Visual analysis method and device for combustion simulation

Abstract

The present application relates to the technical field of combustion state monitoring, and discloses a visual analysis method and device for combustion simulation, which comprises the following steps: under different initial conditions of combustion simulation, generating multiple sets of time sequence flow images for each combustion simulation example; encoding each frame of image in each set of time sequence flow images to obtain corresponding multiple high-dimensional embedding vectors; generating a scatter plot after dimension reduction of the multiple high-dimensional embedding vectors after dimension reduction clustering, and generating multiple time sequence trajectories for each combustion simulation example based on the scatter plot after dimension reduction; searching for a similar time sequence trajectory most similar to a specified time sequence trajectory within a preset convergence radius, and analyzing combustion convergence and stability.The present application makes full use of the correlation and similarity between data, can accurately identify similar combustion evolution modes, helps experts more accurately grasp key characteristics such as stability and oscillation in the combustion process, deeply reveals the combustion law, and improves the precision of ramjet combustion simulation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of combustion state monitoring, and particularly relates to a visual analysis method and device for combustion simulation. BACKGROUND

[0002] Supersonic combustion research in ramjet engines has received extensive attention due to its key role in achieving sustained hypersonic flight. Early modeling of turbulent and reacting flows laid the foundation for modern computational research of ramjet engines. However, the complex coupling between fluid and chemical processes, as well as the resulting high-dimensional data stream, still poses challenges to simulation accuracy and subsequent analysis.

[0003] Although significant progress has been made in related methods, the integration of expert knowledge and semantic automatic interpretation in the field of combustion is still underdeveloped, making it difficult to understand ramjet combustion phenomena and reducing the efficiency of communication about ramjet combustion phenomena, resulting in low simulation accuracy of ramjet combustion. SUMMARY

[0004] Therefore, the present application provides a visual analysis method and device for combustion simulation to solve the problem of low simulation accuracy of ramjet combustion in the prior art.

[0005] In a first aspect, the present application provides a visual analysis method for combustion simulation, which comprises:

[0006] Under different initial conditions of combustion simulation, a plurality of sets of time sequence flow images are generated for each combustion simulation case;

[0007] Each frame of image in each set of time sequence flow images is encoded to obtain a plurality of high-dimensional embedding vectors corresponding thereto;

[0008] After dimensionality reduction clustering of the plurality of high-dimensional embedding vectors, a scatter plot after dimensionality reduction is generated, and a plurality of time sequence trajectories for each combustion simulation case are generated based on the scatter plot after dimensionality reduction;

[0009] Within a preset convergence radius, a similar time sequence trajectory most similar to a specified time sequence trajectory is retrieved, and combustion convergence and stability are analyzed based on the specified time sequence trajectory and the similar time sequence trajectory.

[0010] The combustion simulation visual analysis method provided by the application generates multiple sets of time sequence flow field images under different initial conditions, providing a rich data basis for comprehensively exploring the combustion simulation rules. The images are encoded to obtain high-dimensional embedding vectors, and then dimensionality reduction clustering is performed to generate scatter plots and time sequence trajectories, which converts complex combustion flow field data into intuitive visual forms, greatly reduces the complexity of the data, enables experts to quickly grasp the overall characteristics and distribution rules of the data, and significantly improves the efficiency of data analysis. In terms of analysis accuracy and depth, the time sequence trajectory is generated based on the scatter plot after dimensionality reduction, and the most similar trajectory is searched within the preset convergence radius, and the specified trajectory and the similar trajectory are combined to analyze the combustion convergence and stability. This way fully utilizes the relevance and similarity between data, and can accurately identify similar combustion evolution patterns, helping experts more accurately grasp key characteristics such as stability and oscillation in the combustion process, deeply reveal the combustion rules, provide a more reliable basis for understanding the combustion mechanism, improve the ramjet combustion simulation precision, and solve the problem of low ramjet combustion simulation precision in the prior art.

[0011] In an optional embodiment, the initial conditions include static pressure, static temperature and water vapor concentration.

[0012] Under different initial conditions of combustion simulation, multiple sets of time sequence flow field images of each combustion simulation example are generated, including:

[0013] Different values of static pressure, static temperature and water vapor concentration are combined, and the combustion process is simulated under each combination to generate multiple sets of combustion flow field time slices;

[0014] The same color and scaling ratio are used to render different time sequence flow field components in each set of combustion flow field time slices, remove irrelevant background information, and retain important regions of different time sequence flow field components to obtain multiple sets of time sequence flow field images corresponding to each initial condition. Different time sequence flow field components include pressure, OH radical, Mach number and velocity.

[0015] The combustion simulation visual analysis method provided by the application, static pressure, static temperature and water vapor concentration are key initial parameters affecting the combustion process of ramjet engines, and different combinations of initial parameters can cover various combustion conditions. By generating time sequence flow field images under each combination, the changing characteristics of the combustion flow field over time under different conditions can be fully captured, avoiding missing key combustion phenomena due to incomplete consideration of initial conditions, providing rich and representative basic data for subsequent analysis, and ensuring that the analysis results can reflect various actual combustion scenarios. The same color mapping and scaling ratio are used to render different time sequence flow field components such as pressure, OH radical, Mach number and velocity, ensuring the consistency of different components in visual presentation, eliminating interference caused by differences in rendering methods, and enabling subsequent encoding to focus more accurately on the feature differences of the flow field itself.

[0016] In an optional implementation, each frame image in each group of time sequence streamline images is encoded to obtain a plurality of high-dimensional embedding vectors, including:

[0017] The time sequence streamline field components are cropped into a specified important region, and each frame image of the specified important region is divided into a uniform grid, while a spatial weighting matrix is used to highlight the specified important region, to obtain each frame of time sequence streamline field image after weighting processing;

[0018] Each frame of time sequence streamline field image after weighting processing is input into a pre-trained visual ViT model for encoding to extract a plurality of high-dimensional embedding vectors.

[0019] In an optional implementation, each high-dimensional embedding vector is expressed by a formula as follows:

[0020] ;

[0021] Wherein, represents a pressure field high-dimensional embedding vector of the i-th combustion simulation example at time t, represents a specified important region of the i-th combustion simulation example at time t, represents a spatial weighting matrix, represents a selected ViT encoder.

[0022] The visual analysis method for combustion simulation provided by the application can crop time sequence streamline field components into a specified important region, divide them into a uniform grid, and use a spatial weighting matrix to highlight the region, so as to accurately focus on the key part of the expert's attention, filter irrelevant background information, make the preprocessed image more targeted, reduce the influence of redundant data on the encoding result, and improve the effectiveness of the data. Each frame of time sequence streamline field image after preprocessing is input into a pre-trained visual ViT model for encoding, and with the powerful feature extraction capability of the ViT model, high-dimensional embedding vectors representing the essential features of the streamline field can be efficiently extracted from the images after optimization. These high-dimensional embeddings not only retain the key information of the streamline field, but also convert the image form of the streamline field data into a vector form that is more suitable for subsequent calculation and analysis, laying a solid foundation for subsequent joint dimensionality reduction and clustering, time sequence trajectory construction, etc.

[0023] In an optional implementation, after dimensionality reduction and clustering of the plurality of high-dimensional embedding vectors, a scatter plot after dimensionality reduction is generated, and a plurality of time sequence trajectories of each combustion simulation example are generated based on the scatter plot after dimensionality reduction, including:

[0024] All high-dimensional embedding vectors corresponding to the time sequence streamline field components are dimensionally reduced by using a preset dimensionality reduction algorithm, and all high-dimensional embedding vectors after dimensionality reduction are clustered by using a preset clustering algorithm to obtain a plurality of dimensionally reduced embedding vectors;

[0025] Map the plurality of dimension-reduced embedding vectors to a preset low-dimensional space to obtain corresponding coordinate points, and draw each coordinate point on a plane according to the position of each coordinate point in the low-dimensional space to generate a scatter plot after dimension reduction.

[0026] Connect the scatter plot after dimension reduction into a time sequence path in time sequence to generate a plurality of time sequence trajectories of each combustion simulation example.

