AI multiphase flow detection system training data acquisition device and use method thereof
By designing a training data acquisition device for an AI multiphase flow detection system, the problems of difficulty in acquiring multiphase flow detection data and lack of sensor calibration benchmarks were solved, achieving high-precision multiphase flow detection data acquisition and wide applicability of the AI model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING NORMAL UNIVERSITY
- Filing Date
- 2026-01-20
- Publication Date
- 2026-05-12
AI Technical Summary
In existing technologies, it is difficult to acquire multiphase flow detection data, sensors lack reliable calibration benchmarks, AI training data is scarce and labeling costs are high, and the model has poor generalization ability under different conditions.
An AI multiphase flow detection system training data acquisition device was designed, including an overall frame, a transparent container, a motion module, a rotating drum assembly, an image acquisition module, and a data processing module. By simulating dispersed phases at different velocities, high-quality training data is acquired, and the recognition pattern on the rotating drum is used for correction.
It achieves high-precision and reliable acquisition of multiphase flow detection data, can simulate different working conditions, provides widely applicable training data, and improves the robustness of AI models and the accuracy of sensor calibration.
Smart Images
Figure CN122017282A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of industrial process detection and artificial intelligence technology, specifically relating to an AI multiphase flow detection system training data acquisition device and its usage method. Background Technology
[0002] In many fields such as chemical engineering, the measurement of multiphase flow has always been a challenge because it is in a flowing state and each phase has different physicochemical properties, and the phase interface is complex and variable. Accurately measuring the size and velocity of the dispersed phase is extremely challenging.
[0003] Common testing methods include high-speed imaging, optical tomography, nuclear magnetic resonance (NMR), particle image velocimetry (PIV), and laser Doppler velocimetry (LDV). The accuracy of these methods is affected by a variety of factors, making it difficult to measure the content and velocity of each phase with high precision. Calibration and verification are difficult, and there is a lack of standardized verification methods. Some methods use expensive and complex instruments with safety limitations, making them difficult to apply in online industrial settings.
[0004] A key development trend in multiphase flow detection technology is to compensate for the shortcomings of single measurement techniques and improve the robustness and reliability of the system by using multi-sensor information fusion and artificial intelligence data-driven methods.
[0005] While AI has demonstrated unprecedented advantages in multiphase flow detection, it also faces some significant challenges, including the scarcity of high-quality labeled data, as acquiring accurate and well-labeled datasets covering all flow types under real-world industrial conditions is extremely costly and difficult; and poor generalization ability, where models trained on specific devices and fluid properties may experience severe performance degradation when applied to different pipe diameters, inclinations, media, or sensor configurations. Summary of the Invention
[0006] To address the aforementioned problems, this invention provides a training data acquisition device and its usage method for an AI multiphase flow detection system, thereby overcoming the shortcomings of existing technologies, such as difficulty in acquiring multiphase flow detection data, lack of reliable calibration benchmarks for sensors, scarcity of AI training data, and high labeling costs.
[0007] The specific plan is as follows: A training data acquisition device for an AI multiphase flow detection system includes: The overall frame includes a support bracket for fixing the device and a support bracket for fixing the detection probe of the image acquisition module. A transparent container with mounting holes on the side wall; The motion module includes a vertical motion module that controls the up-and-down movement of the rotary drum assembly and a rotary motion module that drives the rotary drum assembly to rotate. The rotating drum assembly, housed within a transparent container and installed with the motion module, consists of several rotating drums of different diameters and bearing different identification patterns, used to simulate dispersed phases at different velocities. The image acquisition module, used to acquire multiphase flow image information containing the identification pattern on the rotating drum, is installed in the mounting hole and includes an industrial camera, fiber optic sensor and laser sensor; The data processing module is used to identify the dispersed phase in the multiphase flow image information and obtain the image information of the simulated rotating drum. The image information of the simulated rotating drum is used as a benchmark to correct the detection results, and finally the training data of the AI multiphase flow detection system that can be used for intelligent detection is obtained.
[0008] Furthermore, the vertical motion module includes a stepper motor, a lead screw, and a slide table; the lead screw is vertically fixed to the device bracket; the output end of the stepper motor is mounted on the lead screw; the slide table is slidably mounted on the lead screw and has a connecting plate extending horizontally upward toward the transparent container; the rotary motion module includes a rotary motor and a coupling; the rotary motor is mounted on the connecting plate, and its output end is downwardly mounted to the rotating drum assembly through the coupling.
[0009] Furthermore, the identification pattern on the rotating drum consists of several micropores; the diameter of the micropores is 100 to 1000 μm.
[0010] Furthermore, the transparent container is a transparent plexiglass container, and a laser sensor, a fiber optic sensor, and an industrial camera are installed at the mounting holes.
[0011] Furthermore, limit switches are installed at both ends of the lead screw.
