Flow field measurement data compression and storage method and system

Abstract

The invention discloses a flow field measurement data compression and storage method and system. The method comprises the following steps: establishing an experimental device, and carrying out information acquisition on a multiphase flow field through a flow field measurement technology to obtain original flow field data; preprocessing the original flow field data to obtain preprocessed flow field data; constructing a deep neural network model, and detecting a flow field key area in the preprocessed flow field data to obtain geometric parameters of the flow field key area; characterizing the flow field key area to obtain structural parameters of the flow field key area; storing the geometric parameters and the structural parameters in a binary format; and restoring the flow field data according to the geometric parameters and the structural parameters to realize flow field reconstruction. The flow field structure is extracted based on the flow field characteristics, efficient compression and storage are performed on the flow field, and reconstruction of the flow field is realized. The storage space of flow field data is saved while the key structure information of the flow field is reserved.

Description

Technical Field

[0001] The present invention relates to the field of multiphase flow field measurement, and more specifically, to a flow field measurement data compression and storage method and system. Background Art

[0002] Multiphase flow is a typical complex system. The measurement of multiphase flow is a key technical issue in studying its internal dynamic evolution mechanism. At present, multiphase flow measurement technologies include high-speed photography, radiography, ultrasonic imaging, and fiber optic probes, which can obtain the measurement parameters of the flow field flow, and use signal conversion, high-speed Ethernet, and high-speed storage technology to realize the processing, transmission, and storage of flow field measurement data.

[0003] With the development of technology, the temporal resolution and spatial resolution of flow field measurement technology have been rapidly improved. After obtaining the flow field measurement data using the above flow field measurement method, the flow field measurement data must first be transferred to high-speed memory or high-speed storage using a high-bandwidth data transmission method. If the data is to be read, a high-speed reading device or an efficient graphics processing tool is required. The post-processing of the flow field data also requires the use of high-performance computing equipment. If a speed bottleneck occurs in any of the above links, it will directly affect the analysis and research process of the multiphase flow field. If compression is performed after data acquisition, the pressure of data processing can be reduced while preserving key flow field information as much as possible.

[0004] Flow field data is generally stored in the form of images or videos. Different compression methods can be selected according to different scene requirements, ranging from lossless storage to efficient compression. Common image formats include TIFF, JPEG, PNG, BMP, etc. For example, JPEG images are a common image storage method, which can achieve efficient compression of original image information through discrete cosine transform (DCT), and the compression ratio is usually between 10:1 and 40:1. For videos, if a high-speed camera with a resolution and shooting rate of 1280×800@24270 frames / second is used to shoot, 128G image data will be generated in 3.5 seconds. Different encoding methods determine the degree of compression of video content. H.264 is a widely used video compression standard. By dividing video frames into blocks and using motion compensation, intra-frame prediction, transform coding, and entropy coding to compress each block, the size of the video file is reduced, while ensuring high-definition and smooth video playback effects. The compression ratio is usually between 2:1 and 10:1.

[0005] It should be noted that the specific value of the compression ratio may vary due to different compression parameters and encoding settings. However, whether it is images or videos, the existing compression and storage methods are based on pixel processing and do not consider the structural characteristics of the processed objects. For the qualitative and quantitative analysis of the flow field, it is required to save the measurement data as completely as possible for subsequent data processing. The existing pixel-based global compression method cannot cope with the large-scale data generation.

[0006] Taking into account the single background characteristics of multiphase flow and the slight differences in flow field structure, the data storage method based on pixel-level images or videos can neither efficiently utilize the storage space nor reflect the mesoscale structure of the multiphase flow field, thus affecting the transmission, storage, reading and post-processing of flow field data.

[0007] Therefore, it is necessary to develop a measurement data compression and storage method and system that takes into account the flow field structural characteristics.

[0008] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0009] The present invention proposes a flow field measurement data compression and storage method and system, which extracts the flow field structure based on the flow field characteristics, efficiently compresses and stores the flow field, and realizes the reconstruction of the flow field. It retains the key structural information of the flow field and saves the storage space of the flow field data.

