A natural gas component detection system based on multi-modal fusion

Abstract

This invention relates to the field of natural gas detection and discloses a natural gas component detection system based on multimodal fusion. The system includes a test chamber, a multimodal hardware sensing network, an internal main control signal board, a multi-channel data synchronization and preprocessing platform, and a multimodal fusion inversion module. A piston driven by a motor is installed within the sealed test chamber to dynamically adjust the sound path distance and gas volume. The multimodal sensing network integrates a through-beam ultrasonic transducer, a fiber optic microphone, a non-dispersive infrared sensor, a thermal conductivity sensor, and temperature, humidity, and pressure sensors. The internal main control signal board synchronously acquires low-frequency photothermal and environmental signals and adds timestamps; a high-speed data acquisition card independently captures high-frequency acoustic waveforms. The host computer receives data in parallel through dual channels and performs timestamp alignment. After constructing a multimodal feature matrix, the data is sent to the fusion inversion module, which outputs the concentrations of each component of the natural gas, achieving collaborative acquisition and dynamic calibration of multi-dimensional features.

Description

Technical Field

[0001] This invention relates to the field of natural gas detection, and more specifically to a natural gas component detection system based on multimodal fusion. Background Technology

[0002] As a key clean energy source, natural gas faces increasing demand for high-precision online detection of multiple components such as methane, ethane, propane, and carbon dioxide in areas such as accurate calorific value measurement, safe hydrogen blending in pipelines, and optimized industrial combustion. Current technologies primarily rely on gas chromatographs or single-modal sensors for detection. While gas chromatographs offer high precision, they are expensive, complex to operate, and have long detection cycles, typically taking several minutes to over ten minutes, making them unsuitable for real-time online monitoring of natural gas pipelines. Furthermore, sensors based on single physical quantities are generally susceptible to environmental factors such as temperature and pressure, resulting in insufficient data reliability under complex operating conditions.

[0003] In specific single-mode sensing technologies, non-dispersive infrared sensors invert concentration by detecting the absorption intensity of infrared light of a gas at a specific wavelength. However, in a mixed system like natural gas, which is mainly composed of hydrocarbons, the carbon-hydrogen bond absorption lines of alkanes such as methane and ethane easily overlap and are significantly affected by changes in temperature, humidity, and pipeline pressure, leading to severe non-uniqueness and instability in single-mode optical inversion. Thermal conductivity methods use heating elements to detect the thermal conductivity of gases to estimate concentration. This method has high accuracy in pure binary gases, but the thermal conductivity differences among the multiple alkanes in natural gas are not significant, resulting in weak single-mode thermal conductivity signal responses that are easily affected by ambient temperature gradients, making high-precision differentiation difficult. Single acoustic detection methods, based on sound velocity or attenuation to reflect gas density changes, also face the problem of extreme sensitivity to temperature, humidity, and gas pressure. Furthermore, the limited differences in acoustic characteristics among natural gas homologues lead to insufficient resolution. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a natural gas component detection system based on multimodal fusion. This system overcomes the aforementioned deficiencies in existing technologies and solves problems such as the limited information dimensions, susceptibility to environmental parameter interference, and severe cross-coupling of signals from complex mixed gases in existing single-modal natural gas component detection technologies. By actively changing the state parameters of the natural gas to be tested within the test chamber, the system enables the natural gas to generate differentiated multi-physical responses during state changes. Combined with multimodal synchronous acquisition and fusion processing, the system improves the accuracy and stability of detecting complex mixed natural gas components.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A natural gas component detection system based on multimodal fusion includes: The test chamber is cylindrical with its axis set horizontally. A dynamic adjustment mechanism is provided inside the test chamber to adjust the gas volume inside the test chamber, thereby adjusting the gas pressure and sound path distance. A multimodal hardware sensing network is integrated inside and outside the test cavity, and includes at least an acoustic sensing unit, an optical sensing unit, a thermal conductivity sensing unit, and an environmental parameter sensing unit. An internal main control signal board is fixed inside the test cavity and is electrically connected to the optical sensing unit, thermal conductivity sensing unit and environmental parameter sensing unit. It synchronously collects signals from each sensing unit and outputs them after adding timestamps. A multi-channel data synchronization and preprocessing platform is deployed on a host computer, including a first data channel that is communicatively connected to the internal main control signal board and a second data channel that is communicatively connected to the acoustic sensing unit. The preprocessing platform is configured to align the timestamps of the data received from the two channels and construct a multimodal feature matrix. The multimodal fusion inversion module is deployed in the host computer or an edge computing node, connected to the preprocessing platform, receives the multimodal feature matrix, and outputs the predicted concentrations of each component of natural gas.