[0012] Further, the training data is acquired using the device in the following steps; S1. Install the selected rotating drum to the rotary motor via a coupling, and install the rotary motion module with the integrated rotating drum onto the slide of the vertical motion module; install the image acquisition module onto the mounting hole on the side wall of the transparent container; inject working fluid into the transparent container; S2. Place a ruler in a transparent container, use the installed image acquisition module to acquire the image of the ruler, and establish the proportional relationship between the image pixels and the actual physical size through the data processing module. S3. Based on the target motion state of the dispersed phase to be simulated, set the speed of the stepper motor and the rotary motor so that the rotating drum can rotate and move vertically in the working fluid at the same time; and synchronously acquire the multiphase flow image information of the micropore pattern on the rotating drum through the image acquisition module. S4. Process the acquired multiphase flow image information and identify the micro-hole pattern on the rotating drum; calculate the true size of the dispersed phase simulated by each micro-hole based on the proportional relationship established in step S2; calculate the true motion velocity of the simulated dispersed phase based on the displacement and time interval of the same identified pattern in consecutive frame images. S5. Save the dispersed phase data labeled with the true size and true motion velocity obtained in step S4 as a dataset for training the AI model; and use the true size and true motion velocity as benchmark values to compare with the measurement values of other sensors under the same conditions to calculate the correction coefficient of the sensor.
[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention has a clear structure, reliable data, is robust and durable, and is easy to use; 2. The rotating drum uses laser to precisely drill small holes, which is highly accurate and has a small error; different rotating drum radii and different motor speeds can be matched to simulate the movement speed of the dispersed phase at any speed; at the same time, the rotation can move up and down in the vertical plane, which can simulate the simultaneous movement of the dispersed phase in the horizontal and vertical directions. 3. Different liquids can be placed in the transparent container for measurement calibration, avoiding the influence of different physicochemical properties of the liquids on the measurement results; 4. Using this invention, it is possible to obtain diverse data with precise annotations, including the size and a wide range of motion velocities of the dispersed phase, for training of AI intelligent multiphase flow detection systems. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the structure of the transparent container in this invention.
[0015] Explanation of reference numerals in the attached drawings: 1. Device support; 2. Detection probe support; 3. Transparent container; 4. Rotary drum; 5. Industrial camera; 6. Fiber optic sensor; 7. Laser sensor; 8. Stepper motor; 9. Lead screw; 10. Slide table; 11. Rotary motor; 12. Coupling; 13. Aeration head; 14. Liquid receiving cylinder. Detailed Implementation
[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The specific implementation methods of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Example
[0017] like Figure 1 As shown, the present invention is a training data acquisition device for an AI multiphase flow detection system, comprising an overall frame, a transparent container 3, a motion module, a rotating cylinder assembly, an image acquisition module, and a data processing module.
[0018] The overall frame includes a support bracket 1 for fixing the device and a detection probe bracket 2 for fixing the image acquisition module.
[0019] The transparent container 3 is a transparent plexiglass container with mounting holes on the side wall for mounting the image acquisition module.
[0020] The motion module includes a vertical motion module that controls the up-and-down movement of the rotating drum assembly and a rotary motion module that drives the rotating drum assembly to rotate. The vertical motion module includes a stepper motor 8, a lead screw 9, and a slide table 10; the lead screw 9 is vertically fixed to the device bracket 1, and limit switches are installed at both ends; the output end of the stepper motor 8 is mounted on the lead screw 9; the slide table 10 is slidably mounted on the lead screw 9, and a connecting plate extends horizontally upward toward the transparent container 3; the rotary motion module includes a rotary motor 11 and a coupling 12; the rotary motor 11 is mounted on the connecting plate, and its output end is downward and mounted to the rotating drum assembly through the coupling 12.
[0021] The rotating cylinder assembly is set inside the transparent container 3 and installed with the motion module. It consists of several rotating cylinders 4 with different diameters and different recognition patterns to simulate dispersed phases at different speeds. The recognition pattern on the rotating cylinder 4 consists of several micropores with a diameter of 100 to 1000 μm.
[0022] The image acquisition module is used to acquire multiphase flow image information containing the identification pattern on the rotating drum. It is set in the mounting hole and includes an industrial camera 5, a fiber optic sensor 6, and a laser sensor 7. The multiphase flow identified by the industrial camera 5 is used as the standard, and the other sensors perform synchronous detection. The detection results are compared and corrected with the camera recognition results.
[0023] The data processing module is used to identify the dispersed phase in the multiphase flow image information and obtain the image information of the simulated rotating drum. The image information of the simulated rotating drum is used as a benchmark to correct the detection results, and finally the training data of the AI multiphase flow detection system that can be used for intelligent detection is obtained.
