Non-contact liquid level identification system integrating AI vision and thermal imaging

Abstract

The invention discloses a non-contact liquid level recognition system integrating AI vision and thermal imaging, and belongs to the technical field of industrial detection and measurement. In order to overcome the defects that a traditional contact type liquid level meter is prone to corrosion and needs to be installed in a production halt mode, and a pure vision scheme is affected by light, weather and media, the system achieves high-precision monitoring through cooperation of hardware and an AI algorithm. Hardware comprises a dual-spectrum white light high-definition and thermal imaging integrated camera (4 million pixel CMOS, 384 * 288 thermal imaging sensor), NVIDIAJetson edge computing equipment (wide temperature operation at-40 DEG C to 85 DEG C) and a pressure and temperature sensor, protection reaches IP67, ExdIICT6 anti-explosion certification is achieved, and external mounting type non-stop production installation is adopted. The AI algorithm architecture integrates U-Net + + liquid level segmentation, dynamic calibration, GAN anti-interference and multi-sensor fusion technologies. The measuring precision of the system is + / -0.3% FS, all-weather work can be achieved, the functions of leakage detection, sludge interface recognition and the like are expanded, the system is matched with dangerous chemical storage tanks, offshore oil platforms and other scenes, the safety level can be improved, the maintenance cost is reduced, and the liquid level monitoring requirements of multiple industries are met.

Description

Technical Field

[0001] This invention relates to the field of industrial inspection and measurement technology, specifically a non-contact liquid level identification system that integrates AI vision and thermal imaging. Background Technology

[0002] In industries such as petroleum, chemical, food, and wastewater treatment, accurate and reliable real-time monitoring of liquid levels in containers such as storage tanks, reactors, and settling tanks is crucial for ensuring production safety and improving automation levels. Traditional contact-type level gauges (such as float-type, pressure-type, and radar-type) require contact with the medium or installation through openings in the tank body, which poses risks of leakage, susceptibility to medium corrosion, and the need for production shutdowns for installation and maintenance.

[0003] Non-contact visual inspection solutions offer a new approach to address this issue. However, existing pure vision-based solutions face numerous challenges in application: changes in ambient light (such as at night), adverse weather conditions (rain, fog, steam), complex media properties (transparent liquids, foam, suspended matter), and deformation caused by pressure / temperature changes inside the tank can all severely affect the accuracy and stability of measurements. Relying solely on visible light cameras makes it difficult to operate reliably in all weather conditions.

[0004] Therefore, there is an urgent need for a non-contact liquid level identification system that can overcome the above-mentioned defects and achieve high precision, high robustness, and installation-free operation. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a non-contact liquid level recognition system that integrates AI vision and thermal imaging. Through dual-spectral imaging, edge computing, AI algorithms, and multi-sensor fusion technology, it achieves accurate, stable, and non-contact liquid level measurement in various complex industrial environments. This solves the problem that changes in ambient light (such as at night), severe weather (rain, fog, steam), complex media properties (transparent liquids, foam, suspended matter), and deformation caused by pressure / temperature changes inside the tank can severely affect the accuracy and stability of the measurement. Relying solely on visible light cameras makes reliable operation under all weather conditions and operating conditions difficult.

[0007] (II) Technical Solution

[0008] To achieve the aforementioned goal of accurate, stable, and non-contact measurement of liquid levels in various complex industrial environments through dual-spectral imaging, edge computing, AI algorithms, and multi-sensor fusion technology, this invention provides the following technical solution:

[0009] A non-contact liquid level recognition system integrating AI vision and thermal imaging includes two parts: hardware components and AI algorithm architecture. The hardware components include at least a dual-spectrum white light high-definition + thermal imaging integrated camera, an edge computing device, a pressure and temperature sensor, and a liquid surface boundary perception computing module. The AI ​​algorithm architecture includes at least an image acquisition unit, a preprocessing module, a recognition engine, a liquid level calculation unit, and a 3D liquid level reconstruction unit.

[0010] The dual-spectrum white light high-definition + thermal imaging integrated camera is used to acquire white light high-definition images and thermal imaging images of the target tank.

[0011] The edge computing device is used to run the AI ​​algorithm architecture to achieve real-time processing of image data;

[0012] The pressure and temperature sensor is used to collect pressure and temperature data of the target tank.

[0013] The liquid surface boundary perception and calculation module is used to receive image data from the dual-spectrum camera and sensing data from the pressure and temperature sensors. Through the liquid surface segmentation model and the calibration object recognition model in the recognition engine, it completes the extraction of liquid surface edges and the recognition of reference objects. After being processed by the liquid level calculation unit and the 3D liquid level reconstruction unit, the liquid level recognition result is output.