[0010] In a first aspect, an embodiment of the present disclosure provides a flow field measurement data compression and storage method, comprising:

[0011] Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data;

[0012] Preprocessing the original flow field data to obtain preprocessed flow field data;

[0013] Constructing a deep neural network model to detect the key flow field area in the preprocessed flow field data and obtain geometric parameters of the key flow field area;

[0014] Characterizing the key area of ​​the flow field to obtain structural parameters of the key area of ​​the flow field;

[0015] Saving the geometric parameters and the structural parameters in a binary format;

[0016] The flow field data is restored according to the geometric parameters and the structural parameters to achieve flow field reconstruction.

[0017] Preferably, the pretreatment comprises:

[0018] The noise of the raw flow field data is removed by a filter.

[0019] Preferably, the preprocessing further comprises:

[0020] If the original flow field data contains flow field background information, the flow field background information is removed.

[0021] Preferably, constructing a deep neural network model to detect key flow field areas in preprocessed flow field data includes:

[0022] Annotations are added to the preprocessed flow field data as a training set, a deep neural network model is constructed according to the flow field characteristics, the key areas of the flow field are detected, the key areas in the flow field are located, and the category information, confidence and geometric parameters of the key areas of the flow field are output.

[0023] Preferably, characterizing the key flow field area to obtain structural parameters of the key flow field area includes:

[0024] According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely the structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

[0025] In a second aspect, the present disclosure also provides a flow field measurement data compression and storage system, including:

[0026] a memory storing computer executable instructions;

[0027] A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps:

[0028] Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data;

[0029] Preprocessing the original flow field data to obtain preprocessed flow field data;

[0030] Constructing a deep neural network model to detect the key flow field area in the preprocessed flow field data and obtain geometric parameters of the key flow field area;

[0031] Characterizing the key area of ​​the flow field to obtain structural parameters of the key area of ​​the flow field;

[0032] Saving the geometric parameters and the structural parameters in a binary format;

[0033] The flow field data is restored according to the geometric parameters and the structural parameters to achieve flow field reconstruction.

[0034] Preferably, the pretreatment comprises:

[0035] The noise of the raw flow field data is removed by a filter.

[0036] Preferably, the preprocessing further comprises:

[0037] If the original flow field data contains flow field background information, the flow field background information is removed.

[0038] Preferably, constructing a deep neural network model to detect key flow field areas in preprocessed flow field data includes:

[0039] Annotations are added to the preprocessed flow field data as a training set, a deep neural network model is constructed according to the flow field characteristics, the key areas of the flow field are detected, the key areas in the flow field are located, and the category information, confidence and geometric parameters of the key areas of the flow field are output.

[0040] Preferably, characterizing the key flow field area to obtain structural parameters of the key flow field area includes:

[0041] According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely the structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

[0042] The methods and apparatus of the present invention have other features and advantages that will be apparent from, or will be described in detail in, the accompanying drawings and subsequent detailed descriptions incorporated herein, which together serve to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0044] Figure 1 A flow chart showing the steps of a method for compressing and storing flow field measurement data according to an embodiment of the present invention.

[0045] Figure 2 A schematic diagram showing a process of obtaining an original image of a flow field by reading a bubble flow with a high-speed camera according to an embodiment of the present invention.

[0046] Figure 3 A schematic diagram showing the result of preprocessing the original flow field data according to an embodiment of the present invention.

[0047] Figure 4 A schematic diagram showing the result of detecting a gas-liquid flow field image using a detection model according to an embodiment of the present invention.

[0048] Figure 5 A schematic diagram showing a compression process of writing 11 bubble geometric parameters and structural parameters into a binary file according to an embodiment of the present invention.

[0049] Figure 6 A schematic diagram showing a flow field reconstruction result according to an embodiment of the present invention is shown.

[0050] Figure 7 A schematic diagram of a processing flow of a gas-solid two-phase flow field according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0051] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0052] Figure 1 A flow chart showing the steps of a method for compressing and storing flow field measurement data according to an embodiment of the present invention.

[0053] like Figure 1 As shown, the flow field measurement data compression and storage method includes:

[0054] Step 101, constructing an experimental device, collecting information on the multiphase flow field through flow field measurement technology, and obtaining original flow field data;

[0055] Step 102, preprocessing the original flow field data to obtain preprocessed flow field data;

[0056] Step 103, constructing a deep neural network model, detecting the key flow field area in the preprocessed flow field data, and obtaining geometric parameters of the key flow field area;

[0057] Step 104, characterizing the key area of ​​the flow field to obtain structural parameters of the key area of ​​the flow field;

[0058] Step 105, saving the geometric parameters and structural parameters in binary format;

[0059] Step 106, restore the flow field data according to the geometric parameters and the structural parameters to achieve flow field reconstruction.