[0024] In this embodiment, the patterns on the rotating drum 4 are of two types: one is a uniform arrangement of holes of different diameters, and the other is a fixed arrangement of several different diameter holes, forming an identification pattern. Furthermore, the true value of the dispersed phase simulated by the pattern on the rotating drum 4 is compared with the measurement results of other sensors to derive a correction coefficient, thus correcting other multiphase flow detection sensors. When the rotating drum 4 is placed in different types of continuous phases, the different physical properties of these continuous phases will lead to different optical detection data distortions, or optical distortions due to varying installation distances. Therefore, the true values can be used to correct for multiphase flows in the actual detection environment.
[0025] The velocity of the micropores on the rotating drum 4 is determined by the rotational speed of the rotating motor 11 and the radius of the rotating drum 4. The actual velocity of the dispersed phase is: Where R is the radius of the rotating drum 4 and n is the rotational speed.
[0026] Different working solutions can be selected in the transparent container 3 according to the actual working conditions. Commonly used solutions include water, organic solutions, and mixed organic solutions. When passing through different solutions, the refractive index captured by the industrial camera 5 is different, which will cause slight changes in the size of the dispersion terms in the image. A coefficient can be multiplied in the recognition result for uniform correction.
[0027] The industrial camera 5 is also equipped with a telecentric lens and a coaxial light source. The high-resolution camera has a pixel count ranging from 5 million to 20 million. Taking a 12 million pixel lens as an example, it divides the image frame into 4096*3000 grids, with each pixel corresponding to one grid. The higher the pixel count of an image of the same size, the more refined the recognition of image details.
[0028] The lens determines the camera's image sensor size and depth of field. During focusing, only one plane is truly in focus. In the out-of-focus state, where the object plane is in front of or behind the focusing plane, light emitted from a point on the object at different angles falls on the image plane, forming a blurred circle called the circle of confusion. If the diameter of the circle of confusion is smaller than the sensor pixel size, this circle of confusion is called the permissible circle of confusion. There is a distance in front of and behind the focusing plane, and the image falls within the permissible range of the circle of confusion; this distance is called the depth of field. During shooting, with a working distance of 110mm, a magnification of 0.75X, and an object-side depth of field of 0.8mm-3.12mm, adjusting the mounting position can meet the requirements for simulating dispersed phase shooting using the small aperture on the rotating drum.
[0029] The front of the lens is equipped with a coaxial light source. The power of the light source and the lens aperture control the image brightness. However, since the aperture size is inversely proportional to the depth of field, the larger the aperture, the shallower the depth of field. In the camera module, a higher light source brightness and a smaller aperture size are used to prevent the circle of confusion from exceeding the sensor pixels, which would result in a blurry image and affect the sensor accuracy.
[0030] Furthermore, the ratio of the image range to the resolution captured by the industrial camera 5 yields a single-pixel accuracy of 4.6µm. For a 200µm dispersed phase, an error of 2.6% can be calculated, which is within an acceptable range. When the object being photographed is in motion, a rolling shutter camera, at the start of exposure, scans line by line until all lines are exposed, which can cause spherical dispersed phases to be captured as elliptical. A global shutter camera exposes all pixels simultaneously. Therefore, in this embodiment, the industrial camera 5 is a global shutter camera, which can accurately capture the shape of the dispersed phases.
[0031] Since the dispersion term is in motion, the rotation speed of the rotary motor 11 is set to 400 r / min, and the diameter of the rotating drum 4 is 50 mm. Therefore, the simulated speed of the dispersion term is 2093 mm / s, and the corresponding maximum non-fading exposure time is single pixel accuracy / dispersion term speed = 4.6 μm / (2093 mm / s) = 2.2 μs.
[0032] Example 2;
[0033] This embodiment, based on the AI multiphase flow detection system training data acquisition device in Embodiment 1, provides the following steps for acquiring training data: S1. Install the selected rotating drum to the rotary motor via a coupling, and install the rotary motion module with the integrated rotating drum onto the slide of the vertical motion module; install the image acquisition module onto the mounting hole on the side wall of the transparent container; inject working fluid into the transparent container; S2. Place a ruler in a transparent container, use the installed image acquisition module to acquire the image of the ruler, and establish the proportional relationship between the image pixels and the actual physical size through the data processing module. S3. Based on the target motion state of the dispersed phase to be simulated, set the speed of the stepper motor and the rotary motor so that the rotating drum can rotate and move vertically in the working fluid at the same time; and synchronously acquire the multiphase flow image information of the micropore pattern on the rotating drum through the image acquisition module. S4. Process the acquired multiphase flow image information and identify the micro-hole pattern on the rotating drum; calculate the true size of the dispersed phase simulated by each micro-hole based on the proportional relationship established in step S2; calculate the true motion velocity of the simulated dispersed phase based on the displacement and time interval of the same identified pattern in consecutive frame images. S5. Save the dispersed phase data labeled with the true size and true motion velocity obtained in step S4 as a dataset for training the AI model; and use the true size and true motion velocity as benchmark values to compare with the measurement values of other sensors under the same conditions to calculate the correction coefficient of the sensor.