[0014] Preferably, the white light high-definition camera in the dual-spectrum white light high-definition + thermal imaging integrated camera adopts a global shutter CMOS sensor with a resolution of 4 megapixels or higher and is equipped with a wide-angle lens or zoom lens; its night thermal imaging system has anti-strong light interference, infrared night vision and anti-steam interference functions, and the thermal imaging sensor has a pixel of 384*288, a response band of 8-14μm, a focal length of 15mm, and supports nine color palettes including iron red, black and white, and color.

[0015] Preferably, the edge computing device is an NVIDIA Jetson series device that supports a wide operating temperature range of -40℃ to 85℃.

[0016] Preferably, the hardware component further includes a protective structure with an IP67 protection rating and ExdIICT6 explosion-proof certification, suitable for installation and use in hazardous areas; the hardware component adopts an external installation method, which does not require drilling holes in the target tank and supports installation without production stoppage.

[0017] Preferably, the AI ​​algorithm architecture also includes dynamic calibration technology and anti-interference mechanism; the dynamic calibration technology corrects lens distortion in real time through a calibration plate and can adapt to tank deformation caused by temperature or pressure; the anti-interference mechanism uses a GAN generative adversarial network to simulate extreme environments such as rainstorms and dense fog to improve the robustness of the recognition engine model.

[0018] Preferably, the liquid level boundary sensing and calculation module has a multi-sensor fusion function, which integrates the pressure and temperature data collected by the pressure and temperature sensors to compensate for liquid level measurement errors caused by changes in medium density.

[0019] Preferably, the liquid level segmentation model in the recognition engine adopts the U-Net++ segmentation network, and the liquid level measurement error of the network is <0.5%FS; the liquid level measurement accuracy of the system is ±0.3%FS.

[0020] Preferably, the system also has extended functions, including leak detection, impurity identification, sludge interface identification, and foam layer thickness analysis; wherein, the sludge interface identification function is used in wastewater treatment sedimentation tank scenarios and can trigger automatic sludge discharge operation, and the foam layer thickness analysis function is used in food industry fermentation tank scenarios and is adapted to the measurement requirements of sterile environments.

[0021] Preferably, the hardware components further include a server, a hard disk recorder, and an aggregation switch; the server is configured with an Intel i7-12400 CPU, 32GB of RAM, a 512GB solid-state drive, an 8TB hard disk drive, and an RTX-3070 dedicated graphics card; the hard disk recorder has a 12TB hard drive storage capacity and supports H.264 / H.265 / Smart264 / Smart265 decoding formats; the aggregation switch is a 24-port gigabit layer 3 managed switch.

[0022] Preferably, the system also features customized designs for specific application scenarios: when used in hazardous chemical storage tank scenarios, it is equipped with an electrostatic liquid level fluctuation suppression algorithm; when used in offshore oil platform scenarios, it has salt spray corrosion resistance and supports remote monitoring via satellite communication; when used in open-air cement storage tanks and mineral medium liquid level scenarios, it supports complex medium liquid level identification and remote display.

[0023] (III) Beneficial Effects

[0024] Compared with existing technologies, this invention provides a non-contact liquid level recognition system that integrates AI vision and thermal imaging, which has the following advantages:

[0025] 1. Non-contact and installation-free: External installation, no drilling required, no risk of leakage, and can be deployed without interrupting production.

[0026] 2. All-weather operation: It integrates white light and thermal imaging dual spectra, overcoming the problem of pure visible light solutions failing in environments such as nighttime, steam, and strong light.

[0027] 3. High precision and high robustness: Through AI algorithms, multi-sensor fusion and dynamic calibration technology, interference caused by changes in the environment, medium and tank itself is effectively compensated, achieving a high measurement accuracy of ±0.3% FS.

[0028] 4. Feature-rich: It can not only measure liquid level, but also be expanded to include multiple functions such as leak detection, foam identification, and sludge interface identification to meet the personalized needs of different industries.

[0029] 5. Safe and reliable: Equipped with high-level explosion-proof and protection certifications, suitable for various harsh and hazardous industrial environments. Attached Figure Description

[0030] Figure 1 This is a diagram illustrating the overall architecture and workflow of the system of the present invention;

[0031] Figure 2 This is a schematic diagram of the structure of the dual-spectrum integrated camera of the present invention;

[0032] Figure 3 This is a flowchart illustrating the liquid surface recognition and calculation process within the AI ​​algorithm architecture of this invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] like Figure 1-3 As shown, a non-contact liquid level recognition system integrating AI vision and thermal imaging is characterized by comprising two parts: hardware components and AI algorithm architecture.