[0027] The visual analysis method for combustion simulation provided by the application adopts a preset dimension reduction algorithm to reduce the dimension of high-dimensional embedding vectors, which can greatly reduce the data dimension while retaining the core features and simplify the data structure; then, a preset clustering algorithm is used to cluster the embedding vectors after dimension reduction, which can classify flow field states with similar features into one category, making the originally chaotic high-dimensional data ordered, facilitating subsequent analysis of the characteristics of different categories of flow fields, and laying a foundation for mining potential combustion modes. Map the plurality of dimension-reduced embedding vectors to a preset low-dimensional space and draw a scatter plot to convert abstract high-dimensional data into intuitive two-dimensional or three-dimensional graphics. Each coordinate point corresponds to a flow field state, and the distribution of points can directly reflect the similarity and difference between different flow field states. The scatter plot after clustering can also clearly show the aggregation area of different combustion modes, allowing experts to quickly grasp the overall distribution rule of the data and reducing the threshold for understanding complex flow field data. Connect the coordinate points in the scatter plot into a time sequence trajectory in time sequence to completely present the evolution path of the flow field state with time in each combustion simulation example. By observing the trend, fluctuation amplitude and other characteristics of the trajectory, experts can intuitively identify the steady state and oscillation state of the combustion process, and by comparing the time sequence trajectories of different examples, the influence of the initial conditions on the combustion evolution can be further explored, providing a clear dynamic visualization basis for analyzing the convergence and stability of combustion, and significantly improving the efficiency and accuracy of the analysis of the dynamic characteristics of the combustion process.

[0028] In an optional embodiment, the convergence radius is set by the following method:

[0029] Calculate the average of the plurality of dimension-reduced embedding vectors of each combustion simulation example, and calculate the convergence radius of each combustion simulation example based on the plurality of dimension-reduced embedding vectors and the corresponding average.

[0030] The visual analysis method for combustion simulation provided by the application reflects the flow field state of each combustion simulation example at different time points by the dimension reduction embedding vector of the example, and the average number can represent the average level of the flow field state of the example. The convergence radius is calculated based on the dimension reduction embedding vector and the corresponding average number, so that the size of the convergence radius is closely related to the flow field state characteristics of the example itself, rather than using a unified fixed value. This means that different examples can obtain an adaptive convergence range according to their own flow field fluctuation, and for examples with stable flow field state and concentrated dimension reduction embedding vector, the convergence radius will be smaller, and for examples with large flow field fluctuation, the convergence radius will increase accordingly, thereby ensuring the rationality of the set convergence radius and the pertinence to each example.

[0031] In an optional embodiment, the similar time sequence track most similar to the specified time sequence track is retrieved within the preset convergence radius, comprising:

[0032] All time sequence tracks within the preset convergence radius are determined as a candidate set, and the difference degree between each time sequence track in the candidate set and the specified time sequence track is calculated based on a preset difference quantification formula;

[0033] The time sequence tracks in the candidate set are sorted in order of difference degree from small to large, and the first preset number of time sequence tracks with the smallest difference degree are selected as the similar time sequence tracks most similar to the specified time sequence track.

[0034] The visual analysis method for combustion simulation provided by the application determines all time sequence tracks within the preset convergence radius as a candidate set, and effectively reduces the search range with the screening effect of the convergence radius. This operation avoids indiscriminate search of all time sequence tracks and excludes those tracks with low correlation in combustion characteristics. The difference degree between each time sequence track in the candidate set and the specified time sequence track is calculated based on a preset difference quantification formula, which is specially designed for the characteristics of time sequence tracks of combustion simulation and can accurately capture the subtle differences in the combustion evolution process of different tracks. The candidate set tracks are sorted in order of difference degree from small to large, and the first preset number of tracks with the smallest difference degree are selected as the most similar tracks, ensuring the objectivity and pertinence of the selection result.

[0035] In an optional embodiment, the combustion convergence and stability are analyzed based on the specified time sequence track and the similar time sequence track, comprising:

[0036] The distribution of the end points of the specified time sequence track and the similar time sequence track on the scatter plot after dimension reduction is determined, and the combustion convergence is analyzed based on the distribution;

[0037] The fluctuation degree of the specified time sequence track and the similar time sequence track is obtained on the scatter plot after dimension reduction, and the stability of the combustion is analyzed based on the fluctuation degree.

[0038] The visual analysis method of combustion simulation provided by the present application can intuitively and accurately reflect the convergence characteristics of the combustion process by determining the distribution of the end points of the specified time sequence track and the similar time sequence track on the scatter plot. When the end points are concentrated in a smaller area, it indicates that under the corresponding initial conditions, the combustion process tends to similar stable states and has good convergence. If the end points are distributed dispersedly, it indicates that the convergence results of the combustion process in different examples are quite different. The fluctuation degree of the specified track and the similar track on the scatter plot is obtained and the stability is analyzed accordingly, which has strong pertinence and accuracy. Small track fluctuation degree means that the flow field state changes gently in the combustion process and the combustion is stable. Large track fluctuation indicates that the combustion has obvious oscillation and the stability is poor.

[0039] In an optional embodiment, the visual analysis method of combustion simulation further comprises:

[0040] visualizing the full cycle processing of each combustion simulation example;

[0041] embedding representative cluster centers based on expert field description annotated time sequence fluent images, and constructing a combustion mode knowledge base based on the representative cluster centers;

[0042] inputting the combustion mode knowledge base, the time sequence fluent image and the representative cluster centers into a preset time sequence language model to generate a descriptive explanation of the time sequence fluent image, and generating an interactive graphic summary report of each combustion simulation example based on the time sequence fluent image and the corresponding descriptive explanation.

[0043] The combustion simulation visual analysis method provided by the application can visually display the full cycle processing of each combustion simulation example, and can enable experts to completely and intuitively understand the overall situation of the entire combustion simulation process. From the initial state to the final state, the flow field changes and time sequence trajectory evolution of each stage are clearly presented, which helps experts grasp the overall context of the combustion process, find key nodes or abnormal changes that may exist in the full cycle, provide a more comprehensive perspective for in-depth analysis of the combustion law, and make up for the information omission that may be caused by only focusing on part of the stage. Representative cluster centers are embedded based on expert field description annotation time sequence flow field image coding, and a combustion mode knowledge base is constructed, which fully combines the professional experience of experts. The expert field description has strong professionalism and pertinence, and can accurately extract the core features of the combustion mode represented by each cluster center. The knowledge base constructed on this basis becomes an effective carrier for storing and inheriting the knowledge in the combustion field, which not only provides a reliable field background for subsequent image analysis, but also helps beginners quickly understand the characteristics of different combustion modes. The combustion mode knowledge base, time sequence flow field image and representative cluster center are input into the preset visual language model to generate a descriptive explanation of the time sequence flow field image, and an interactive graphic summary report of each combustion simulation example is generated by reasoning, realizing the conversion from data to knowledge. The interactive graphic summary report integrates various forms of information such as images and texts, and experts can flexibly interact, conveniently check, filter and compare the report contents of different examples. This not only provides a convenient way for the presentation and exchange of combustion simulation results, but also helps experts quickly summarize the combustion characteristics and rules of different examples, provides more intuitive and practical reference for the optimization of combustion simulation and the improvement of engine design, and further improves the practicality and intelligent level of the entire visual analysis system.

[0044] In a second aspect, the application provides a combustion simulation visual analysis device, which comprises:

[0045] A time sequence flow image generation module is configured to generate a plurality of groups of time sequence flow images of each combustion simulation example under different initial conditions of combustion simulation.

[0046] A high-dimensional embedding vector generation module is configured to encode each frame of image in each group of time sequence flow images to obtain a plurality of high-dimensional embedding vectors corresponding thereto.

[0047] A time sequence trajectory generation module is configured to generate a scatter plot after dimension reduction of the plurality of high-dimensional embedding vectors, and generate a plurality of time sequence trajectories of each combustion simulation example based on the scatter plot after dimension reduction.

[0048] A similar time sequence trajectory retrieval and combustion performance analysis module is configured to retrieve the most similar time sequence trajectory to a specified time sequence trajectory within a preset convergence radius, and analyze the combustion convergence and stability based on the specified time sequence trajectory and the similar time sequence trajectory.

[0049] In a third aspect, the present application provides a computer device, comprising a memory and a processor, the memory and the processor being communicatively connected with each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the visual analysis method of combustion simulation according to the first aspect or any of the corresponding embodiments thereof.

[0050] In a fourth aspect, the present application provides a computer readable storage medium, storing computer instructions for causing a computer to perform the visual analysis method of combustion simulation according to the first aspect or any of the corresponding embodiments thereof.

