration to check if there are any anomalies in the prediction. If any anomalies are observed, the prediction without the anomaly is considered for performing the simulation.

[0043] Once the behavior is predicted for the second duration, the simulation engine 120 can generate a subsequent simulation for the reaction flow by using simulation techniques. In one example, the subsequent simulation is performed for a duration after the second duration. In one embodiment, if a correct prediction is observed from the machine learning model for 10ms, the subsequent duration will be continuous for 50ms (first duration) + 10ms (second duration), i.e., 60ms. In another embodiment, simulation and prediction can alternate after every 10ms.

[0044] In a preferred embodiment, the prediction engine 122 can provide data corresponding to the predicted behavior and subsequent simulations as input to the machine learning model to predict the behavior of the reaction flow for subsequent time periods. Thus, the input data set provided to the machine learning model is increased so that the machine learning model can provide accurate predictions.

[0045] In a preferred embodiment, the simulation engine 120 can determine whether the reaction between the reaction streams is complete. When it is determined that the reaction of the reaction stream is not complete, the simulation engine 120 can generate subsequent simulations for one or more consecutive durations. Thereafter, the prediction engine 122 can use the simulation as input to predict the behavior of the reaction stream until the reaction of the reaction stream is complete.

[0046] When the reaction is complete, the report generation engine 124 can generate a report based on the simulation and prediction results. For example, the report can depict the time saved by alternating between simulation and prediction. The report can also indicate the type of reaction flow for which the simulation was performed.

[0047] Therefore, the present invention facilitates saving costs and time while speeding up reactive flow simulations, especially for industrial applications. The ConvLSTM model coupled with CFD simulation ensures that the results of the simulation are accurate while speeding up the simulation.

[0048] Figure 2 2 is a graphical user interface (GUI) 200 according to one embodiment of the present invention, which shows an initial simulation of a reaction flow using a numerical simulation technique in a first duration. As described above, for the first duration, a numerical simulation technique (such as a CFD technique) is used to simulate the reaction of the reaction flow. In one example, the first duration is predefined, such as 50 ms. Figure 2 As depicted in , the simulation represents the behavior of an unsteady reacting flow subject to operating conditions. GUI 200 depicts a plurality of images captured over a first duration during an initial simulation. A plurality of images may be captured for at least one region of interest.

[0049] For example, in a preferred embodiment, multiple images can be captured to collect temperature contours of flow characteristics (P, T, V) and species concentrations (O2, CO2, CO, H2O) of the reaction flow. In one example, multiple images can be captured by a high-speed video camera. Each of the multiple images corresponds to the behavior of the reaction flow at each instance of a first duration. For example, in a duration of 50 ms, 500 images can be captured, such that each image is captured at intervals of 0.1 ms. This data is collected at multiple time intervals (called t0, t1, t2, ..., tn) during the simulation process.

[0050] Figure 33 is a graphical user interface 300 according to an embodiment of the present invention, which shows the prediction of the reaction flow in the second duration. In a preferred embodiment of the present invention, a machine learning model (such as a ConvLSTM model) is used to predict the behavior of the reaction flow in the second duration. The ConvLSTM model is a hybrid of a convolutional neural network (CNN) and a long short-term memory (LSTM) network. CNN is a neural network that is very suitable for extracting features from images and correlating features. LSTM network is a recurrent neural network (RNN) that is very suitable for time series prediction. The ConvLSTM model combines the powerful capabilities of the CNN & LSTM framework and is very suitable for predicting reaction flows with flow behaviors that evolve in space & time.

[0051] like Figure 3 As depicted in , a plurality of images captured during a first duration are provided as input to a machine learning model to predict the behavior of a reacting flow for a second duration. The second duration is continuous with the first duration. For example, 500 images corresponding to a simulation of 50ms are provided as input to a ConvLSTM model to predict the behavior of the reacting flow for the second duration. The ConvLSTM model can predict subsequent frames or images of the transient flow of the reacting flow based on the input. The predicted behavior 302 is then compared with the simulation 304 performed for the second duration to check if there are any anomalies in the prediction. The prediction is performed for the same characteristics provided as training data. For example, when a temperature contour image is provided as input to the ConvLSTM model, the ConvLSTM model can predict the temperature contours of the reacting flow in subsequent frames. Similarly, other flow or reaction physical properties (such as pressure, velocity, species concentration, etc.) can be provided as input to obtain corresponding predictions. As shown from Figure 3 As can be seen, after a few milliseconds, there is a deviation in the prediction and the simulation. Therefore, the period until the predicted behavior 302 is similar to the simulation 304 is considered for subsequent simulations.