[0060] In one example, preprocessing includes:

[0061] The noise of the original flow field data is removed by filtering.

[0062] In one example, preprocessing also includes:

[0063] If the original flow field data contains flow field background information, the flow field background information is removed.

[0064] In one example, building a deep neural network model to detect key flow field areas in preprocessed flow field data includes:

[0065] Annotations are added to the preprocessed flow field data as a training set, and a deep neural network model is constructed according to the flow field characteristics to detect the key areas of the flow field, locate the key areas in the flow field, and output the category information, confidence and geometric parameters of the key areas of the flow field.

[0066] In one example, the key area of ​​the flow field is characterized, and the structural parameters of the key area of ​​the flow field include:

[0067] According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

[0068] Specifically, an experimental device is built and flow field measurement technology is used to collect information on the multiphase flow field. The collection technology includes but is not limited to high-speed photography, X-ray imaging, ultrasonic imaging, fiber optic probes, etc. The collected data is called raw flow field data.

[0069] Due to the limitations of the experimental device, impurities in the mobile phase, interference from experimental conditions and other non-human factors, the original flow field data contains uneven noise or background information. Therefore, the original flow field data must be preprocessed to remove the influence of non-human factors and factors unrelated to the key structure of the flow field, and to retain the main information of the flow field structure as much as possible.

[0070] The preprocessing of raw flow field data is divided into two parts: (1) Flow field background processing: If the detection source is visible light, the flow field background has a greater impact on the flow field data. This step considers the propagation path of real light, uses the Beer-Lambert Law to simulate the light path, and removes the flow field background information. (2) Flow field image denoising: The noise information generated simultaneously with the data is removed by the filter method. The flow field data is obtained after the raw flow field data is preprocessed.

[0071] In a multiphase flow system, due to the interaction between phases, the parameters describing multiphase flow are more complex and diverse than those of single-phase flow, such as multiphase flow type, flow velocity of each phase, phase content, interphase drag and other parameters. Therefore, a flow field structure detection method should be designed to facilitate the characterization of the flow field structure.

[0072] The flow field data obtained in the previous step is added with human annotations as a training set, and a deep neural network model is constructed based on the flow field characteristics. The input of the neural network is the preprocessed flow field data, and the output is the information of the key areas in the flow field. The network model is trained using high-performance computing equipment, and the training results are evaluated using a loss function during the training process to continuously correct the model parameters. After the training is completed, the flow field data can be input into the neural network to detect the key areas of the flow field, locate the key areas in the flow field, and output the category information, confidence, and geometric parameters of the key areas.

[0073] Characterizing the pixel area information of the key area in the flow field as low-dimensional structural parameters with actual physical meaning is the key method for flow field data compression. Using the OpenCV visual tool library for processing, including threshold segmentation, edge detection, ellipse fitting and other methods, we can obtain the characterization parameters related to the flow field structure with actual physical meaning. Each parameter is a quantitative parameter of specific physical information in the flow field structure.

[0074] In order to facilitate storage, reading and transmission, the geometric parameters and structural parameters of the key areas of the flow field detected and characterized are saved in the storage medium in binary format. Compared with the conventional decimal storage format, the biggest advantage of binary is that the flow field information is compressed again. In this scenario, for the same data, the size of the decimal file is about 5 times the size of the binary file.

[0075] Flow field reconstruction is the process of restoring the geometric parameters and structural parameters of the key areas of the flow field to flow field data. Using the parameters obtained by detection and characterization, discarding data irrelevant to the flow field, and using lower storage and computing costs to reproduce the original structure of the flow field. Use the OpenCV visual tool library to reconstruct the main structure of the flow field.

[0076] This method reconstructs the flow field for bubbly flow fields. Bubbly flow fields are flow fields with elliptical structures, and the outer contours of their key structures can be fitted using elliptical structures. Bubbly flows include gas-liquid two-phase flow fields, gas-solid two-phase flow fields, liquid-liquid two-phase flow fields, etc.