[0051] In a fifth aspect, the present application provides a computer program product, comprising computer instructions for causing a computer to perform the visual analysis method of combustion simulation according to the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 is a flowchart of the visual analysis method of combustion simulation according to an embodiment of the present application;

[0054] Figure 2 is a flowchart of another visual analysis method of combustion simulation according to an embodiment of the present application;

[0055] Figure 3 is a flowchart of still another visual analysis method of combustion simulation according to an embodiment of the present application;

[0056] Figure 4 is a flowchart of yet another visual analysis method of combustion simulation according to an embodiment of the present application;

[0057] Fig. 5(a) is a schematic diagram of a filter panel of a system display interface in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0058] Fig. 5(b) is a time sequence track view of a system display interface in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0059] Fig. 5(c) is a similar track view of a system display interface in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0060] Figure 5 (d) is a report view of the system display interface in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0061] Figure 5 (e) is a detail view of the system display interface in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0062] Figure 6 (a1) is a schematic diagram of clustering results in a time sequence trajectory view in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0063] Figure 6 (a2) is a schematic diagram of a flow field frame corresponding to Figure 6 (a1) according to an embodiment of the present application;

[0064] Figure 6 (b1) is a schematic diagram of a first most similar time sequence trajectory calculated in a similar trajectory view in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0065] Figure 6 (b2) is a schematic diagram of a flow field frame corresponding to Figure 6 (b1) according to an embodiment of the present application;

[0066] Figure 6 (c1) is a schematic diagram of a second most similar time sequence trajectory calculated in a similar trajectory view in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0067] Figure 6 (c2) is a schematic diagram of a flow field frame corresponding to Figure 6 (c1) according to an embodiment of the present application;

[0068] Figure 6 (d) is a schematic diagram of the geometry of a ramjet engine used in a combustion simulation example in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0069] Figure 7 is a schematic diagram of a pressure field of stable combustion in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0070] Figure 8 is a schematic diagram of an interactive text and picture summary report of a pressure field of a combustion simulation example in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0071] Figure 9 is a schematic diagram of OH field analysis in the visual analysis method of combustion simulation according to an embodiment of the present application;

[0072] Figure 10 is a structural block diagram of a visual analysis device of combustion simulation according to an embodiment of the present application;

[0073] Figure 11 is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0074] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0075] The development of deep learning, especially Vision Transformer (ViT), has demonstrated strong capabilities in extracting semantic features from scientific images. ViT has achieved leading performance in standard vision tasks by treating images as sequences of image patches. The application of ViT in fluid dynamics tasks, such as partial differential equation operator learning and particle image velocimetry estimation, further highlights its potential for modeling complex flow structures.

[0076] To make high-dimensional embedding spaces more understandable to humans, dimensionality reduction methods such as t-SNE (t-distributed Stochastic Neighbor Embedding) and UMAP (Uniform Manifold Approximation and Projection) have become indispensable. Combined with density-based clustering algorithms such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), these tools can reveal consistent flow patterns and transient phenomena from large-scale simulation data. Interactive visual analysis frameworks further support the association between clustering results and high-dimensional original data, provide multi-view exploration, and enable tracking of dynamic changes in the reduced dimension space.

[0077] Despite significant progress in related methods, the integration of expert knowledge and automatic semantic interpretation in the field of combustion remains underdeveloped. In recent years, Vision Language Models (VLMs) have shown promising prospects in text generation from multi-modal data, but their application in scientific visualization, especially in contexts requiring specific terminology and physical knowledge, remains to be explored.

[0078] Understanding the complex combustion dynamics in ramjet engines is crucial for efficient engine propulsion design. However, the large amount of high-dimensional time-series flow field data generated by numerical simulations poses significant challenges for feature identification and comparative analysis.

[0079] The embodiment of the present application provides a visual analysis method for combustion simulation, which is based on hundreds of combustion simulation examples of ramjet engines under different initial conditions, and each example generates a series of time sequence flow field images. High-dimensional embedding vectors of each frame of image are extracted by a plurality of pre-trained ViT models, and then dimension reduction and clustering are performed to identify different combustion modes, and time sequence trajectories of each example are constructed in the embedding space to help expert users understand and track the evolution trend of the combustion process and further judge whether the ramjet engine tends to be stable. The combustion law is deeply revealed, a more reliable basis for understanding the combustion mechanism is provided, and the effect of improving the simulation precision of ramjet combustion is achieved.

[0080] According to the embodiment of the present application, a visual analysis method for combustion simulation is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0081] In the present embodiment, a visual analysis method for combustion simulation is provided, which can be used in the computer device described above, Figure 1 is a flowchart of the visual analysis method for combustion simulation according to the embodiment of the present application, as Figure 1 shown, the flow includes the following steps:

[0082] Step S101, under different initial conditions of combustion simulation, a plurality of time sequence flow images of each combustion simulation example are generated.

[0083] Specifically, taking the combustion simulation of a ramjet engine as an example, each time the ramjet engine combustion simulation has a set of initial conditions, and different values of static pressure, static temperature and water vapor concentration are combined to form different sets of initial conditions.

[0084] Each combustion simulation is taken as a combustion simulation example, and under different sets of initial conditions, each example generates hundreds of sets of combustion flow field time slices during combustion simulation to reflect the combustion process changing with time, and the hundreds of sets of combustion flow field time slices are taken as a plurality of time sequence flow images of each combustion simulation example.

[0085] Step S102, each frame of image in each set of time sequence flow images is encoded to obtain a plurality of corresponding high-dimensional embedding vectors.

[0086] Specifically, for each time sequence flow field component (such as pressure field ) in the time sequence flow field image, a key region (such as the flame core region of the combustion chamber and the region near the nozzle) is cut according to expert experience to obtain a cut time sequence flow field component ; adjust each frame image in the cropped time-series flow field component to a fixed length and width (e.g., 224x224 pixels), and divide it into uniform grids (e.g., 16x16 grids).

[0087] construct a spatial weighting matrix W P (e.g., the core region weight is 1.0, and the edge region weight is 0.3), and perform weighted processing on the cropped flow field component through matrix multiplication, highlighting important regions in the flow field component. Each frame of the flow field image after weighted processing is input into the pre-trained Vision Transformer model, and the local and global features of the flow field are captured through the self-attention mechanism of the model to output a high-dimensional embedding vector (e.g., 768 dimensions).

[0088] In step S103, a scatter plot after dimensionality reduction is generated after dimensionality reduction clustering of a plurality of high-dimensional embedding vectors, and a plurality of time-series trajectories of each combustion simulation example are generated based on the scatter plot after dimensionality reduction.

[0089] Specifically, the high-dimensional embedding vectors of all combustion simulation examples are collected.

[0090] Taking the pressure field as an example, it is denoted as: .

[0091] The plurality of high-dimensional embedding vectors are dimensionally reduced by a pre-set dimensionality reduction algorithm (e.g., PCA dimensionality reduction), and are clustered by a pre-set clustering algorithm (e.g., KMeans or DBSCAN clustering algorithm), and similar flow field component states are classified into the same class. The clustered dimensionally reduced embedding is further mapped to a 2D space to obtain low-dimensional coordinates ; all coordinate points are plotted on the plane to form a scatter plot after dimensionality reduction, wherein points of different clusters are marked with different colors (e.g., cluster 1 is green, cluster 2 is purple, etc.).

[0092] For each combustion simulation example i, the low-dimensional coordinates of all time steps are connected by line segments in chronological order to form a time-series trajectory.

[0093] In step S104, the most similar similar time-series trajectory to the specified time-series trajectory is searched within a pre-set convergence radius, and the combustion convergence and stability are analyzed based on the specified time-series trajectory and the similar time-series trajectory.

[0094] Specifically, the pre-set convergence radius is set based on the flow field characteristics of each combustion simulation example, which is not specifically limited herein. The most similar similar time-series trajectory to the specified time-series trajectory is searched within the pre-set convergence radius based on a pre-set trajectory difference quantification formula.

[0095] The combustion convergence and stability are analyzed based on the specified time-series trajectory and the similar time-series trajectory, exemplarily:

[0096] Convergence analysis: Observe the distribution of the end points of the specified trajectory and similar trajectories in the scatter plot: if all the end points are densely clustered in a small area centered on pi (and all within the convergence radius), it indicates that the combustion process has good convergence under this class of initial conditions; if the end points are scattered or beyond the radius range, it indicates that the convergence is poor. Compare the convergence speed of the trajectory: calculate the time step from the starting point to the entry into the convergence radius range, the shorter the time, the faster the convergence speed.

[0097] Stability analysis: Calculate the ratio of the total length (cumulative distance of line segments) of the specified trajectory and similar trajectories to the time step, the smaller the ratio, the smoother the trajectory, the higher the stability; if the trajectory appears a large shock (such as the length of a line segment suddenly increases by 3 times), it indicates that the combustion is transiently unstable.

[0098] The visual analysis method of combustion simulation provided in this embodiment generates multiple sets of time series flow field images under different initial conditions, providing a rich data basis for comprehensively exploring the combustion simulation rules. The images are encoded to obtain high-dimensional embedding vectors, and then dimensionality reduction clustering is performed to generate scatter plots and time series trajectories, which converts complex combustion flow field data into intuitive visual forms, greatly reduces the complexity of the data, enables experts to quickly grasp the overall characteristics and distribution rules of the data, and significantly improves the efficiency of data analysis. In terms of accuracy and depth of analysis, time series trajectories are generated based on the scatter plot after dimensionality reduction, and the most similar trajectories are retrieved within the preset convergence radius, and the convergence and stability of combustion are analyzed by combining the specified trajectory and similar trajectories. This way makes full use of the relevance and similarity between data, and can accurately identify similar combustion evolution patterns, helping experts more accurately grasp key characteristics such as steady state and oscillation in the combustion process, and deeply reveal the combustion rules, providing a more reliable basis for understanding the combustion mechanism, improving the accuracy of ramjet combustion simulation, and solving the problem of low accuracy of ramjet combustion simulation in the prior art.