[0035] The hardware components include:

[0036] Dual-spectrum white light high-definition + thermal imaging integrated camera: used to simultaneously acquire white light high-definition images and thermal images of the target tank. Its white light component uses a CMOS sensor with a global shutter of 4 megapixels or higher, equipped with a wide-angle or zoom lens; the thermal imaging component features strong light interference resistance, infrared night vision, and anti-steam interference capabilities. The sensor has a pixel count of 384*288, a wavelength response of 8-14μm, a focal length of 15mm, and supports multiple color palettes.

[0037] Edge computing devices: Utilizing NVIDIA Jetson series devices, supporting a wide operating temperature range (-40℃-85℃), used to deploy and run AI algorithm architectures to achieve real-time processing and analysis of image data.

[0038] Pressure and temperature sensors: used to collect internal pressure and temperature data of the target tank, providing compensation parameters for liquid level calculation.

[0039] Liquid Surface Boundary Sensing Computation Module: This module receives image data and sensor data and is the functional module that executes the core algorithm.

[0040] Protective Structure: Provides protection for the system hardware with an IP67 protection rating and Ex d IIC T6 explosion-proof certification, ensuring safe use in hazardous areas. The entire system adopts an external installation method, requiring no drilling into the target tank and supporting installation without production downtime.

[0041] Auxiliary equipment includes servers (configured with Intel i7-12400 CPU, 32GB RAM, RTX-3070 graphics card, etc.), hard disk recorders (12TB storage, supporting multi-channel decoding and playback), and aggregation switches (24-port gigabit managed type) for data storage, backup, display, and network management.

[0042] AI algorithm architecture includes:

[0043] Image acquisition unit: controls the dual-spectrum camera to acquire images.

[0044] Preprocessing module: Performs preprocessing operations such as noise reduction and enhancement on the acquired images.

[0045] Recognition Engine: Built-in liquid surface segmentation model and reference object recognition model based on U-Net++ segmentation network, used to accurately extract liquid surface edges and identify reference objects, with a liquid level measurement error of <0.5% FS.

[0046] Liquid level calculation unit: Calculates the liquid level height by combining the size of the calibration object and the geometric relationship of the image.

[0047] 3D Liquid Level Reconstruction Unit: Generates a three-dimensional model of the liquid surface, providing more intuitive liquid level information.

[0048] Dynamic calibration technology: It corrects lens distortion in real time through a calibration plate and can adapt to tank deformation caused by temperature or pressure changes, ensuring the accuracy of the measurement reference.

[0049] Anti-interference mechanism: The model is trained by simulating extreme environments such as rainstorms and dense fog using a generative adversarial network (GAN), which greatly improves the robustness of the recognition engine under harsh conditions.

[0050] Multi-sensor fusion function: Integrates pressure and temperature data, and uses algorithms to compensate for liquid level measurement errors caused by changes in medium density, ultimately achieving an overall system measurement accuracy of ±0.3% FS.

[0051] In addition, the system boasts powerful expansion capabilities, including leak detection, impurity identification, sludge interface identification (which can be used in wastewater treatment and trigger automatic sludge discharge), and foam layer thickness analysis (suitable for aseptic environments in the food industry). The system supports deep customization, such as configuring electrostatic liquid level fluctuation suppression algorithms for hazardous chemical storage tanks; adding salt spray corrosion resistance and satellite communication capabilities to offshore platform equipment; and optimizing algorithms for complex media identification scenarios such as mineral slurries.

[0052] The system of this invention is deployed on-site as follows: A dual-spectrum integrated camera, protective housing, and edge computing device are integrated into one unit and fixedly installed at an appropriate position on the side of the target tank using a bracket, ensuring that its field of view clearly covers the liquid surface inside the tank and pre-set calibration objects (such as a scale float). Pressure and temperature sensors are installed on corresponding interfaces on the tank according to the characteristics of the medium. All devices are connected to the industrial network through a convergence switch to communicate with the backend server and hard disk recorder.

[0053] During system operation, a dual-spectrum camera simultaneously acquires white light and thermal imaging video streams from the tank, which are then transmitted to the edge computing device via a switch. The AI ​​algorithm architecture within the edge computing device then starts running. The image acquisition unit obtains the video stream, and the preprocessing module performs image enhancement and filtering.

[0054] Subsequently, the liquid surface segmentation model (U-Net++ network) in the recognition engine analyzes the preprocessed image to accurately segment the liquid surface edges. Simultaneously, the calibration object recognition model identifies reference objects in the image. The liquid level calculation unit calculates the current liquid level height based on the actual size of the reference object and its pixel position in the image, combined with camera calibration parameters. The 3D liquid level reconstruction unit further generates the three-dimensional shape of the liquid surface.