[0077] The parameters for bubbly flow field detection are the ellipse parameters obtained by fitting the outer contour of the key structure, including the centroid position coordinates, major semi-axis, minor semi-axis, rotation angle, etc.

[0078] The present invention also provides a flow field measurement data compression and storage system, comprising:

[0079] a memory storing computer executable instructions;

[0080] A processor executes the computer executable instructions in the memory to perform the following steps:

[0081] Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data;

[0082] Preprocessing the original flow field data to obtain preprocessed flow field data;

[0083] A deep neural network model is constructed to detect the key flow area in the preprocessed flow field data and obtain the geometric parameters of the key flow area.

[0084] Characterize the key areas of the flow field and obtain the structural parameters of the key areas of the flow field;

[0085] Save the geometric parameters and structural parameters in binary format;

[0086] The flow field data is restored according to the geometric parameters and structural parameters to achieve flow field reconstruction.

[0087] In one example, preprocessing includes:

[0088] The noise of the original flow field data is removed by filtering.

[0089] In one example, preprocessing also includes:

[0090] If the original flow field data contains flow field background information, the flow field background information is removed.

[0091] In one example, building a deep neural network model to detect key flow field areas in preprocessed flow field data includes:

[0092] Annotations are added to the preprocessed flow field data as a training set, and a deep neural network model is constructed according to the flow field characteristics to detect the key areas of the flow field, locate the key areas in the flow field, and output the category information, confidence and geometric parameters of the key areas of the flow field.

[0093] In one example, the key area of ​​the flow field is characterized, and the structural parameters of the key area of ​​the flow field include:

[0094] According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

[0095] Specifically, an experimental device is built and flow field measurement technology is used to collect information on the multiphase flow field. The collection technology includes but is not limited to high-speed photography, X-ray imaging, ultrasonic imaging, fiber optic probes, etc. The collected data is called raw flow field data.

[0096] Due to the limitations of the experimental device, impurities in the mobile phase, interference from experimental conditions and other non-human factors, the original flow field data contains uneven noise or background information. Therefore, the original flow field data must be preprocessed to remove the influence of non-human factors and factors unrelated to the key structure of the flow field, and to retain the main information of the flow field structure as much as possible.

[0097] The preprocessing of raw flow field data is divided into two parts: (1) Flow field background processing: If the detection source is visible light, the flow field background has a greater impact on the flow field data. This step considers the propagation path of real light, uses the Beer-Lambert Law to simulate the light path, and removes the flow field background information. (2) Flow field image denoising: The noise information generated simultaneously with the data is removed by the filter method. The flow field data is obtained after the raw flow field data is preprocessed.

[0098] In a multiphase flow system, due to the interaction between phases, the parameters describing multiphase flow are more complex and diverse than those of single-phase flow, such as multiphase flow type, flow velocity of each phase, phase content, interphase drag and other parameters. Therefore, a flow field structure detection method should be designed to facilitate the characterization of the flow field structure.

[0099] The flow field data obtained in the previous step is added with human annotations as a training set, and a deep neural network model is constructed based on the flow field characteristics. The input of the neural network is the preprocessed flow field data, and the output is the information of the key areas in the flow field. The network model is trained using high-performance computing equipment, and the training results are evaluated using a loss function during the training process to continuously correct the model parameters. After the training is completed, the flow field data can be input into the neural network to detect the key areas of the flow field, locate the key areas in the flow field, and output the category information, confidence, and geometric parameters of the key areas.

[0100] Characterizing the pixel area information of the key area in the flow field as low-dimensional structural parameters with actual physical meaning is the key method for flow field data compression. Using the OpenCV visual tool library for processing, including threshold segmentation, edge detection, ellipse fitting and other methods, we can obtain the characterization parameters related to the flow field structure with actual physical meaning. Each parameter is a quantitative parameter of specific physical information in the flow field structure.

[0101] In order to facilitate storage, reading and transmission, the geometric parameters and structural parameters of the key areas of the flow field detected and characterized are saved in the storage medium in binary format. Compared with the conventional decimal storage format, the biggest advantage of binary is that the flow field information is compressed again. In this scenario, for the same data, the size of the decimal file is about 5 times the size of the binary file.