[0099] In this embodiment, a visual analysis method of combustion simulation is provided, which can be used in the above-mentioned computer device, Figure 2 is a flowchart of the visual analysis method of combustion simulation according to an embodiment of the present application, as Figure 2 shown, the flowchart includes the following steps:

[0100] Step S201, under different initial conditions of combustion simulation, generate multiple sets of time series flow field images for each combustion simulation example.

[0101] Specifically, taking the combustion of a ramjet engine as an example, the initial conditions include static pressure, static temperature and water vapor concentration; the above step S201 includes:

[0102] Step S2011: Combine different values ​​of static pressure, static temperature and water vapor concentration, and simulate the combustion process under each combination to generate multiple sets of combustion flow field time slices.

[0103] Specifically, such as Figure 4 As shown, the initial conditions include: static pressure p, temperature (also known as static temperature) T, and water vapor concentration w. Multiple discrete values ​​are set for each parameter (e.g., static pressure is p1, p2, ..., pm, static temperature is T1, T2, ..., Tn, and water vapor concentration is w1, w2, ..., wk).

[0104] Different values ​​of static pressure, static temperature, and water vapor concentration are combined either fully or orthogonally to form various initial condition combinations. For each initial condition combination, a ramjet engine combustion simulation program is run to simulate the complete cycle of the combustion process from start-up to stabilization (or termination), generating multiple combustion flow field time slice sequences containing flow field components such as pressure field, OH (Hydroxyl Radical, OH, hydroxyl radical) free radical field, Mach number field, and velocity field.

[0105] Step S2012: Render different time-series flow field components in each group of combustion flow field time slices using the same color and scaling ratio, remove irrelevant background information, retain important regions of different time-series flow field components, and obtain multiple time-series flow field images corresponding to each group of initial conditions; different time-series flow field components include pressure, OH radicals, Mach number and velocity.

[0106] Specifically, the simulated flow field time-slice sequence is standardized, and each flow field component (including pressure field, OH field, etc.) is rendered using a uniform color mapping (e.g., red for high-temperature regions and blue for low-temperature regions) and scaling. Irrelevant background information is removed, and the core flow field region is retained. This results in multiple sets of time-series flow field images corresponding to each set of initial conditions, denoted as [images of the time-series flow field]. ,in This represents the pressure field of the i-th combustion simulation case at time t. Let represent the OH field of the i-th combustion simulation case at time t.

[0107] The visual analysis method for combustion simulation provided in this embodiment uses static pressure, static temperature, and water vapor concentration as key initial parameters affecting the combustion process of a ramjet engine. Different combinations of these initial parameters can cover various combustion conditions. By generating time-series flow field images under each combination, the changing characteristics of the combustion flow field over time under different conditions can be comprehensively captured. This avoids missing key combustion phenomena due to incomplete initial conditions, providing rich and representative basic data for subsequent analysis and ensuring that the analysis results reflect various actual combustion scenarios. Using the same color mapping and scaling ratio to render different time-series flow field components such as pressure, OH radicals, Mach number, and velocity ensures consistency in visual presentation of different components, eliminates interference caused by differences in rendering methods, and allows subsequent encoding to more accurately focus on the characteristic differences of the flow field itself.

[0108] Step S202: Encode each frame of each group of time-series smooth images to obtain multiple corresponding high-dimensional embedding vectors.

[0109] Specifically, step S202 includes:

[0110] Step S2021: The different temporal flow field components are cropped into specified important regions, and each frame of the specified important region is divided into a uniform grid. At the same time, a spatial weighting matrix is ​​used to highlight the specified important regions, resulting in a weighted temporal flow field image for each frame.

[0111] Specifically, such as Figure 4 As shown, for each flow field component (such as pressure field) in the time-series flow field image... Based on expert experience, key areas (such as the core area of ​​the combustion chamber flame and the area near the nozzle) are selected to obtain the tailored flow field. The cropped flow field frame is adjusted to a fixed length and width (e.g., 224×224 pixels) and divided into a uniform grid (e.g., 16×16 grid).

[0112] Constructing a spatial weighted matrix W p (For example, the weight of the core region is 1.0 and the weight of the edge region is 0.3). The cropped flow field image is weighted by matrix multiplication to highlight the important regions in the flow field components (such as regions with drastic changes in pressure gradient and regions with high concentration of OH radicals), and the weighted flow field image of each frame is obtained.

[0113] Step S2022: Input each frame of the time-series flow field image after weighted processing into the pre-trained visual ViT model for encoding and extract multiple high-dimensional embedding vectors.

[0114] Specifically, each frame of the flow field image after weighted processing is input into a pre-trained Vision Transformer model (ViT model). The model's self-attention mechanism captures the local and global features of the flow field and outputs a high-dimensional embedding vector. (e.g., 768 dimensions).

[0115] Each high-dimensional embedding vector is expressed by the formula:

[0116] (1);

[0117] in, This represents the high-dimensional embedding vector of the pressure field at time t for the i-th combustion simulation example. This represents a specified important region for the i-th combustion simulation case at time t. Represents a spatial weighted matrix. This indicates the selected ViT encoder.

[0118] Step S203 involves performing dimensionality reduction and clustering on multiple high-dimensional embedding vectors to generate a dimensionality-reduced scatter plot, and then generating multiple time-series trajectories for each combustion simulation example based on the dimensionality-reduced scatter plot. For details, please refer to [link to details]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0119] Step S204: Within a preset convergence radius, retrieve the most similar time-series trajectory to the specified time-series trajectory, and analyze combustion convergence and stability based on the specified time-series trajectory and the similar time-series trajectory. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0120] The visual analysis method for combustion simulation provided in this embodiment clips different temporal flow field components into designated important regions, divides them into uniform grids, and uses a spatial weighting matrix to highlight these regions. This accurately focuses on key parts of interest to experts, filters out irrelevant background information, makes the preprocessed images more targeted, reduces the impact of redundant data on the encoding results, and improves data effectiveness. Each frame of the preprocessed temporal flow field image is input into a pre-trained visual ViT model for encoding. Leveraging the powerful feature extraction capabilities of the ViT model, high-dimensional embedding vectors representing the essential features of the flow field can be efficiently extracted from the optimized images. These high-dimensional embeddings not only preserve the key information of the flow field but also transform the image-based flow field data into a vector form more suitable for subsequent calculations and analysis, laying a solid foundation for subsequent joint dimensionality reduction and clustering, temporal trajectory construction, and other operations.

[0121] This embodiment provides a visual analysis method for combustion simulation, which can be used with the aforementioned computer equipment. Figure 3This is a flowchart of a visual analysis method for combustion simulation according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0122] Step S301: Under different initial conditions of the combustion simulation, generate multiple sets of time-series smooth images for each combustion simulation case. See details below. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0123] Step S302: Encode each frame of each group of temporally smooth images to obtain multiple corresponding high-dimensional embedding vectors. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0124] Step S303: After performing dimensionality reduction and clustering on multiple high-dimensional embedding vectors, a dimensionality-reduced scatter plot is generated, and multiple time-series trajectories for each combustion simulation case are generated based on the dimensionality-reduced scatter plot.

[0125] Specifically, step S303 includes:

[0126] Step S3031: All high-dimensional embedding vectors corresponding to the time-series flow field components are reduced in dimensionality using a preset dimensionality reduction algorithm, and all high-dimensional embedding vectors after dimensionality reduction are clustered using a preset clustering algorithm to obtain multiple dimensionality-reduced embedding vectors.

[0127] Specifically, the preset dimensionality reduction algorithm adopts PCA (Principal Component Analysis). PCA is a dimensionality reduction algorithm that projects high-dimensional data into a low-dimensional space through linear transformation while preserving the main features of the data. Its core is to find the principal components of the data (i.e., the directions with the most information) through eigenvalue decomposition, thereby reducing the feature dimensionality and removing redundant information.

[0128] The preset clustering algorithm uses either K-means or DBSCAN. K-means clustering is a partition-based unsupervised learning algorithm that iteratively divides the dataset into K clusters, ensuring that data points within the same cluster are similar and data points in different clusters are significantly different. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm that can automatically identify dense regions in the data and divide them into clusters, while also being robust to noisy data.

[0129] Collect the high-dimensional embedding vectors of all combustion simulation examples, denoted as follows: (For example, for the pressure field:) , This represents the stopping time of the i-th combustion simulation case.