[0055] Throughout the process, dynamic calibration technology continuously operates to correct deviations caused by the lens and temperature. Real-time data collected by pressure and temperature sensors is fed into the liquid level perception and calculation module. A multi-sensor fusion algorithm performs density compensation on the visually calculated liquid level value, ultimately outputting a liquid level value with an accuracy of ±0.3% FS. The results can be transmitted via network to the central control room for display and recording.

[0056] For scenarios requiring expanded functionality, the system can load corresponding AI models. For example, in a wastewater treatment plant's settling tank, the system can additionally run a sludge interface recognition model to monitor sludge thickness while monitoring the liquid level, and output a signal to control the sludge discharge pump when a threshold is reached.

[0057] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A non-contact liquid level identification system integrating AI vision and thermal imaging, characterized in that, It comprises two parts: hardware components and AI algorithm architecture. The hardware components include at least a dual-spectrum white light high-definition + thermal imaging integrated camera, an edge computing device, a pressure and temperature sensor, and a liquid surface boundary perception computing module. The AI ​​algorithm architecture includes at least an image acquisition unit, a preprocessing module, a recognition engine, a liquid level calculation unit, and a 3D liquid level reconstruction unit. The dual-spectrum white light high-definition + thermal imaging integrated camera is used to acquire white light high-definition images and thermal imaging images of the target tank. The edge computing device is used to run the AI ​​algorithm architecture to achieve real-time processing of image data; The pressure and temperature sensor is used to collect pressure and temperature data of the target tank. The liquid surface boundary perception and calculation module is used to receive image data from the dual-spectrum camera and sensing data from the pressure and temperature sensors. Through the liquid surface segmentation model and the calibration object recognition model in the recognition engine, it completes the extraction of liquid surface edges and the recognition of reference objects. After being processed by the liquid level calculation unit and the 3D liquid level reconstruction unit, the liquid level recognition result is output.

2. The non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The dual-spectrum white light high-definition + thermal imaging integrated camera uses a global shutter CMOS sensor with over 4 megapixels and is equipped with a wide-angle lens or zoom lens. Its night thermal imaging system has strong light interference resistance, infrared night vision and anti-steam interference functions. The thermal imaging sensor has a pixel size of 384*288, a response band of 8-14μm, a focal length of 15mm, and supports nine color palettes including iron red, black and white, and color.

3. The non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The edge computing device uses NVIDIA Jetson series devices that support a wide operating temperature range of -40℃ to 85℃.

4. The non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The hardware components also include a protective structure with an IP67 protection rating and ExdIICT6 explosion-proof certification, making it suitable for installation and use in hazardous areas. The hardware components are installed externally, eliminating the need to drill holes in the target tank and supporting installation without production stoppage.

5. A non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The AI ​​algorithm architecture also includes dynamic calibration technology and anti-interference mechanism; the dynamic calibration technology corrects lens distortion in real time through a calibration plate and can adapt to tank deformation caused by temperature or pressure; the anti-interference mechanism uses GAN generative adversarial network to simulate extreme environments such as rainstorms and dense fog to improve the robustness of the recognition engine model.

6. The non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The liquid level boundary sensing and computing module has a multi-sensor fusion function. By integrating the pressure and temperature data collected by the pressure and temperature sensors, it compensates for liquid level measurement errors caused by changes in medium density.

7. A non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The liquid level segmentation model in the recognition engine uses a U-Net++ segmentation network, and the liquid level measurement error of this network is <0.5%FS; the liquid level measurement accuracy of the system is ±0.3%FS.

8. A non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The system also has extended functions, including leak detection, impurity identification, sludge interface identification, and foam layer thickness analysis. The sludge interface identification function is used in wastewater treatment sedimentation tanks and can trigger automatic sludge discharge. The foam layer thickness analysis function is used in food industry fermentation tanks and is adapted to aseptic environment measurement requirements.

9. A non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, The hardware components also include a server, a hard disk recorder, and an aggregation switch; the server is configured with an Intel i7-12400 CPU, 32GB of RAM, a 512GB solid-state drive, an 8TB hard disk drive, and an RTX-3070 dedicated graphics card; the hard disk recorder has a 12TB hard drive storage capacity and supports H.264 / H.265 / Smart264 / Smart265 decoding formats; the aggregation switch is a 24-port gigabit layer 3 managed switch.

10. A non-contact liquid level identification system integrating AI vision and thermal imaging according to claim 1, characterized in that, For specific application scenarios, the system also features customized designs: when used in hazardous chemical storage tank scenarios, it is equipped with an electrostatic liquid level fluctuation suppression algorithm; when used in offshore oil platform scenarios, it has salt spray corrosion resistance and supports remote monitoring via satellite communication; when used in open-air cement storage tanks and mineral medium liquid level scenarios, it supports complex medium liquid level identification and remote display.

Images