[0102] Flow field reconstruction is the process of restoring the geometric parameters and structural parameters of the key areas of the flow field to flow field data. Using the parameters obtained by detection and characterization, discarding data irrelevant to the flow field, and using lower storage and computing costs to reproduce the original structure of the flow field. Use the OpenCV visual tool library to reconstruct the main structure of the flow field.

[0103] To facilitate understanding of the solutions and effects of the embodiments of the present invention, three specific application examples are given below. Those skilled in the art should understand that the examples are only for facilitating understanding of the present invention, and any specific details thereof are not intended to limit the present invention in any way.

[0104] Example 1

[0105] In this embodiment, the present invention is further explained by taking the use of high-speed camera technology to measure the bubble flow field as an example.

[0106] Bubble flow in gas-liquid two-phase flow is widely present in many fields such as chemical engineering, environmental engineering, health care, nuclear industry and marine engineering. For the measurement of bubble flow field, high-speed camera technology is a commonly used traditional characterization technology for analyzing fast-moving objects. With the help of signal conversion, high-speed Ethernet and high-speed storage technology, flow field images can be converted, transmitted and stored.

[0107] Figure 2 A schematic diagram showing a process of obtaining an original image of a flow field by reading a bubble flow with a high-speed camera according to an embodiment of the present invention.

[0108] The LED light board emits background light, and the light passes through the bubbling device, carrying the flow field data to be collected by the high-speed camera. In this embodiment, a high-speed camera is used to collect signals of the bubble flow in the cylindrical container. The size of the experimental device is diameter D m=100mm cylinder, gas phase is air, liquid phase is water. The high-speed camera uses Phantom T2410 produced by Vision Research. The actual shooting time is 1 second, and a total of 10,000 frames are captured, that is, the sampling frequency is 10,000 FPS, and the interval is 100 μm. The video file is in CINE format, and the file size is 11.9G (12,801,650,556 bytes). If the video additional information is not considered, the image resolution is 1280×800, and the color depth per pixel is 10 bits. Therefore, the storage space required to save each frame of the image is 10240000 bits (1280000 bytes).

[0109] Figure 3 A schematic diagram showing the result of preprocessing the original flow field data according to an embodiment of the present invention.

[0110] The background information collected synchronously during information acquisition includes asymmetric light source information, floating dust in the liquid phase, stains and scratches on the wall, etc., which will affect the flow field information analysis. Therefore, the original flow field data must be preprocessed. This step is divided into flow field background processing and flow field image denoising. (1) Flow field background processing: The process of light passing through the bubble flow channel is a light absorption process. While taking the gas-liquid two-phase flow field image, a pure background image of the stable water flow is taken. By simulating the real light path, the Beer-Lambert Law is used to perform theoretical solution to obtain the preprocessed bubble flow information. (2) Flow field image denoising: The median filter method is used to denoise the flow field data. The value of a point in the digital image is replaced by the median of the values ​​of each point in a neighborhood of the point, thereby eliminating isolated noise points. Similarly, there are also methods such as mean filtering and Gaussian filtering that use filters for denoising.

[0111] Figure 4 A schematic diagram showing the result of detecting a gas-liquid flow field image using a detection model according to an embodiment of the present invention.

[0112] This example uses a deep learning framework to detect rotating targets, and the detection targets are all bubble structures visible in the flow field. The following steps are required to achieve the detection target:

[0113] (1) Dataset annotation: Use image annotation tools to manually annotate the flow field data. Relevant tools include but are not limited to roLabelImg and lebalImg2. The annotation box uses the rotating target box to annotate each rotating target (bubble). The annotation content is the position information, size and rotation angle of all visible bubbles in the flow field. Bubbles are divided into two categories: single bubbles and overlapping bubbles, such as Figure 3The two bubbles on the left are overlapping bubbles, and the three bubbles on the right are single bubbles. A total of 400 flow field images were annotated and divided into training set and validation set according to 9:1.

[0114] (2) Construct a deep neural network model based on the flow field characteristics. The bubble structure obtained by the photographic measurement technology is a two-dimensional elliptical image with a random angle. Considering the detection model of rotating targets, the bubbles can be accurately located. Common rotating target detection models include MMRotate, YOLOv5-obb, Oriented R-CNN, etc. The training method of the detection model is similar to the inference method, so it will not be repeated. Therefore, the input of the neural network is the flow field data, and the output is the category information, confidence and geometric parameters of the key areas in the flow field.