[0130] The PCA algorithm is used for dimensionality reduction (e.g., to 10 dimensions) to retain the main variance of each time-series flow field component (e.g., cumulative variance contribution rate ≥ 95%). KMeans clustering is used on the dimensionality-reduced embeddings (the number of clusters K is adjusted by experts according to the number of combustion modes, e.g., K=5) to group similar flow field states into the same class, resulting in multiple dimensionality-reduced embedding vectors.

[0131] Step S3032: Map multiple dimension-reduced embedding vectors to a preset low-dimensional space to obtain corresponding coordinate points, and draw them on a plane according to the position of each coordinate point in the low-dimensional space to generate a dimension-reduced scatter plot.

[0132] Specifically, the clustered dimensionality-reduced embeddings are further mapped to 2D space using t-SNE or UMAP to obtain low-dimensional coordinates. Plot all coordinate points on the plane to form a scatter plot after dimensionality reduction, with points of different clusters marked with different colors (e.g., green for cluster 1, purple for cluster 2, etc.).

[0133] Step S3033: Connect the dimensionality-reduced scatter plots into a time-series path according to the time sequence to generate multiple time-series trajectories for each combustion simulation example.

[0134] Specifically, for each combustion simulation example i, the low-dimensional coordinates of all its time steps are... Connect the time segments with line segments to form a time-series trajectory. The slope of the time-series trajectory represents the rate of change of the flow field state, and the density of the time-series trajectory reflects the stability of the combustion state (e.g., a densely intertwined trajectory indicates stable combustion).

[0135] Step S304: Search for the most similar time-series trajectory within the preset convergence radius, and analyze the combustion convergence and stability based on the specified time-series trajectory and the similar time-series trajectory.

[0136] Specifically, step S304 includes:

[0137] Step S3041, the convergence radius is set in the following way: calculate the average of multiple dimensionality-reduced embedding vectors for each combustion simulation, and calculate the convergence radius of each combustion simulation case based on the multiple dimensionality-reduced embedding vectors and the corresponding average.

[0138] Specifically, based on practical experience, the time trajectory of combustion simulation examples that achieve steady-state combustion will oscillate and converge within a small region of the latent space. The convergence radius of the i-th combustion simulation example is defined as:

[0139] (2);

[0140] (3);

[0141] in, Let represent the convergence radius of the i-th combustion simulation case, and k represent the number of similar time-series trajectories within the preset radius. This represents the average of multiple dimension-reduced embedding vectors. This represents the stopping time of the i-th combustion simulation case.

[0142] Step S3042: Determine all time-series trajectories within the preset convergence radius as a candidate set, and calculate the difference between each time-series trajectory in the candidate set and the specified time-series trajectory based on the preset difference quantification formula.

[0143] Specifically, from the time-series trajectories generated by all combustion simulation examples, one trajectory requiring focused analysis is selected as the designated time-series trajectory. This trajectory is formed by connecting the dimensionality-reduced embedding representations of the corresponding examples in chronological order, reflecting the dynamic evolution of the combustion process in that example.

[0144] The spatial range for retrieval is defined based on a preset convergence radius. Only other time-series trajectories that fall within the preset convergence radius of the corresponding computational instance of the specified time-series trajectory are included in the candidate trajectory set.

[0145] For each time-series trajectory in the candidate trajectory set, a pre-defined difference quantification formula is used for calculation. This formula focuses on the differences between the trajectories as they approach the end of the simulation. By calculating the distance between the dimensionality-reduced embedded representations at the corresponding time points and combining the change magnitudes of adjacent time points of the trajectory, the difference degree value between each candidate trajectory and the specified time-series trajectory is obtained. The smaller the difference degree, the higher the similarity between the two trajectories.

[0146] Step S3043: Sort the time series trajectories in the candidate set in ascending order of difference, and select the time series trajectory with the smallest difference among the first preset number of time series trajectories as the most similar time series trajectory to the specified time series trajectory.

[0147] Specifically, the candidate trajectories are sorted in ascending order of their difference, and the top k trajectories with the smallest difference (k is a preset number that can be adjusted by experts according to analysis needs) are selected as the most similar time series trajectories to the specified time series trajectory.

[0148] To identify similar combustion simulation patterns, the time-series trajectories of the first k and the i-th combustion simulation examples are retrieved. The most similar time series trajectories are denoted as follows: Combustion simulation examples Correspondingly, to quantify the difference between the time trajectories of combustion simulation case i and combustion simulation case j, multiple formulas were repeatedly explored and designed in collaboration with domain experts. Through a series of evaluations, the degree of difference between the time trajectories of combustion simulation case i and combustion simulation case j was obtained, and its formula is defined as:

[0149] (4);

[0150] in, and These represent the termination times of combustion simulation examples i and j, respectively. The difference index of the time-series trajectories highlights the differences in the embedded time-series trajectories as the simulation approaches termination, which is crucial for understanding the convergence and stability characteristics of the combustion process.

[0151] The visual analysis method for combustion simulation provided in this embodiment determines all time-series trajectories within a preset convergence radius as a candidate set, effectively narrowing the search scope by leveraging the filtering effect of the convergence radius. This operation avoids indiscriminate searching of all time-series trajectories, excluding those trajectories with low correlation to the specified trajectory in terms of combustion characteristics. Based on a preset difference quantification formula, the difference degree between each time-series trajectory in the candidate set and the specified time-series trajectory is calculated. This formula is specifically designed for the characteristics of time-series trajectories in combustion simulation, accurately capturing subtle differences between different trajectories during combustion evolution. The candidate set trajectories are sorted in ascending order of difference degree, and the trajectory with the smallest difference degree among the first preset number is selected as the most similar trajectory, ensuring the objectivity and specificity of the selection results.

[0152] Step S3044: Determine the distribution of the endpoints of the specified time-series trajectory and similar time-series trajectories on the dimensionality-reduced scatter plot, and analyze the combustion convergence based on the distribution.

[0153] Specifically, observe the distribution of trajectory endpoints: On the scatter plot after dimensionality reduction, examine the endpoint positions of the specified time-series trajectory and similar time-series trajectories. If these endpoints are concentrated in a small region, and this region is within the preset convergence radius, it indicates that the combustion process of these examples eventually tends to a similar stable state, and the combustion convergence is good; if the endpoints are scattered and exceed the preset convergence radius, it indicates that the combustion convergence is poor.

[0154] Comparing trajectory convergence speeds: Analyze the evolution of a specified trajectory and similar trajectories from the starting point to the ending point, and calculate the time it takes for the trajectory to enter the preset convergence radius and stabilize. The shorter the time, the faster the combustion convergence speed; conversely, the longer the time, the slower the convergence speed. By comparing the convergence speeds of different trajectories, the influence of initial conditions on the combustion convergence speed can be explored.

[0155] Step S3045: On the dimensionality-reduced scatter plot, obtain the fluctuation degree of the specified time-series trajectory and similar time-series trajectories, and analyze the stability of combustion based on the fluctuation degree.

[0156] Specifically, examine the trajectory oscillation: On the dimensionality-reduced scatter plot, observe the degree of fluctuation between the specified time-series trajectory and similar time-series trajectories. If the trajectory exhibits small fluctuations within the convergence region, meaning the distance between points is relatively short and there are no obvious large oscillations, it indicates that the combustion process is relatively stable. If the trajectory fluctuates significantly back and forth within the convergence region, or even exceeds the preset convergence radius and then returns, it indicates that the combustion exhibits oscillation phenomena and has poor stability.

[0157] Correlation between trajectory similarity and stability: If the specified time-series trajectory is highly similar to similar time-series trajectories and they both have good stability characteristics (such as small fluctuations), it indicates that the stability of the combustion process is consistent under such initial conditions; if some trajectories among the similar trajectories have good stability and some have poor stability, the differences in the initial conditions corresponding to these trajectories can be further analyzed to explore the key factors affecting combustion stability.

[0158] Judgment based on changes in the convergence radius: For a given trajectory and similar trajectories, calculate the convergence radius at different time points. If the convergence radius gradually decreases and stabilizes at a small value over time, it indicates that the stability of the combustion process is continuously improving; if the convergence radius fluctuates wildly without a stable trend, it indicates poor combustion stability.

[0159] The visual analysis method for combustion simulation provided in this embodiment can intuitively and accurately reflect the convergence characteristics of the combustion process by determining the distribution of the endpoints of a specified time-series trajectory and similar time-series trajectories on a scatter plot. When these endpoints are concentrated in a small area, it indicates that under the corresponding initial conditions, the combustion process eventually tends to a similar stable state, with good convergence; if the endpoints are scattered, it indicates that the convergence results of the combustion process vary greatly in different examples. Obtaining the fluctuation degree of the specified trajectory and similar trajectories on the scatter plot and analyzing the stability accordingly is highly targeted and accurate. Small trajectory fluctuations mean that the flow field state changes gently during combustion, and the combustion is relatively stable; while large trajectory fluctuations indicate that there are obvious oscillations in the combustion, and the stability is poor.