[0006] In this invention, preferably, the dynamic adjustment mechanism includes a piston, an axial adjustment push rod, and a driving component. The piston is slidably engaged with the inner wall of the test chamber. One end of the axial adjustment push rod is fixedly connected to the piston, and the other end of the axial adjustment push rod passes through the side wall of the test chamber and is connected to the driving component. The driving component drives the axial adjustment push rod to cause the piston to slide along the axial direction of the test chamber.

[0007] In this invention, preferably, the acoustic sensing unit includes an ultrasonic transmitting probe and an ultrasonic receiving probe. The ultrasonic transmitting probe is fixedly installed in the test chamber, and the ultrasonic receiving probe is installed on the piston via a movable bracket. When the piston moves, the ultrasonic transmitting probe and the ultrasonic receiving probe move relative to each other to adjust the sound path distance.

[0008] In this invention, preferably, the acoustic sensing unit further includes a passive fiber optic probe and a fiber optic microphone; the passive fiber optic probe is mounted on the movable bracket and is used to penetrate into the gas to be tested to capture changes in sound pressure; the fiber optic microphone is installed at the boundary of the test cavity and is connected to the passive fiber optic probe via an optical fiber, and is used to convert optical parametric signals into electrical signals and transmit them to the second data channel.

[0009] In this invention, preferably, the internal main control signal board adopts a modular architecture, and its power adapter module has a modular ground isolation design to provide the thermal conductivity sensing unit with an analog power supply independent of the digital power supply; the internal main control signal board integrates a microcontroller module and a serial communication module, the microcontroller is configured to perform digital filtering and moving average processing on the collected signal and then package it into a data frame, and the serial communication module is used to transmit the data frame to the first data channel.

[0010] In this invention, preferably, the first data channel is a serial port acquisition link, configured to receive low-frequency data from the internal main control signal board, the low-frequency data including the light absorptivity output by the optical sensing unit, the mixed thermal conductivity output by the thermal conductivity sensing unit, and the temperature, humidity, and pressure data output by the environmental parameter sensing unit; the second data channel is a direct memory access (DAQ) acquisition link, configured to capture high-frequency acoustic waveform data output by the acoustic sensing unit without obstruction.

[0011] In this invention, preferably, the low-frequency data received by the first data channel is parsed and stored in the serial port data storage unit, the high-frequency acoustic waveform data stream captured by the second data channel is disked to the NI data storage unit, and then the data in the two storage units are spliced ​​into an aligned multimodal feature matrix according to a unified timestamp.

[0012] In this invention, preferably, the test chamber is provided with at least one air inlet port, one purge port and one vacuum exhaust port; the vacuum exhaust port is used to connect to an external vacuum pump, the purge port is used to connect to a high-purity nitrogen source, and the air inlet port is used to connect to an external mass flow controller to inject the natural gas to be tested according to a preset ratio.

[0013] In this invention, preferably, the microcontroller of the internal main control signal board also integrates a multi-source synchronous triggering unit. The multi-source synchronous triggering unit is configured to simultaneously trigger the excitation pulse of the ultrasonic transmitting probe, the light source modulation signal of the optical sensing unit, and the sample-and-hold circuit of the thermal conductivity sensing unit with the same hardware clock source, and write the triggering time as a timestamp reference into the data frame header.