[0115] (3) The network model is trained on a high-performance computing device using a manually labeled rotation target dataset. During the training process, the forward propagation calculates the output results through the training data and weight parameters, and the training results are evaluated through the loss function. The loss function consists of three parts: bubble category confidence loss, bubble coordinate regression loss, and bubble detection target confidence loss. The confidence loss and category loss are calculated using binary cross entropy, and the coordinate regression loss uses CIOU Loss. Similar loss functions also include IOU, GIOU, DIOU, etc., which are used to calculate the overlap between bubbles with different aspect ratios. Backward propagation calculates the gradient of the loss function for each parameter through the derivative chain rule, and updates the parameters based on the gradient calculated according to the total loss value. During the training process of the neural network, forward propagation and back propagation are performed alternately to continuously update the model parameters until the set number of training rounds is completed and the training is stopped.

[0116] (4) After the training, the flow field data is input into the neural network to abstract the flow field structure. The model can output results similar to the annotated data set, that is, all two types of bubble structures visible in the entire flow field. The output results are the category information, confidence and geometric parameters of the key areas of the flow field. The geometric parameters are the results of using a standard ellipse to approximate the outer contour of the bubble, including position information, size information and rotation angle. The results after detection are as follows: Figure 4 The two bubbles on the left are overlapping bubbles, represented by black elliptical outlines, and the three bubbles on the right are independent single bubbles, represented by gray elliptical outlines.

[0117] The detection of key areas of the flow field can obtain seven parameters, including each bubble category, confidence, and outer contour quasi-ellipse parameters. More specific structural information of the bubble needs to be described using other parameters. Five parameters related to bubble structure characterization are introduced here: eccentricity, circularity, overlap rate, shadow ratio, and edge gradient, as shown in Table 1.

[0118] Table 1

[0119]

[0120] The above five structural parameters can all be obtained by processing using the OpenCV visual tool library. In addition to the above structural parameters, parameters such as center of mass, roundness, aspect ratio, skewness, etc. can also be used to describe the bubble structure, which are not additionally introduced in this embodiment. The five structural parameters are combined with the bubble category and geometric parameters to obtain a total of 11 parameters. Flow field structure characterization is actually a flow field parameterization process.

[0121] In addition to the above 11 bubble geometric parameters and structural parameters, the current interface gas holdup can also be estimated during the detection process. The size of the experimental device in this example is diameter D m = 100 mm cylinder, the width of the flow field in the image is D p =1248pix, each pixel represents the distance PPM (Pixels per meter) calculated as follows:

[0122]

[0123] Image acquisition area height H p =800pix, actual length H m The calculation is as follows:

[0124] H m =PPM×H p =64.1024mm

[0125] The volume of the flow field captured is calculated as follows:

[0126]

[0127] Through flow field structure detection, all bubble types and geometric parameters in the interface can be obtained. The geometric parameters are the major semi-axis, minor semi-axis and rotation angle of the ellipse fitting the outer contour of the bubble.

[0128] Estimate the volume of the three-dimensional ellipsoid through the two-dimensional ellipse, take the major semi-axis a of the two-dimensional ellipse as the length of the third axis of the ellipsoid, and obtain the volume occupied by all bubbles:

[0129]

[0130] Where n is the total number of bubbles in the current interface, a i ,b i are the major and minor semi-axes of the two-dimensional ellipse fitted to the i-th bubble. Then the gas holdup is obtained:

[0131]

[0132] Figure 5 A schematic diagram showing a compression process of writing 11 bubble geometric parameters and structural parameters into a binary file according to an embodiment of the present invention.