[0160] Step S305: Visualize the full-cycle processing of each combustion simulation case; embed representative cluster centers based on expert domain description annotation of the time-series smooth image encoding, and construct a combustion mode knowledge base based on the representative cluster centers; input the combustion mode knowledge base, time-series smooth images, and representative cluster centers into a preset time-series language model to generate descriptive explanations of the time-series smooth images, and generate an interactive graphic summary report for each combustion simulation case based on the time-series smooth images and corresponding descriptive explanations.

[0161] Specifically, to enhance interpretability, experts were invited to annotate representative cluster centers of the flow field embeddings using domain descriptions, thereby constructing a combustion pattern knowledge base. The cluster centers are composed of… (Clustering with label k) and The point closest to the mean is calculated, that is:

[0162] (5);

[0163] in, This represents the high-dimensional embedding vector of the pressure field at time t for the i-th combustion simulation example. This represents the k-th cluster. Represents the k-th cluster The number of high-dimensional embedding vectors contained in This indicates a set of k clusters. All high-dimensional embedding vectors Sum.

[0164] In this visual analysis system for ramjet engine combustion simulation, the latent space refers to the low-dimensional feature space obtained by dimensionality reduction processing of the high-dimensional flow field embedding.

[0165] This embodiment designs a system display interface for visually displaying the entire lifecycle processing of each combustion simulation example. The system display interface includes:

[0166] 1. Filtering Panel:

[0167] As shown in Figure 5(a), the filter panel is used to flexibly filter and organize simulation data based on initial conditions (parameters) and clustering parameters. In the "Initial Condition Filter" area (a1 in Figure 5(a)), users can limit the initial pressure P, temperature T, and water vapor concentration H2O of the simulation dataset using sliders and input boxes, in order to focus on specific combustion states (such as high-pressure lean combustion) and reduce interference in subsequent visualizations.

[0168] Below the filter, the "Selected Case Table" (a2 in Figure 5(a)) displays the simulated cases and their initial parameters for the current filtering results, facilitating user comparison. Users can select multiple cases by checking the checkboxes in the table, and all subsequent visualizations will be updated after clicking the "Plot Scatter Plot" button. Users can also sort the cases by clicking the table header.

[0169] In the “Clustering Parameters” area (a3 in Figure 5(a)), users can set parameters of the DBSCAN clustering algorithm, such as the minimum number of samples (minSamples) and the neighborhood radius (eps), and select flow field variables (such as pressure, OH concentration or Mach number) for feature embedding, thereby exploring the impact of different physical perspectives on the clustering results.

[0170] 2. Temporal Trajectory View:

[0171] The time-series trajectory view (Figure 5(b)) shows the distribution of high-dimensional image embeddings in two-dimensional space after dimensionality reduction processing (such as UMAP or t-SNE). Each point represents the embedding result of a frame of combustion simulation image, and the color indicates the cluster to which it belongs. When a user clicks on a point in the main scatter plot, the system automatically connects all frames belonging to the same calculation case to form a time-series trajectory, helping experts analyze the evolution of combustion dynamics in the potential space.

[0172] This view supports pattern recognition at both macroscopic and microscopic levels. At the macroscopic level, users can identify clusters representing different combustion modes; at the microscopic level, time-series trajectories reveal how the combustion process transitions between clusters, stabilizes within a particular cluster, or exhibits periodic or chaotic behavior. This information is crucial for assessing ignition success, flame stability, and steady-state transition.

[0173] Users can interactively select a single point or the entire trajectory to view the corresponding image and perform further analysis. Clicking the "Show / Hide Cluster Centers" button controls the display of cluster centers in the main scatter plot.

[0174] 3. Similar Trajectories View:

[0175] To support case-based reasoning, the similar trajectory view (Figure 5(c)) automatically calculates the trajectory of the currently selected case on the main scatter plot and displays the three trajectories that are most similar to it.

[0176] Each similar trajectory is presented as a mini scatter plot, with the target trajectory highlighted against a faded background, allowing experts to visually compare their shapes and evolutionary characteristics and identify similar combustion development paths.

[0177] This view helps experts understand the current simulation within a broader design space and also facilitates model validation by comparing it with known or labeled examples.

[0178] 4: Details View

[0179] The rightmost detail view (Figure 5(e)) is an image browser used to display the complete time-series flow field image sequence of the selected example. When the user clicks on a point in the main scatter plot or similar trajectory plot, this view displays the flow field image (such as pressure or temperature distribution) of the example at each time step. Clicking on any frame image will embed the corresponding image in the highlighted area of ​​the scatter plot, and display the corresponding description (AI-generated or expert-annotated) in the "Report View" below.

[0180] This module, through close integration of images and embeddings, allows users to correlate trajectories in the potential space with physical phenomena, verify the rationality of the embedded structure, discover anomalies, or locate key events (such as ignition or extinction). The image browser supports scrolling preview and hover zoom viewing.

[0181] 5. Report View:

[0182] The report view at the bottom of the interface (Figure 5(d)) displays a text description of the currently selected image, generated by AI or manually annotated. The system first retrieves the nearest cluster center to the target image in the latent space and uses its annotated image as context input to the Visual Language Model (VLM) to generate a natural language description for the current image. This method effectively narrows the semantic gap between low-level visual images and high-level expert knowledge. Based on expert annotations, the description generated by the system is closer to domain terminology and cognitive habits, improving the accuracy and readability of the explanation.

[0183] The visual analysis method for combustion simulation provided in this embodiment visualizes the entire lifecycle of each combustion simulation case, allowing experts to gain a complete and intuitive understanding of the entire combustion simulation process. From the initial state to the final state, the flow field changes and temporal trajectory evolution at each stage are clearly presented, helping experts grasp the overall context of the combustion process, identify potential key nodes or abnormal changes throughout the entire lifecycle, and provide a more comprehensive perspective for in-depth analysis of combustion laws, compensating for information omissions that may result from focusing only on certain stages. Based on expert domain descriptions and annotations of representative cluster centers embedded in the temporal flow field image encoding, a combustion mode knowledge base is constructed, fully integrating the expert's professional experience. The expert's domain descriptions are highly professional and targeted, accurately extracting the core features of the combustion modes represented by various cluster centers. The knowledge base built on this basis has become an effective carrier for storing and passing on knowledge in the field of combustion. It not only provides a reliable domain background for subsequent image analysis, but also helps novices quickly understand the characteristics of different combustion modes. By inputting the combustion mode knowledge base, time-series flow field images and representative cluster centers into the preset visual language model, descriptive explanations of time-series flow field images are generated, and interactive graphic summary reports of each combustion simulation case are generated through reasoning, realizing the transformation from data to knowledge.

[0184] As one or more specific application embodiments of the present invention, in conjunction with Figures 6(a1) to Figure 9 The visual analysis method for combustion simulation provided by this invention will be described in further detail below:

[0185] Over 200 ramjet engine combustion scenarios with different initial conditions were simulated using a dedicated CFD (Cadence Fidelity) program, generating approximately 3TB of data. Each simulation case contained multiple flow field frames generated at millisecond intervals until a stable combustion state was reached. The initial conditions for each case included static pressure (P in the filter panel, simulation range 0.8 MPa to 2.1 MPa), static temperature (range 565 K to 830 K), and inlet water vapor concentration (H2O, range 7.8% to 14%). Each generated flow field contained multiple components including pressure, OH radicals, Mach number, and velocity. This embodiment primarily focuses on the pressure field and the OH field.

[0186] Example 1: Pressure Field Analysis

[0187] Figures 6(a1) to 6(c2) illustrate the study of the combustion pressure field of the ramjet engine. The cropped pressure field images were processed by InternViT and clustered using the DBSCAN algorithm. Figure 6(a1) shows the clustering results in the time-series trajectory view; Figures 6(b1) and 6(c1) show the two most similar trajectories calculated in the similar trajectory view; Figures 6(a2), 6(b2), and 6(c2) are the flow field frames corresponding to the cases in Figures 6(a1), 6(b1), and 6(c1); Figure 6(d) shows the geometry of the ramjet engine used in the combustion simulation.

[0188] The pressure field image is cropped by region and assigned different weights to compare the clustering results with the time trajectory, focusing on the isolator and cavity regions where pressure changes are most significant. The cropped images are then processed by extracting high-dimensional embedding vectors using multiple pre-trained ViT models, and then input into the system for further processing.