[0014] In this invention, preferably, the system is configured to execute a baseline calibration mode, a dynamic detection mode, or a cleaning and reset mode. In the baseline calibration mode, the test chamber is evacuated to a vacuum state through the vacuum exhaust port, and then high-purity nitrogen is introduced through the purge port. The drive motor moves the piston to multiple preset sound path nodes, and the ultrasonic flight time is recorded synchronously. The inherent time delay of the system is eliminated by multi-node differential calculation. In dynamic detection mode, natural gas to be tested is injected through the air inlet. After the valve is closed, the drive motor pushes the piston to compress the gas to simulate the pressure rise of the pipeline network. During the compression process, data from the acoustic sensing unit, optical sensing unit, thermal conductivity sensing unit and environmental parameter sensing unit are collected simultaneously. In the cleaning and reset mode, after the test is completed, the drive piston returns to the maximum volume position, and high-purity nitrogen is introduced through the purge port until the baseline readings of each sensor return to zero.

[0015] The beneficial effects of this invention are: 1. This invention constructs a multimodal hardware sensing network integrating acoustics, optics, thermal conductivity, and environmental sensing. Within a single sealed test chamber, it achieves delay-free collaborative acquisition of multiple dimensions of natural gas characteristics, including sound velocity, sound attenuation, infrared absorptivity, thermal conductivity, and temperature, humidity, and pressure. Through a separate design of the through-beam ultrasonic probe and fiber optic microphone, the physical decoupling of sound velocity and sound attenuation is achieved. Simultaneously, the fiber optic sensing architecture ensures intrinsic safety, explosion protection, and electromagnetic interference resistance in hazardous gas environments, providing a reliable physical platform for high-precision detection in complex natural gas mixtures. This invention does not simply combine multiple sensors; rather, it simultaneously acquires acoustic, optical, thermal conductivity, and environmental parameter responses during the same dynamic pressure excitation process. This allows different physical modes to jointly describe the response patterns of natural gas components as pressure changes, improving the ability to distinguish features among complex mixed natural gas components. 2. This invention designs a sealed test chamber with dynamic pitch and pressure variation functions. By driving a piston to translate via a drive motor, the sound path distance of the through-beam ultrasonic probe can be continuously adjusted, thereby eliminating the inherent time delay of the system and improving the accuracy of sound velocity inversion through multi-node differential calculation. It can also change the gas volume within the chamber to simulate various pressure conditions of natural gas pipelines from atmospheric pressure to high pressure. This design enables the system to perform online baseline calibration, pressure coupling characteristic testing, and deep cleaning and reset, significantly improving the calibration accuracy, environmental adaptability, and long-term stability of the detection system. Compared with methods using multiple fixed pressure states for detection, this invention continuously changes the pressure state through dynamic volume variation within a single sealed chamber, allowing the detection system to obtain a continuous pressure response curve, rather than a limited number of measurement points under discrete pressure states, thereby improving the uniqueness of component inversion. 3. This invention proposes a preprocessing architecture with dual-channel parallel acquisition and timestamp alignment. The first channel acquires low-frequency data such as optical, thermal conductivity, and environmental data via serial port, while the second channel captures high-frequency waveform data from ultrasonic waves and fiber optic microphones without obstruction using a high-speed data acquisition card. This achieves physical isolation and synchronous transmission between high- and low-frequency heterogeneous sensors. The host computer strictly matches the two data streams based on a unified timestamp, eliminating phase misalignment caused by differences in the hardware response times of different sensors. This provides a high-quality, time-aligned multimodal input matrix for the backend fusion algorithm. By using a unified hardware time reference to correlate sensors with different sampling frequencies, data mismatch caused by response delays of different sensors during dynamic pressure changes is avoided, enabling the fusion model to analyze multiple physical responses under the same state.