[0052] Figure 4 4 is a graphical user interface 400 according to one embodiment of the present invention, which shows a chart 402 depicting a comparison of the runtime involved in simulation techniques and machine learning models. The chart 402 includes a line 404 indicating the runtime involved in prediction, and a line 406 indicating the runtime involved in simulation. As can be seen, the use of the ConvLSTM model in a coupled manner with the CFD simulation technique can significantly reduce the simulation runtime.

[0053] For example, referring again to the example above, a simulation for 50 milliseconds (ms) on 120 CPUs with a time step of 0.1 microseconds (μs) may take about 50 hours. This means that it takes about 1 hour to progress through a 1ms simulation. For a good converged solution, the simulation needs to run for at least 150ms (~150 hours of clock time). Assume that after the first 50ms of the CFD simulation, the process switches between ConvLSTM prediction and CFD simulation every 5ms. Such an alternating arrangement results in a speedup of at least 33% in the simulation process.

[0054] Advantageously, the present invention reduces the costs involved in the product development & optimization phase by accelerating the simulation of reactive flows using machine learning models and numerical simulation techniques.

[0055] Those skilled in the art will recognize that, for simplicity and clarity, the complete structure and operation of all data processing systems suitable for use with the present disclosure are not depicted or described herein. Instead, only those of the computing system 102 that are unique to the present disclosure or necessary for understanding the present disclosure are depicted and described. The remaining construction and operation of the computing system 102 may conform to any of the various current implementations and practices known in the art.

[0056] Figure 5 is a process flow diagram illustrating an exemplary method 500 of performing transient prediction in a reactive fluid flow simulation, according to one embodiment of the present invention.

[0057] At step 502, method 500 includes generating an initial simulation of a reaction stream for a first duration. In one example, a simulation technique, such as a numerical simulation technique, is used to generate the initial simulation. In a preferred embodiment, the numerical simulation technique is a computational fluid dynamics (CFD) simulation technique. The CFD simulation technique can be implemented by the simulation engine 120 of the computing system 102 to simulate the mixing process of multiple fluids in any given hybrid. The geometry of the hybrid can be represented by a digital three-dimensional (3D) model. The digital 3D model can be in the form of an electronic file for computer-aided design (CAD), computer-aided manufacturing (CAM), computer-aided engineering (CAE) or other suitable applications.

[0058] The digital 3D model of the mixture can be provided to a computer-implemented simulation method along with other simulation model parameters to simulate the process of mixing reactive fluid flows. The simulation results can be further processed to generate a spatial distribution of fluid concentrations of the mixture.

[0059] In a preferred embodiment, the first duration for performing the initial simulation is 50 milliseconds (ms). During the initial simulation, multiple images associated with at least one region of interest may be captured during the first time period. The multiple images may be captured by a high-speed camera of the computing system 102. Each of the multiple images may correspond to the behavior of each instance of the reaction flow at the first duration. For example, 500 images may be captured in a duration of 50 ms. In addition, the multiple images correspond to temperature contours of flow characteristics, such as pressure (P), temperature (T), volume (V), and species concentration (O2, CO2, H2O) of the reaction flow.

[0060] At step 504, based on the initial simulation, the behavior of the reaction flow is predicted during the second duration. In one example, the prediction engine 122 can use a machine learning model to predict the behavior of the reaction flow for the second duration. In one example, the second duration is continuous with the first duration. In a preferred embodiment, multiple images captured during the initial simulation can be fed to the machine learning model. In one example, the machine learning model is a recursive neural network (RNN) model. Specifically, the RNN model is a convolutional long short-term memory (ConvLSTM) model. Multiple images are fed to the ConvLSTM model to extract features related to the multiple images and predict future frames. For example, based on multiple images, the ConvLSTM model can extract features such as flow trajectories, flame dispersion, mixing. Based on the extracted features, the ConvLSTM model can predict future frames.

[0061] For image sequences, ConvLSTM combines the features and advantages of LSTM and CNN. Prediction engine 122 can store the predicted output as prediction data 130. Although the present invention is described with reference to the ConvLSTM model, those skilled in the art will understand that prediction engine 122 can use any other machine learning model alone or in combination with the ConvLSTM model to predict the behavior of the reaction flow.