[0189] After filtering a specific range of initial parameters in the filter panel, the red-blue boundary region in the flow field was observed to represent a pressure surge. The surge locations in Figures 6(a2), 6(b2), and 6(c2) all occur before the engine cavity, as shown in Figure 6(d), indicating that this state is a sub-combustion state. When the engine approaches stable combustion, the pressure distribution remains stable or experiences only slight fluctuations, as shown in the flow field frame 6 ms later. If the surge is located between the isolator and the cavity, it indicates the entry into the combustion mode. The first two frames of Figures 6(a2), 6(b2), and 6(c2) represent typical combustion, but it is not yet stable.

[0190] When the case studies tend towards steady-state combustion, their trajectories in the embedded space tend to stabilize and oscillate within a small region, as shown by the dashed circles in Figures 6(a1), 6(b1), and 6(c1). Furthermore, it can be observed that cases with similar time trajectories also exhibit highly similar pressure field evolution. The three combustion simulation examples in Figures 6 (a1, b1, and c1) all start from an unstable mode and rapidly transition to stable subsonic combustion.

[0191] like Figure 7 As shown in Figures 6(a1), 6(b1), and 6(c1), the three combustion simulation examples finally reach a stable combustion state. Their time trajectories are concentrated in a small cluster region in the embedding space, as shown in Figure 6(a1), 6(b1), and 6(c1). Figure 7 The enlarged view of (a1) is shown. Figure 7 The image (af) shows the flow field frame for case (a), where the pressure surge always occurs between the isolator and the cavity. Under the InternViT embedding, trajectories with stable combustion characteristics are always concentrated in areas such as... Figure 7 The small circular clusters shown in (a)-(d) exhibit only minor differences. Figure 7The irregular large clusters shown in (e) primarily contain frames in a stable sub-burning state. These trajectories typically begin with a small cluster and aggregate after stabilization. Figure 7 Within the area shown in (e).

[0192] In this embodiment, descriptions for cluster center frames were manually written in collaboration with experts. Clicking the "Save Description" button saves the description of a specific cluster center frame. Figure 8 The bottom left corner shows the automatic interpretation generated by the VLM model based on the nearest cluster center image and description of this flow field frame. Figure 8 The right side provides a summary description of the entire case study, also generated by integrating the descriptions of each frame using the VLM model. Users can click on any frame in the report view or the corresponding point in the time-series trajectory view to view the flow field frame and its description.

[0193] Example 2: OH Field Analysis

[0194] like Figure 9 As shown, combustion modes are identified through OH field time-series trajectory clustering. The trajectory of the selected case is displayed at (s1), and its two most similar trajectories are located at (s2) and (s3). The corresponding OH field images are shown at (f1)-(f3). Two main combustion modes are identified: the red cluster corresponds to shear layer combustion, characterized by a narrow and layered flame structure formed along the interface between the combustion gas and air (e.g., ...). Figure 9 (D)-(E) circled areas); the yellow clusters represent jet-wake combustion, characterized by a broad, curved flame trail extending from the nozzle (e.g., Figure 9 (Select the area in (a)-(b)).

[0195] This embodiment focuses on the OH field image reflecting combustion activity. The image is cropped to preserve the engine structure and the background area is removed. Based on domain knowledge, the area above the engine is given a higher weight (because the OH distribution changes significantly there), while the area on the left with smaller changes is given a lower weight. Subsequently, the pre-trained BLIP-2 ViT model (a vision-language pre-trained model; BLIP-2 uses a Vision Transformer (ViT) as an image encoder, responsible for decomposing the image into multiple time slices and generating corresponding visual feature vectors) is used to extract embedding vectors and capture the spatial features of the combustion process.

[0196] By adjusting the initial parameter range in the filter panel and DBSCAN's clustering parameters (such as neighborhood radius ep and minimum number of samples minSamples), more meaningful clustering results can be obtained in collaboration with experts.

[0197] like Figure 9As shown, the time trajectory of the selected case is at (s1), and its two most similar trajectories are located at (s2) and (s3), with the corresponding OH field frames shown as (f1)-(f3). The clustering results effectively distinguish different combustion modes. Among them, the red clusters represent shear layer combustion, where the flame distribution is along the fuel-air interface, forming a slender layered structure (see the circled areas in (d) and (e)). The yellow clusters represent wake combustion, where the flame is ejected from the fuel nozzle and extends downstream to form a wide wake (as shown in the diagram). Figure 9 (Selected areas in (a) and (b)).

[0198] It can also be observed that even when the pressure field has stabilized, the OH field trajectories may still fluctuate within the embedding space, but they always remain within the same cluster. This indicates that even under steady-state conditions, the spatial structure of the OH field may change significantly, but the overall combustion mode remains unchanged.

[0199] The visual analysis method for combustion simulation provided in this embodiment is based on hundreds of combustion simulation examples of ramjet engines under different initial conditions. Each set of examples generates a series of time-series flow field images. High-dimensional embedding representations of each frame are extracted using various pre-trained ViT (VisionTransformer) models, followed by dimensionality reduction and clustering to identify different combustion modes. A time-series trajectory for each example is constructed in the embedding space to help expert users understand and track the evolution trend of the combustion process, and thus determine whether the ramjet engine is approaching a steady state. Furthermore, the method invites domain experts to provide text annotations for the flow field images corresponding to a few representative cluster centers. These annotations serve as contextual input to the visual language model, further generating descriptive summaries for arbitrary flow field images. The system also supports natural language-based example retrieval and can generate AI-assisted graphic reports for single-frame images or complete simulation processes.

[0200] This embodiment also provides a visual analysis device for combustion simulation, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0201] This embodiment provides a visual analysis device for combustion simulation, such as... Figure 10 As shown, it includes:

[0202] The time-series smooth image generation module 1001 is used to generate multiple sets of time-series smooth images for each combustion simulation case under different initial conditions of combustion simulation.

[0203] The high-dimensional embedding vector generation module 1002 is used to encode each frame of each group of time-series smooth images to obtain multiple corresponding high-dimensional embedding vectors.

[0204] The time-series trajectory generation module 1003 is used to generate a dimensionality-reduced scatter plot after performing dimensionality reduction and clustering on multiple high-dimensional embedding vectors, and to generate multiple time-series trajectories for each combustion simulation example based on the dimensionality-reduced scatter plot.

[0205] The similar time-series trajectory retrieval and combustion performance analysis module 1004 is used to retrieve the most similar time-series trajectory to the specified time-series trajectory within a preset convergence radius, and to analyze combustion convergence and stability based on the specified time-series trajectory and the similar time-series trajectory.

[0206] In some optional implementations, the initial conditions include static pressure, static temperature, and water vapor concentration; the time-series smooth image generation module 1001 includes:

[0207] The combustion flow field time slice generation unit is used to combine different values ​​of static pressure, static temperature and water vapor concentration, and simulate the combustion process under each combination to generate multiple sets of combustion flow field time slices.

[0208] The time-series flow field image generation unit is used to render different time-series flow field components in each group of combustion flow field time slices using the same color and scaling ratio, remove irrelevant background information, retain important regions of different time-series flow field components, and obtain multiple sets of time-series flow field images corresponding to each set of initial conditions; different time-series flow field components include pressure, OH radicals, Mach number and velocity.

[0209] In some alternative implementations, the high-dimensional embedding vector generation module 1002 includes:

[0210] The weighted processing unit is used to crop different temporal flow field components into specified important regions, divide each frame of the specified important region into a uniform grid, and use a spatial weighting matrix to highlight the specified important regions, thus obtaining each frame of the temporal flow field image after weighted processing.

[0211] The high-dimensional embedding vector extraction unit is used to input the weighted temporal flow field image of each frame into a pre-trained visual ViT model for encoding, and extract multiple high-dimensional embedding vectors. Each high-dimensional embedding vector is expressed by the formula:

[0212] ;

[0213] in, This represents the high-dimensional embedding vector of the pressure field at time t for the i-th combustion simulation example. This represents a specified important region for the i-th combustion simulation case at time t. Represents a spatial weighted matrix. This indicates the selected ViT encoder.

[0214] In some optional implementations, the time-series trajectory generation module 1003 includes:

[0215] The dimension reduction clustering unit is used to reduce the dimension of all high-dimensional embedding vectors corresponding to the time-series flow field components using a preset dimension reduction algorithm, and then use a preset clustering algorithm to cluster all the high-dimensional embedding vectors after dimension reduction to obtain multiple dimension-reduced embedding vectors.

[0216] The scatter plot generation unit is used to map multiple dimensionality-reduced embedding vectors to a preset low-dimensional space to obtain the corresponding coordinate points, and draw them on a plane according to the position of each coordinate point in the low-dimensional space to generate a dimensionality-reduced scatter plot.

[0217] The time-series trajectory generation unit is used to connect the dimensionality-reduced scatter plots into time-series paths in chronological order, generating multiple time-series trajectories for each combustion simulation example.