[0016] 4. The system of this invention has a high degree of modularity and adjustable parameters, possessing a complete detection workflow. The system automatically executes baseline calibration mode, dynamic detection mode, and cleaning and reset mode, forming a fully automatic closed-loop operation from vacuuming and nitrogen filling, multi-point sound path calibration, to dynamic pressure synchronous acquisition, and then to nitrogen purging and baseline zeroing. This not only reduces errors caused by manual intervention but also ensures the consistency of the system's initial state before each test, effectively eliminating the memory effect of gas sensors.

[0017] 5. The system of this invention integrates a multimodal fusion inversion module at the top level, which can receive the preprocessed multimodal feature matrix and output the predicted concentrations of each component of natural gas. This module is compatible with various fusion algorithms, enabling the system to have a complete closed-loop capability from bottom-level physical perception to top-level intelligent decision-making, providing a high-precision, real-time, and robust detection solution for natural gas calorific value measurement, hydrogen blending safety monitoring, and combustion optimization in industrial settings. Attached Figure Description

[0018] Figure 1 This is a schematic cross-sectional view of the test chamber in this invention; Figure 2 This is a schematic diagram of the connection structure of the internal main control signal board in this invention; Figure 3 This is a flowchart of the overall data flow and processing architecture of the system in this embodiment of the invention. Reference numerals in the attached diagram: 1. Test chamber; 2. Ultrasonic transmitting probe; 3. Internal main control signal board; 4. Host computer; 5. Fiber optic probe; 6. Ultrasonic receiving probe; 7. High-speed data acquisition card; 8. Axial adjustment push rod; 9. Drive motor; 10. Pressure transmitter; 11. First air inlet port; 12. Second air inlet port; 13. Purge port; 14. Vacuum exhaust port; 15. Fiber optic microphone; 16. Piston. Detailed Implementation

[0019] 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.

[0020] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Please also see Figures 1 to 3 The natural gas component detection system based on multimodal fusion provided by the present invention includes a sealed test chamber 1, a multimodal hardware sensor acquisition network, an internal main control signal board 3, a multi-channel data synchronization and preprocessing platform, and a multimodal fusion inversion module.

[0023] The test chamber 1 adopts a horizontal cylindrical structure and is made of metal. An axial adjustment push rod 8, driven by a drive motor 9, is located on the right side of the test chamber 1. The push rod penetrates the right side wall of the test chamber 1 and is rigidly connected to the piston 16 inside. A dynamic sealing ring is installed on the edge of the piston 16 to ensure absolute physical isolation between the inside and outside of the test chamber 1 during the sliding process of the piston 16. The drive motor 9, in conjunction with the transmission mechanism, drives the piston 16 to translate along the axis of the test chamber 1. The movement of the piston 16 can synchronously change the physical position of the receiver probe mounted on it, thereby achieving continuous and stepwise adjustment of the sound wave flight distance, i.e., the sound path distance. Simultaneously, with the test chamber 1 completely sealed and the total gas volume fixed, the translation of the piston 16 can change the gas volume inside the chamber, achieving dynamic rise and fall of gas pressure, simulating different pressure conditions of a natural gas pipeline network.

[0024] A multimodal hardware sensing network is integrated inside and outside the test chamber 1, including an acoustic sensing unit, an optical sensing unit, a thermal conductivity sensing unit, and an environmental parameter sensing unit. The acoustic sensing unit comprises an ultrasonic transmitting probe 2 fixedly mounted in the stationary area on the left side of the test chamber 1, and an ultrasonic receiving probe 6 mounted on a movable bracket on the right side. The ultrasonic transmitting probe 2 and the ultrasonic receiving probe 6 are positioned opposite each other, forming a complete through-beam acoustic detection link, specifically used to accurately pinpoint the time of flight of sound waves to invert macroscopic sound speed. In addition, the acoustic sensing unit also includes a passive fiber optic probe 5 and a fiber optic microphone 15. The passive fiber optic probe 5 is a fiber optic sensor that requires no power supply and senses changes in sound pressure solely through the principle of light wave interference. The passive fiber optic probe 5 is also mounted on the movable bracket on the right side, extending into the mixed gas mass with the bracket, and uses the principle of light wave interference to capture the weak sound pressure changes caused by the molecular relaxation effect of natural gas components. The passive fiber optic probe 5 transmits the sensed optical parameter signals through optical fiber to the fiber optic microphone 15 installed at the boundary of the test chamber 1. The fiber optic microphone 15 performs photoelectric demodulation and signal amplification, and transmits the demodulated analog electrical signal to the external high-speed data acquisition card 7. This fiber optic sensing architecture, which separates the probe and microphone, achieves physical decoupling of sound velocity and sound attenuation characteristics at the underlying level, while ensuring intrinsically safe explosion protection and electromagnetic interference resistance in hazardous gas spaces. Intrinsically safe explosion protection limits the energy of the circuit so that the electrical sparks and thermal effects generated under normal operation or specified fault conditions cannot ignite explosive gas mixtures.