[0133] First, convert all parameters to integers, convert 1 category parameter to integers, convert 10 geometric parameters and structure parameters to floating-point numbers, and then use to store all data in binary form in a binary file. The compression ratio is calculated as follows: the resolution of the high-speed camera is 1280×800, and the color depth of each pixel is 10 bits. If only the image pixels are considered, and the file header size is not considered, the storage space occupied by each frame of uncompressed image is:

[0134] Uncompressed Size=1280×800×10bit=10240000bit

[0135] =1280000byte

[0136] After extracting the bubble structure, each bubble can be represented by 10 geometric parameters and structural parameters (floating point numbers) plus 1 category (integer number). Each floating point number occupies 4 bytes, and each integer number occupies 4 bytes. Assuming that there are 100 bubbles in each image, the size of the compressed image file is:

[0137] Compressed Size=100×(4×1+4×10)=4400byte

[0138] For video, each frame is saved as a binary file. The compression ratio is calculated as follows:

[0139]

[0140] This compression and storage method of extracting flow field image data into low-dimensional bubble geometric parameters and structural parameters according to the bubble characteristics in the flow field is different from the traditional pixel-based data compression method. The actual compression ratio is determined by the number of bubbles in each frame of the image. The flow field can be deconstructed from the perspective of the gas-liquid two-phase principle to obtain physically meaningful low-dimensional abstract structural parameters, thereby achieving efficient compression of flow field data.

[0141] Figure 6 A schematic diagram of the flow field reconstruction result according to an embodiment of the present invention is shown, wherein the black ellipse represents overlapping bubbles and the gray ellipse represents a single bubble. The parameters required for reconstruction are the category of each bubble, the centroid position and the parameter information of the outer contour fitting ellipse, and the ellipse information includes the major semi-axis, the minor semi-axis and the rotation angle. The above 6 parameters can be used to draw the bubble contours in each frame image using the OpenCV visual tool library.

[0142] Example 2

[0143] Figure 7 A schematic diagram of a processing flow of a gas-solid two-phase flow field according to an embodiment of the present invention is shown.

[0144] The processing of the gas-solid flow field is similar to that of the gas-liquid flow field. The main structure of the flow field is a closed structure in a two-dimensional space. Therefore, the detection, characterization and reconstruction methods of the flow field structure are similar to those in Example 1.

[0145] Figure 7 (a) shows the measured flow field data, which is a bubbling flow in a gas-solid flow. The following methods are used to detect the flow field: numerical simulation, high-speed camera photography, and X-ray tomography technology.

[0146] Figure 7 (b) shows the result of the original flow field data after preprocessing, and the bubble body is clearly distinguished from the flow field background. In this embodiment, the flow field image denoising and threshold segmentation method are used to preprocess the flow field data.

[0147] (1) Flow field image denoising: Use the median filter method to denoise the flow field data. The value of a point in the digital image is replaced by the median of the values ​​of all points in a neighborhood of the point, thereby eliminating isolated noise points. Similarly, there are also methods that use filters such as mean filtering and Gaussian filtering to denoise.

[0148] (2) Image threshold segmentation: Because there is non-steady-state background information in the gas-solid flow field, it is impossible to stably segment the background information. Therefore, the image threshold segmentation algorithm can be used for preprocessing. This can ensure that in the absence of a stable flow field background image, certain bubble information is retained while removing the background effect. The calculation formula is as follows:

[0149]

[0150] Among them, T0(i,j) refers to the pixel value of the image at (i,j), T(i,j) refers to the pixel value of the image at (i,j) after removing the background, and T th Refers to the segmentation threshold. The threshold is calculated using the maximum between-class variance method (OTSU). The segmentation threshold can also be determined by the bimodal method, iterative method, dynamic threshold method, etc.

[0151] Figure 7 (c) shows the result of detecting the flow field structure using the rotating target detection model. The relevant data set preparation, model design, and training methods are the same as those described in Example 1. The model input is the preprocessed flow field data, and the output result is a total of 6 geometric parameters, including each bubble type and the outer contour quasi-ellipse parameters in the flow field. The bubble structure of the gas-solid system is relatively simple, so there is no need to characterize the detected bubble area to obtain the structural parameters.

[0152] Figure 7 (d) shows the result after gas-solid flow field reconstruction. The flow field information compression storage and flow field structure reconstruction are implemented in the same way as the gas-liquid system.

[0153] Example 3

[0154] a memory storing computer executable instructions;

[0155] A processor executes the computer executable instructions in the memory to perform the following steps:

[0156] Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data;

[0157] Preprocessing the original flow field data to obtain preprocessed flow field data;

[0158] A deep neural network model is constructed to detect the key flow area in the preprocessed flow field data and obtain the geometric parameters of the key flow area.