[0062] In one example, before predicting the behavior of a reactive stream, a machine learning model can be trained based on simulations generated using simulation techniques. The ConvLSTM network consists of multiple ConvLSTM layers followed by batch normalization layers sandwiched together in varying numbers. Each layer has different parameters inside, followed by a 2D convolutional layer. The input data set is divided into 80% training and 20% validation. Once predicted, the accuracy of the prediction can be verified by visually comparing the prediction to a simulation performed for the same duration. For the next steps, the output of the ConvLSTM model that matches the simulation is considered, and the output in which deviations from the simulation are observed is discarded. The trained machine learning model can be stored in a memory, such as memory 114. The prediction engine 122 can store information about any deviations or anomalies as anomaly data 132.

[0063] At step 506, a subsequent simulation is generated for the reaction flow for the subsequent duration. It will be apparent to one skilled in the art that the subsequent simulation is performed by simulation techniques and that the subsequent duration is continuous with the second time period. In one embodiment, the processor(s) 110 may switch between simulation using numerical techniques and prediction using machine learning models to reduce the time involved in simulating complex problems. The simulation engine 120 may store the output of the simulation as simulation data 128.

[0064] At step 508, data corresponding to the predicted behavior and subsequent simulations are provided as input to the machine learning model to predict the behavior of the reaction flow for subsequent time periods. Thus, the prediction engine 122 can dynamically train the machine learning model by feeding it with large data sets to make predictions in an accurate manner. Thus, for each subsequent prediction, not only the data of the previous simulation will be provided as input to the machine learning model, but also data related to the previous prediction will be provided as input.

[0065] At step 510, it is determined whether the reaction of the reacting flow is complete. For example, based on the flow characteristics and the species concentrations, the processor(s) 110 may determine whether the reaction of the reacting flow is complete.

[0066] At step 512, upon determining that the simulation of the reacting flow is not complete, subsequent simulation of the reacting flow and prediction of behavior are repeated for one or more successive durations until a complete simulation of the reacting flow is achieved.

[0067] Advantageously, use of the ConvLSTM model in a coupled manner with CFD simulation techniques can significantly reduce the time involved in performing simulations of the behavior of reacting flows, especially unsteady reacting flows in various industrial applications.

[0068] Figure 6 6 is a schematic representation of a cloud computing system 600 according to one embodiment of the present invention. The cloud computing system 600 includes a processing unit 602, a memory unit 604, a storage unit 606, a communication interface 608, and a cloud interface 610.

[0069] The processing unit 602 can be one or more processors (e.g., servers). The processing unit 602 is capable of executing machine-readable instructions stored on a computer-readable storage medium such as a memory unit 604 to perform one or more functionalities described in the foregoing description, including but not limited to the simulation of optimized reactive flows. The memory unit 604 includes a cloud computing platform 612 stored in the form of machine-readable instructions and executable by the processing unit 602. Alternatively, the cloud computing platform 612 can take the form of hardware, such as a processor with embedded software. The cloud computing platform 612 includes a simulation engine 120, a prediction engine 122, and a report generation engine 124 stored in the form of machine-readable instructions executable by the processing unit 602.

[0070] The storage unit 606 may be a volatile or non-volatile storage device. In a preferred embodiment, the storage unit 606 includes a database 106 for storing training data related to various predictions of the reaction fluid flow. The storage unit 606 may also store untrained data received from the prediction of the reaction flow. The communication interface 608 acts as an interconnection device between different components of the cloud computing system 600. The communication interface 608 may enable communication between the processing unit 602, the memory unit 604, and the storage unit 606. The processing unit 602, the memory unit 604, and the storage unit 606 may be located in the same location or at different locations.

[0071] The cloud interface 610 is configured to establish and maintain a communication link with the industrial application. Furthermore, the cloud interface 610 is configured to maintain a communication channel between the cloud computing system 600 and the user device 104 .

[0072] In various embodiments, the cloud computing system 600 provides a platform for transient prediction of various reactive fluid flow simulations. The cloud computing system 600 enables simulation and prediction of the behavior of reactive fluid flows used in various industrial applications such as mixing tanks, burners, etc. The cloud computing system 600 combines numerical simulation techniques with predictions of machine learning models to reduce the overall time required to simulate the behavior of reactive fluid flows.