[0025] The optical sensing unit employs a non-dispersive infrared sensor, detecting wavelengths including the characteristic absorption wavelength of the target gas and a reference wavelength. The detection optical path length can be adjusted via the position of piston 16. The thermal conductivity sensing unit uses a thermal conductivity chip based on a Wheatstone bridge structure. The environmental parameter sensing unit includes a temperature sensor, a humidity sensor, and a pressure transmitter 10. All of the aforementioned optical sensing unit, thermal conductivity sensing unit, and environmental parameter sensing unit are integrated onto the internal main control signal board 3.

[0026] The internal main control signal board 3 is fixed in the static area inside the test chamber 1 and adopts a modular architecture design. Its power adapter module uses an analog-to-digital ground isolation design to provide low-ripple analog power to the thermal conductivity sensing unit. The internal main control signal board 3 integrates a microcontroller module and a serial communication module. The microcontroller, in conjunction with the analog-to-digital converter, is responsible for synchronously acquiring the light absorptivity output from the optical sensing unit, the bridge differential signal output from the thermal conductivity sensing unit, and the temperature, humidity, and pressure data output from the environmental parameter sensing unit. The microcontroller performs digital filtering and data packaging, adding a unified timestamp to the heterogeneous sensing data. The serial communication module is responsible for outputting the processed data frames through a serial signal.

[0027] A multi-channel data synchronization and preprocessing platform is deployed on the host computer 4. This platform includes a first data channel communicating with the serial communication module of the internal main control signal board 3, and a second data channel communicating with the high-speed data acquisition card 7. The first data channel is a serial acquisition link used to receive low-frequency data from the internal main control signal board 3. This low-frequency data includes the light absorptivity output by the non-dispersive infrared sensor, the mixed thermal conductivity output by the thermal conductivity sensor, and temperature, humidity, and pressure data. The second data channel is a data acquisition link used to capture high-frequency acoustic waveform data output by the ultrasonic transducer and the fiber optic microphone 15. The preprocessing platform stores the low-frequency data received by the first data channel into the serial data storage unit and the high-frequency acoustic waveform data captured by the second data channel into the high-speed data storage unit. Then, the preprocessing platform concatenates the data from the two storage units into an aligned multimodal feature matrix based on a unified timestamp.

[0028] Because the internal state parameters of the test chamber continuously change during dynamic detection, and different sampling times correspond to different gas detection states, the multi-channel data synchronization and preprocessing platform not only needs to achieve data synchronization between different sensing modes, but also needs to maintain the temporal correlation between the detection data and the corresponding state change process. Specifically, the first data channel acquires low-frequency detection data output by the optical sensing unit, thermal conductivity sensing unit, and environmental parameter sensing unit, while the second data channel acquires high-frequency detection data output by the acoustic sensing unit. The multi-channel data synchronization and preprocessing platform correlates data from different sampling frequencies based on a unified time reference, forming a multi-modal temporal feature matrix corresponding to the continuous state change process.

[0029] The multimodal fusion and inversion module is deployed on the host computer or an edge computing node and connected to the preprocessing platform. This module receives the multimodal feature matrix output from the preprocessing platform and uses an ensemble learning model based on feature decoupling and dynamic weights to output the predicted concentrations of each component of natural gas. In this embodiment, any fusion algorithm capable of mapping from the multimodal feature matrix to gas component concentrations can be deployed in this module.