[0218] In some optional implementations, the similar time-series trajectory retrieval and combustion performance analysis module 1004 includes:

[0219] The convergence radius setting unit is used to calculate the average of multiple dimensionality-reduced embedding vectors for each combustion simulation, and to calculate the convergence radius of each combustion simulation case based on the multiple dimensionality-reduced embedding vectors and their corresponding averages.

[0220] The difference calculation unit is used to determine all time-series trajectories within a preset convergence radius as a candidate set, and calculate the difference between each time-series trajectory in the candidate set and the specified time-series trajectory based on a preset difference quantification formula.

[0221] The most similar time-series trajectory selection unit is used to sort the time-series trajectories in the candidate set in ascending order of difference, and select the time-series trajectory with the smallest difference among the first preset number of time-series trajectories as the most similar time-series trajectory to the specified time-series trajectory.

[0222] The convergence analysis unit is used to determine the distribution of the endpoints of a specified time-series trajectory and similar time-series trajectories on the dimensionality-reduced scatter plot, and analyze the combustion convergence based on the distribution.

[0223] The stability analysis unit is used to obtain the fluctuation degree of a specified time-series trajectory and similar time-series trajectories on the dimensionality-reduced scatter plot, and analyze the stability of combustion based on the fluctuation degree.

[0224] In some alternative implementations, the visual analysis apparatus for combustion simulation further includes:

[0225] The system interface design module is used to visualize the entire lifecycle processing of each combustion simulation example.

[0226] The interactive graphic report generation module is used to embed representative cluster centers based on expert domain description annotations of time-series fluent image encoding, and to construct a combustion mode knowledge base based on the representative cluster centers. The combustion mode knowledge base, time-series fluent images, and representative cluster centers are input into a preset time-series language model to generate descriptive explanations of the time-series fluent images. Based on the time-series fluent images and the corresponding descriptive explanations, an interactive graphic summary report for each combustion simulation example is generated.

[0227] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0228] In this embodiment, the visual analysis device for combustion simulation is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0229] This invention also provides a computer device having the above-described features. Figure 10 The visual analysis device for combustion simulation shown.

[0230] Please see Figure 11 , Figure 11 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 11 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 11 Take a processor 10 as an example.

[0231] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0232] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0233] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0234] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0235] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 11 Taking the example of a connection between China and Israel via a bus.

[0236] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0237] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0238] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0239] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A visual analysis method for combustion simulation, characterized in that, The method includes: Under different initial conditions of the combustion simulation, multiple sets of time-series smooth images are generated for each combustion simulation case; the initial conditions include static pressure, static temperature, and water vapor concentration. The process of generating multiple sets of time-series smooth images for each combustion simulation case under different initial conditions includes: By combining different values ​​of static pressure, static temperature and water vapor concentration, and simulating the combustion process under each combination, multiple sets of combustion flow field time slices are generated. The same color and scaling ratio are used to render different time-series flow field components in each combustion flow field time slice. Irrelevant background information is removed, and important regions of different time-series flow field components are retained to obtain multiple sets of time-series flow field images corresponding to each set of initial conditions. The different time-series flow field components include pressure, OH radicals, Mach number and velocity. Encode each frame of each group of time-series smooth images to obtain multiple corresponding high-dimensional embedding vectors; After performing dimensionality reduction and clustering on the multiple high-dimensional embedding vectors, a dimensionality-reduced scatter plot is generated, and multiple time-series trajectories for each combustion simulation case are generated based on the dimensionality-reduced scatter plot. Within a preset convergence radius, the system retrieves the most similar time-series trajectory to the specified time-series trajectory and analyzes the combustion convergence and stability based on the specified time-series trajectory and the similar time-series trajectory.

2. The method according to claim 1, characterized in that, The process of encoding each frame in each group of temporally smooth images yields multiple corresponding high-dimensional embedding vectors, including: Different temporal flow field components are cropped into specified important regions, and each frame of the specified important region is divided into a uniform grid. At the same time, a spatial weighting matrix is ​​used to highlight the specified important regions, resulting in a weighted temporal flow field image for each frame. Each frame of the time-series flow field image after weighted processing is input into a pre-trained visual ViT model for encoding, and multiple high-dimensional embedding vectors are extracted.

3. The method according to claim 2, characterized in that, Each high-dimensional embedding vector is expressed by the formula: ； in, This represents the high-dimensional embedding vector of the pressure field at time t for the i-th combustion simulation example. This represents a specified important region for the i-th combustion simulation case at time t. Represents a spatial weighted matrix. This indicates the selected ViT encoder.

4. The method according to claim 2, characterized in that, The step of generating a dimensionality-reduced scatter plot after performing dimensionality reduction and clustering on the multiple high-dimensional embedding vectors, and generating multiple time-series trajectories for each combustion simulation example based on the dimensionality-reduced scatter plot, includes: All high-dimensional embedding vectors corresponding to the time-series flow field components are reduced in dimensionality using a preset dimensionality reduction algorithm, and then all the high-dimensional embedding vectors after dimensionality reduction are clustered using a preset clustering algorithm to obtain multiple dimensionality-reduced embedding vectors. The multiple dimensionality-reduced embedding vectors are mapped to a preset low-dimensional space to obtain the corresponding coordinate points, and then plotted on a plane according to the position of each coordinate point in the low-dimensional space to generate a dimensionality-reduced scatter plot. The dimensionality-reduced scatter plots are connected in chronological order to form a time-series path, generating multiple time-series trajectories for each combustion simulation example.

5. The method according to claim 1, characterized in that, The convergence radius is set in the following manner: Calculate the average of multiple dimensionality-reduced embedding vectors for each combustion simulation, and calculate the convergence radius of each combustion simulation case based on the multiple dimensionality-reduced embedding vectors and their corresponding averages.

6. The method according to claim 1, characterized in that, Retrieve the most similar time series trajectory to the specified time series trajectory within a preset convergence radius, including: All time-series trajectories within the preset convergence radius are identified as a candidate set, and the degree of difference between each time-series trajectory in the candidate set and a specified time-series trajectory is calculated based on a preset difference quantification formula. The time series trajectories in the candidate set are sorted in ascending order of their difference, and the time series trajectory with the smallest difference among the first preset number is selected as the most similar time series trajectory to the specified time series trajectory.

7. The method according to claim 1, characterized in that, The analysis of combustion convergence and stability based on specified time-series trajectories and similar time-series trajectories includes: Determine the distribution of the endpoints of the specified time-series trajectory and similar time-series trajectories on the dimensionality-reduced scatter plot, and analyze the combustion convergence based on the distribution. On the dimensionality-reduced scatter plot, the fluctuation degree of the specified time-series trajectory and similar time-series trajectories is obtained, and the stability of combustion is analyzed based on the fluctuation degree.

8. The method according to claim 1, characterized in that, The method further includes: The entire lifecycle processing of each combustion simulation case is visualized. Representative cluster centers are embedded based on expert domain description annotation of temporal smooth image encoding, and a combustion pattern knowledge base is constructed based on the representative cluster centers. The combustion mode knowledge base, time-series smooth images, and representative cluster centers are input into a preset time-series language model to generate descriptive explanations of the time-series smooth images. Based on the time-series smooth images and the corresponding descriptive explanations, an interactive graphic summary report for each combustion simulation example is generated.

9. A visual analysis device for combustion simulation, characterized in that, The device includes: The time-series smooth image generation module is used to generate multiple sets of time-series smooth images for each combustion simulation case under different initial conditions of combustion simulation; the initial conditions include static pressure, static temperature, and water vapor concentration. The process of generating multiple sets of time-series smooth images for each combustion simulation case under different initial conditions includes: By combining different values ​​of static pressure, static temperature and water vapor concentration, and simulating the combustion process under each combination, multiple sets of combustion flow field time slices are generated. The same color and scaling ratio are used to render different time-series flow field components in each combustion flow field time slice. Irrelevant background information is removed, and important regions of different time-series flow field components are retained to obtain multiple sets of time-series flow field images corresponding to each set of initial conditions. The different time-series flow field components include pressure, OH radicals, Mach number and velocity. The high-dimensional embedding vector generation module is used to encode each frame of each group of time-series smooth images to obtain multiple corresponding high-dimensional embedding vectors. The time-series trajectory generation module is used to generate a dimensionality-reduced scatter plot after performing dimensionality reduction and clustering on the multiple high-dimensional embedding vectors, and to generate multiple time-series trajectories for each combustion simulation case based on the dimensionality-reduced scatter plot. The similar time-series trajectory retrieval and combustion performance analysis module is used to retrieve the most similar time-series trajectory to the specified time-series trajectory within a preset convergence radius, and to analyze combustion convergence and stability based on the specified time-series trajectory and the similar time-series trajectory.

Images