[0159] Characterize the key areas of the flow field and obtain the structural parameters of the key areas of the flow field;

[0160] Save the geometric parameters and structural parameters in binary format;

[0161] The flow field data is restored according to the geometric parameters and structural parameters to achieve flow field reconstruction.

[0162] In one example, preprocessing includes:

[0163] The noise of the original flow field data is removed by filtering.

[0164] In one example, preprocessing also includes:

[0165] If the original flow field data contains flow field background information, the flow field background information is removed.

[0166] In one example, building a deep neural network model to detect key flow field areas in preprocessed flow field data includes:

[0167] Annotations are added to the preprocessed flow field data as a training set, and a deep neural network model is constructed according to the flow field characteristics to detect the key areas of the flow field, locate the key areas in the flow field, and output the category information, confidence and geometric parameters of the key areas of the flow field.

[0168] In one example, the key area of ​​the flow field is characterized, and the structural parameters of the key area of ​​the flow field include:

[0169] According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

[0170] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any given examples.

[0171] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A flow field measurement data compression and storage method, characterized in that: include: Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data; Preprocessing the original flow field data to obtain preprocessed flow field data; Constructing a deep neural network model to detect the key flow field area in the preprocessed flow field data and obtain geometric parameters of the key flow field area; Characterizing the key area of ​​the flow field to obtain structural parameters of the key area of ​​the flow field; Saving the geometric parameters and the structural parameters in a binary format; The flow field data is restored according to the geometric parameters and the structural parameters to achieve flow field reconstruction.

2. The flow field measurement data compression and storage method according to claim 1, wherein: The pre-processing comprises: The noise of the raw flow field data is removed by a filter.

3. The flow field measurement data compression and storage method according to claim 1, wherein: The pre-processing further comprises: If the original flow field data contains flow field background information, the flow field background information is removed.

4. The flow field measurement data compression and storage method according to claim 1, wherein: Constructing a deep neural network model to detect key flow field areas in preprocessed flow field data includes: Annotations are added to the preprocessed flow field data as a training set, a deep neural network model is constructed according to the flow field characteristics, the key areas of the flow field are detected, the key areas in the flow field are located, and the category information, confidence and geometric parameters of the key areas of the flow field are output.

5. The flow field measurement data compression and storage method according to claim 1, wherein: Characterizing the key area of ​​the flow field, obtaining structural parameters of the key area of ​​the flow field includes: According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely the structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

6. A flow field measurement data compression and storage system, characterized in that: The system includes: a memory storing computer executable instructions; A processor, wherein the processor runs the computer executable instructions in the memory to perform the following steps: Build an experimental device, collect information on the multiphase flow field through flow field measurement technology, and obtain original flow field data; Preprocessing the original flow field data to obtain preprocessed flow field data; Constructing a deep neural network model to detect the key flow field area in the preprocessed flow field data and obtain geometric parameters of the key flow field area; Characterizing the key area of ​​the flow field to obtain structural parameters of the key area of ​​the flow field; Saving the geometric parameters and the structural parameters in a binary format; The flow field data is restored according to the geometric parameters and the structural parameters to achieve flow field reconstruction.

7. The flow field measurement data compression and storage system according to claim 6, wherein: The pre-processing comprises: The noise of the raw flow field data is removed by a filter.

8. The flow field measurement data compression and storage system according to claim 6, wherein: The pre-processing further comprises: If the original flow field data contains flow field background information, the flow field background information is removed.

9. The flow field measurement data compression and storage system according to claim 6, wherein: Constructing a deep neural network model to detect key flow field areas in preprocessed flow field data includes: Annotations are added to the preprocessed flow field data as a training set, a deep neural network model is constructed according to the flow field characteristics, the key areas of the flow field are detected, the key areas in the flow field are located, and the category information, confidence and geometric parameters of the key areas of the flow field are output.

10. The flow field measurement data compression and storage system according to claim 6, wherein: Characterizing the key area of ​​the flow field, obtaining structural parameters of the key area of ​​the flow field includes: According to the preprocessed flow field data, characterization parameters with actual physical significance related to the flow field structure are obtained, namely the structural parameters, which are quantitative parameters of specific physical information in the flow field structure.

Images