[0030] The following section provides a complete explanation of the overall usage process, combining the system's three working modes.

[0031] In baseline calibration mode, the system first opens the vacuum exhaust port 14 to evacuate the test chamber 1 to a vacuum state. Then, high-purity nitrogen gas is introduced through the purge port 13 to atmospheric pressure as the background reference gas. Under atmospheric pressure, the host computer 4 controls the piston 16 to move to multiple preset discrete distance nodes via the drive motor 9. At each distance node, the host computer 4 triggers the acquisition card to synchronously record the ultrasonic flight time and the actual position of the piston 16. A multi-point differential calculation method is used to fit the linear relationship between flight time and sound path, thereby eliminating the fixed system delay caused by the acoustic probe and the underlying hardware circuit response, ensuring the absolute accuracy of subsequent natural gas sound velocity calculations.

[0032] In dynamic detection mode, the system injects the mixed natural gas to be tested into the test chamber 1 through the first air inlet port 11, the second air inlet port 12, and an external mass flow controller, according to the target molar ratio. After closing the air inlet and outlet valves, the system is allowed to stand until the sensor readings stabilize. The host computer 4 sends a command to start the dynamic pressure change program: the drive motor 9 pushes the piston 16 to move according to the preset adjustment process, thereby continuously changing the internal volume of the test chamber 1 and the corresponding gas state parameters. The pressure transmitter 10 records the dynamic rise curve of the absolute pressure inside the chamber in real time. While the pressure is continuously rising, the system simultaneously collects data from all sensing units: the non-dispersive infrared sensor records the collision broadening phenomenon of the absorption spectrum, the thermal conductivity sensor records the baseline drift characteristics, the ultrasonic transmitting and receiving probes record the flight time and amplitude, the fiber optic probe 5 and the fiber optic microphone 15 record the change trend of the sound attenuation rate with the increase of gas density, and the temperature, humidity and pressure sensors record environmental parameters. All data are accompanied by a unified hardware timestamp. After receiving the complete data packet transmitted from the lower layer, the host computer 4 performs strict time-series matching and alignment between the low-frequency features transmitted through the serial port and the high-frequency acoustic features captured by the acquisition card based on the timestamp, constructing a multimodal feature matrix under the current operating conditions. This feature matrix is ​​sent to the multimodal fusion and inversion module, which outputs the concentration of each component. Unlike acquiring only single-state response data under fixed detection conditions, this embodiment dynamically changes the internal state parameters of the test chamber, causing the natural gas under test to produce continuously changing multi-physical responses under different state conditions, thereby obtaining continuous detection data that can characterize the process of natural gas component feature changes.

[0033] In the cleaning and reset mode, after a single test cycle, the system drives piston 16 to return to the initial maximum volume position, automatically opens the exhaust port to discharge exhaust gas, and simultaneously continuously introduces high-purity nitrogen from purge port 13 for rinsing until the baseline readings of each sensor return to zero, completely replacing the residual hydrocarbons in the chamber and eliminating the memory effect of the gas sensor.

[0034] By combining the three modes described above, the system of this invention achieves a fully automated closed-loop workflow from baseline calibration and dynamic pressurization to deep cleaning. The overall data flow of the system is as follows: Multimodal sensors at the physical layer acquire data in parallel, which is then aggregated into timestamped low-frequency data frames by the internal main control signal board 3. Simultaneously, a high-speed acquisition card independently captures high-frequency acoustic waveforms. The dual-channel parallel acquisition link of the host computer 4 receives both types of data, which are then timestamped to form a unified multimodal feature matrix. This matrix is ​​ultimately sent to the multimodal fusion inversion module, which outputs the concentration values ​​of each component of the natural gas. The entire system integrates bottom-level physical sensing with top-level intelligent inversion, providing a complete hardware and software collaborative solution for high-precision online detection of multiple components in complex natural gas mixing environments.