[0034] Example 3;
[0035] This embodiment is based on the AI multiphase flow detection system training data acquisition device provided in Embodiment 1, and further improves upon it: An aeration head 13 is added inside the transparent container 3. Gas (such as air) is introduced through the aeration head 13, which can generate a realistic and randomly distributed bubble flow in the continuous phase liquid. The size distribution and rising speed of these bubbles are naturally formed, making them more random and realistic, and able to simulate more complex multiphase flow conditions. At the same time, a liquid inlet 15 is added to the top of the transparent container 3 to facilitate the addition of continuous phase liquid; a liquid receiving pipe is connected to the bottom of the transparent container 3, and a valve that can control the on / off state is installed on the liquid receiving pipe, which is connected to the liquid receiving cylinder 14 to facilitate cleaning of the transparent container 3 after the experiment.
[0036] The above embodiments are for illustrative purposes only and are not intended to limit the scope of this invention. Although this invention has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of this invention do not depart from the spirit and scope of the technical solutions of this invention and should be covered within the scope of the claims of this invention.
Claims
1. A training data acquisition device for an AI multiphase flow detection system, characterized in that, include: The overall frame includes a support bracket for fixing the device and a support bracket for fixing the detection probe of the image acquisition module. Transparent container with mounting holes on the side wall; The motion module includes a vertical motion module that controls the up-and-down movement of the rotary drum assembly and a rotary motion module that drives the rotary drum assembly to rotate. The rotating drum assembly, housed within a transparent container and installed with the motion module, consists of several rotating drums of different diameters and bearing different identification patterns, used to simulate dispersed phases at different velocities. The image acquisition module, used to acquire multiphase flow image information containing the identification pattern on the rotating drum, is installed in the mounting hole and includes an industrial camera, fiber optic sensor and laser sensor; The data processing module is used to identify the dispersed phase in the multiphase flow image information and obtain the image information of the simulated rotating drum. The image information of the simulated rotating drum is used as a benchmark to correct the detection results, and finally the training data of the AI multiphase flow detection system that can be used for intelligent detection is obtained.
2. The training data acquisition device for an AI multiphase flow detection system according to claim 1, characterized in that, The vertical motion module includes a stepper motor, a lead screw, and a slide table; the lead screw is vertically fixed to the device bracket; the output end of the stepper motor is mounted on the lead screw; the slide table is slidably mounted on the lead screw and has a connecting plate extending horizontally upward toward the transparent container; the rotary motion module includes a rotary motor and a coupling; the rotary motor is mounted on the connecting plate, and its output end is downwardly mounted to the rotating drum assembly through the coupling.
3. The training data acquisition device for an AI multiphase flow detection system according to claim 2, characterized in that, The identification pattern on the rotating drum consists of several micropores; the diameter of the micropores is 100 to 1000 μm.
4. The training data acquisition device for an AI multiphase flow detection system according to claim 3, characterized in that, The transparent container is a transparent plexiglass container, and a laser sensor, a fiber optic sensor, and an industrial camera are installed at the mounting holes.
5. The training data acquisition device for an AI multiphase flow detection system according to claim 4, characterized in that, Limit switches are installed at both ends of the lead screw.
6. The training data acquisition device for an AI multiphase flow detection system according to claim 5, characterized in that, Use this device to acquire training data by following these steps: S1. Install the selected rotating drum to the rotary motor via a coupling, and install the rotary motion module with the integrated rotating drum onto the slide of the vertical motion module; install the image acquisition module onto the mounting hole on the side wall of the transparent container; inject working fluid into the transparent container; S2. Place a ruler in a transparent container, use the installed image acquisition module to acquire the image of the ruler, and establish the proportional relationship between the image pixels and the actual physical size through the data processing module. S3. Based on the target motion state of the dispersed phase to be simulated, set the speed of the stepper motor and the rotary motor so that the drum can rotate and move vertically in the working fluid at the same time. The multiphase flow image information of the micro-hole pattern on the rotating drum is acquired synchronously through the image acquisition module; S4. Process the acquired multiphase flow image information and identify the micropore pattern on the rotating drum; Based on the proportional relationship established in step S2, the true size of the dispersed phase simulated by each micropore is calculated; The true velocity of the simulated dispersed phase is calculated based on the displacement and time interval of the same identification pattern in consecutive frame images. S5. Save the dispersed phase data labeled with the true size and true motion velocity obtained in step S4 as a dataset for training the AI model; and use the true size and true motion velocity as benchmark values to compare with the measurement values of other sensors under the same conditions to calculate the correction coefficient of the sensor.