[0073] The present invention may take the form of a computer program product including a program module accessible from a computer-usable or computer-readable medium storing program code, which is used by or in combination with one or more computers, processing units, or instruction execution systems. For the purposes of this description, a computer-usable or computer-readable medium may be any device capable of containing, storing, communicating, propagating, or transmitting a program used by or in combination with an instruction execution system, device, or apparatus. The medium may be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or the propagation medium in and of itself as a signal carrier that is not included in the definition of a physical computer-readable medium includes semiconductor or solid-state memory, magnetic tape, removable computer floppy disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disk (such as compact disk read-only memory (CD-ROM), compact disk read / write, and DVD). As known to those skilled in the art, both the processing unit and the program code for implementing each aspect of the technology may be centralized or distributed (or a combination thereof).

[0074] Although the present invention has been described in detail with reference to certain embodiments, it should be appreciated that the present invention is not limited to those embodiments. In view of this disclosure, many modifications and variations will exist for those skilled in the art without departing from the scope of the various embodiments of the present invention as described herein. Therefore, the scope of the present invention is indicated by the following claims rather than by the preceding description. All changes, modifications and variations within the meaning and scope of the equivalents of the claims are to be considered within their scope. All advantageous embodiments claimed in the method claims may also be applied to the system / device claims.

Claims

1. A method (500) for optimizing a simulation of a reaction flow, the method (500) being performed by a processing unit (110, 602), the method comprising: generating, using a simulation technique, an initial simulation of the reaction flow for a first duration; predicting, by a machine learning model, a behavior of the reaction flow during a second duration based on the initial simulation, wherein the second duration is continuous with the first duration; generating, using the simulation technique, subsequent simulations of the reaction flow for subsequent durations; providing data corresponding to the predicted behavior and subsequent simulations as input to the machine learning model to predict the behavior of the reaction flow for a subsequent time period; determining whether the reaction between the reaction streams is complete; as well as Upon determining that the reaction of the reacting flow is not complete, repeatedly generating the subsequent simulation and predicting the behavior of the reacting flow for one or more successive durations until the reaction of the reacting flow is complete.

2. The method (500) according to claim 1, wherein: Predicting the behavior of the reaction flow further includes: During each generated simulation, a plurality of images associated with at least one region of interest are captured during the first duration.

3. The method (500) according to claim 1 or claim 2, wherein: The plurality of images correspond to temperature contours of flow characteristics of the reacting flow.

4. The method (500) according to claims 1 to 3, wherein: Each of the plurality of images corresponds to behavior of the reactive flow at each instance of the first duration.

5. The method (500) according to one of claims 1 to 4, wherein: Generating the subsequent simulation for the subsequent duration further comprises: After completion of the second duration, it is determined that the predictions made by the machine learning model deviate from actual behavior of the reaction flow.

6. The method (500) according to one of claims 1 to 5, wherein: The method further comprises: Based on the determination, predictions that deviate from the actual behavior of the reaction flow are discarded.

7. The method (500) of claim 1, wherein: The simulation technique is computational fluid dynamics (CFD) technique.

8. The method (500) of claim 1, wherein: The method further includes generating a report based on the simulation and prediction results.

9. The method (500) of claim 1, wherein: The machine learning model is a recurrent neural network (RNN) model.

10. The method (500) of claim 9, wherein: The RNN model is a convolutional long short-term memory (ConvLSTM) model.

11. The method (500) of claim 1, wherein: The method includes training the machine learning model based on simulations generated for the reacting flow using the simulation technique before predicting the behavior of the reacting flow.

12. A computing system (102) having a processing unit (110) for executing the method according to one of claims 1 to 11.

13. A computer program product having a computer program code which, when executed by a processing unit, causes the processing unit to perform the method according to one of claims 1 to 11.

14. A computer-readable medium comprising a computer program product, the computer program product comprising computer program code which, when executed by a processing unit, causes the processing unit to perform the method according to one of claims 1 to 11.

15. A cloud computing system (600), the cloud computing system (600) comprising: one or more processing units (602); as well as At least one memory unit (604) communicatively coupled to the one or more processing units (602), wherein the at least one memory unit (604) includes a simulation optimization module stored in the form of machine-readable instructions executable by the one or more processing units, and wherein the simulation optimization module is configured to execute or initiate a method according to one of claims 1 to 11.

Images