[0035] The overall data flow and processing architecture flowchart of the system in this embodiment can be found in [reference needed]. Figure 3 The process is strictly divided into four core stages from top to bottom, specifically including: During the experimental environment configuration phase, the system first performs a gas purging operation in the gas chamber by evacuating or purging with high-purity background gas to ensure that the baseline inside the chamber is zero and to eliminate the experimental memory effect. Subsequently, a mixed gas injection operation is performed, injecting the target natural gas components and interfering gases into the chamber according to a preset ratio to build a stable physical testing environment for subsequent multimodal detection.

[0036] In the sensor data acquisition phase, the main responsibility is to acquire the characteristics of multi-source heterogeneous gases in the physical space. Specifically, three major sensing modules are deployed in parallel at the front end of the system: the NDIR / TCS sensor module is responsible for targeted extraction of the optical absorption characteristics and mixed thermal conductivity TCS characteristics of natural gas; the temperature, humidity, and pressure sensor modules simultaneously acquire environmental reference parameters within the test chamber, providing environmental basis for subsequent multimodal compensation; and the ultrasonic sensor module operates independently, dedicated to transmitting, receiving, and extracting high-frequency acoustic waveform characteristics.

[0037] During the host computer data acquisition phase, this stage constitutes a distinctive dual-channel parallel acquisition link of the system, achieving physical isolation and synchronous transmission of high- and low-frequency data. On one hand, for low-frequency characteristic data such as NDIR, TCS, and temperature, humidity, and pressure, the system aggregates and reads them through the serial port acquisition link, and finally formats and writes them to the serial port data storage unit. On the other hand, for the massive amount of high-frequency acoustic data generated by the ultrasonic sensor, the system bypasses the conventional serial port and directly calls the high-speed DAQ data acquisition module for non-blocking capture, and streams it to the NI data storage unit.

[0038] The algorithm inversion stage is the central brain of the entire system. The host computer platform extracts and concatenates the time-aligned multidimensional feature matrices from the serial port data storage and NI data storage, and inputs them uniformly into the top-level Stacking algorithm inversion module. Through this ensemble learning model, such as combining feature decoupling and dynamic drift MSE weight allocation, the cross-interference signals in complex natural gas mixtures are deeply decoupled, and finally, high-precision and robust multi-component concentration inversion results are output, realizing closed-loop detection data.

[0039] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A natural gas component detection system based on multimodal fusion, characterized in that, include: The test chamber is equipped with a dynamic adjustment mechanism. The dynamic adjustment mechanism changes the gas volume inside the test chamber to generate a continuously changing pressure state of the natural gas to be tested, and simultaneously adjusts the acoustic path distance of the acoustic detection path. A multimodal hardware sensing network is integrated inside and outside the test cavity, including at least an acoustic sensing unit, an optical sensing unit, a thermal conductivity sensing unit, and an environmental parameter sensing unit. An internal main control signal board is fixed inside the test cavity and is electrically connected to the optical sensing unit, thermal conductivity sensing unit and environmental parameter sensing unit. It synchronously collects the continuous signals of each sensing unit during the execution of the dynamic adjustment mechanism and outputs them after adding timestamps. A multi-channel data synchronization and preprocessing platform is deployed on a host computer, including a first data channel that is communicatively connected to the internal main control signal board and a second data channel that is communicatively connected to the acoustic sensing unit. The preprocessing platform is configured to align the data received from the two channels with timestamps and construct a continuous multimodal feature matrix. The multimodal fusion inversion module is deployed in the host computer or at least one edge computing node, connected to the preprocessing platform, receives the multimodal feature matrix, and outputs the predicted concentrations of each component of natural gas.

2. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The dynamic adjustment mechanism includes a piston, an axial adjustment push rod, and a driving component. The piston is slidably engaged with the inner wall of the test chamber. One end of the axial adjustment push rod is fixedly connected to the piston, and the other end of the axial adjustment push rod passes through the side wall of the test chamber and is connected to the driving component. The driving component drives the axial adjustment push rod to drive the piston to slide along the axial direction of the test chamber, thereby changing the gas volume inside the test chamber.

3. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The acoustic sensing unit includes an ultrasonic transmitting probe and an ultrasonic receiving probe. The ultrasonic transmitting probe is fixedly installed in the test chamber, and the ultrasonic receiving probe is installed on the piston via a movable bracket. When the piston moves, the ultrasonic transmitting probe and the ultrasonic receiving probe move relative to each other to adjust the sound path distance.

4. The natural gas component detection system based on multimodal fusion according to claim 3, characterized in that, The acoustic sensing unit also includes a passive fiber optic probe and a fiber optic microphone; the passive fiber optic probe is mounted on the movable bracket and is used to penetrate into the gas to be tested to capture changes in sound pressure; the fiber optic microphone is installed at the boundary of the test cavity and is connected to the passive fiber optic probe via an optical fiber, and is used to convert optical parametric signals into electrical signals and transmit them to the second data channel.

5. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The internal main control signal board adopts a modular architecture, and its power adapter module has a modular ground isolation design to provide the thermal conductivity sensing unit with an analog power supply independent of the digital power supply. The internal main control signal board integrates a microcontroller module and a serial communication module. The microcontroller is configured to perform digital filtering and moving average processing on the collected signal and then package it into a data frame. The serial communication module is used to transmit the data frame to the first data channel.

6. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The first data channel is a serial port acquisition link, configured to receive low-frequency data from the internal main control signal board. The low-frequency data includes the light absorption rate output by the optical sensing unit, the mixed thermal conductivity output by the thermal conductivity sensing unit, and the temperature, humidity, and pressure data output by the environmental parameter sensing unit. The second data channel is a direct memory access data acquisition card acquisition link, configured to capture high-frequency acoustic waveform data output by the acoustic sensing unit.

7. The natural gas component detection system based on multimodal fusion according to claim 6, characterized in that, The multi-channel data synchronization and preprocessing platform stores the low-frequency data received by the first data channel in the first data storage unit and stores the high-frequency acoustic waveform data acquired by the second data channel in the second data storage unit. The low-frequency data and high-frequency acoustic waveform data are then spliced ​​together according to a unified timestamp to form the multimodal feature matrix.

8. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The test chamber is provided with at least one air inlet port, one purge port and one vacuum exhaust port; the vacuum exhaust port is used to connect to an external vacuum pump, the purge port is used to connect to a high-purity nitrogen source, and the air inlet port is used to connect to an external mass flow controller to inject the natural gas to be tested according to a preset ratio.

9. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The microcontroller of the internal main control signal board also integrates a multi-source synchronous triggering unit. The multi-source synchronous triggering unit is configured to simultaneously trigger the excitation pulse of the ultrasonic transmitting probe, the light source modulation signal of the optical sensing unit, and the sample-and-hold circuit of the thermal conductivity sensing unit with the same hardware clock source, and write the triggering time as a unified timestamp reference into the data frame header.

10. The natural gas component detection system based on multimodal fusion according to claim 1, characterized in that, The system is configured to perform baseline calibration mode, dynamic detection mode, or cleaning and reset mode. In the baseline calibration mode, the test chamber is evacuated to a vacuum state through the vacuum exhaust port, and then nitrogen is introduced through the purge port; the drive motor drives the piston to move to multiple preset sound path nodes, and the ultrasonic flight time is recorded simultaneously. The inherent time delay of the system is eliminated by multi-node differential calculation. In dynamic detection mode, natural gas to be tested is injected through the air inlet. After the valve is closed, the drive motor pushes the piston to compress the gas to simulate the pressure rise of the pipeline network. During the compression process, data from the acoustic sensing unit, optical sensing unit, thermal conductivity sensing unit and environmental parameter sensing unit are collected simultaneously. In the cleaning and reset mode, after the test is completed, the drive piston returns to the maximum volume position, and high-purity nitrogen is introduced through the purge port until the baseline readings of each sensor return to zero.

Images