Flame arrester and AI-based flame arresting efficiency real-time monitoring method

Abstract

The invention discloses a flame arrester and an AI-based flame arresting efficiency real-time monitoring method, and relates to the technical field of flame arresting efficiency detection. The monitoring method comprises the following steps: detecting data through the multi-mode sensor, and transmitting the data to the edge calculation module; the edge calculation module preprocesses the data and outputs a fire retardance efficiency score through an AI model; judging whether the score triggers the protection action of the dynamic suppression module or not; the edge calculation module uploads the data and updates parameters of a digital twin model in the cloud management platform; a digital twin model analyzes fault causes and optimizes working parameters. According to the invention, by combining three kinds of data of temperature, sound wave and image, the flame arresting state of the flame arrester body can be comprehensively monitored, misjudgment of a single data source can be avoided, the robustness of the monitoring system in a complex environment is improved, fusion analysis is carried out on multiple data through the AI model, and the accuracy of the monitoring system is improved. And the reliability and the accuracy of early warning of fire arresting failure of the fire arrester are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire resistance performance monitoring, and in particular to a fire arrester and a real-time monitoring method for fire resistance performance. BACKGROUND

[0002] With the large-scale development of China in the fields of petrochemical industry, new energy storage, etc., as a key safety device, the demand for real-time monitoring of the performance of the fire arrester is increasingly urgent. The current industry generally adopts the standard of GB / T13347-2010 Oil and Gas Pipeline Fire Arrester, but the standard only specifies the static performance indicators of the fire arrester and does not involve real-time health assessment under dynamic working conditions. In complex industrial scenarios, the fire arrester core is easily affected by high temperature creep, dust accumulation, etc., resulting in a decrease in porosity and a decrease in fire resistance efficiency, and the existing monitoring methods have significant defects: single parameter monitoring limitation: the mainstream differential pressure sensor (such as Rosemount 3051 series) only monitors the overall pressure drop and cannot locate local defects (such as 1-3mm microcracks at the edge of the corrugated plate affecting the gradient of the fire resistance performance). This single parameter monitoring method cannot fully reflect the health status of the fire arrester core, resulting in potential risks that cannot be discovered in time. The temperature inside the new energy storage battery module fluctuates violently (-40℃ to 120℃), the existing sensor network data loss rate is high, and there is a lack of multi-parameter fusion analysis capability. This technical shortcoming greatly reduces the reliability of the fire arrester in this complex environment.

[0003] According to industry statistics, more than 50% of the fire arrester failures in the world are caused by the fact that the performance decay is not discovered in time. Therefore, developing a real-time monitoring method for fire resistance performance that integrates multi-modal sensing and AI algorithms to realize quantitative evaluation of the health status of the fire arrester core, precise positioning of defects and life prediction has become the key to breaking through the safety bottleneck of the industry. SUMMARY

[0004] In view of the fact that the monitoring of the fire resistance performance of the fire arrester at present only relies on single data parameters for monitoring, which cannot fully reflect the fire resistance state of the fire arrester, has safety hazards, and does not have the ability of multi-parameter data fusion analysis in complex environments, the present application provides a fire arrester and a real-time monitoring method for fire resistance performance.

[0005] In order to solve the above technical problems, the present application adopts the following technical solutions: The fire arrester includes a fire arrester body, a multi-modal sensing module, an edge computing module, a dynamic suppression module, and a cloud management platform. The multi-modal sensing module and the dynamic suppression module are arranged on the fire arrester body. The edge computing module is connected with the multi-modal sensing module, the dynamic suppression module, and the cloud management platform. The dynamic suppression module includes a motor, a piston, a storage bin, and a solenoid valve. The piston is arranged in the storage bin and is in transmission connection with the motor. A plurality of metal foam particles are arranged in the storage bin and in a fire arrester section of the fire arrester body. The storage bin is in communication with the fire arrester section through a connecting channel, and the solenoid valve is arranged in the connecting channel.

[0006] Further, the multi-modal sensing module includes a fiber optic sensor, an acoustic wave detector, and an infrared thermal imager. The end of the fiber optic sensor is embedded in the fire arrester body. The acoustic wave detector is arranged on the outer wall of the fire arrester. The infrared thermal imager is adjacent to the flange and is embedded in the fire arrester body. The fire arrester body further includes an inlet section, an expanding section, a contraction section, and an outlet section. The flanges are fixed to the inlet section and the outlet section. A plurality of fiber optic sensors are arranged in the inlet section, the expanding section, the contraction section, and the outlet section, and are uniformly arranged along the axial direction of the fire arrester body. A plurality of acoustic wave detectors are fixed to the outer walls of the inlet section and the outlet section. A plurality of infrared thermal imagers are embedded in the inlet section and the outlet section, and each infrared thermal imager is provided with a high-temperature-resistant quartz window.

[0007] Further, the edge computing module adopts an industrial-grade embedded controller and integrates an FPGA acceleration unit. The operating temperature range of the edge computing module is between -40℃ and 85℃.

[0008] An AI-based real-time monitoring method for fire arrester efficiency is applied to the fire arrester as claimed in any one of the preceding claims. The method includes the following steps: S1, the multi-modal sensing module monitors the temperature, sound waves, and thermal distribution image data of the fire arrester body in real time and transmits the data to the edge computing module; S2, the edge computing module pre-processes the data and inputs the pre-processed data into an AI model to output a fire arrester efficiency score through the AI model; S3, the edge computing module determines whether the fire arrester efficiency score is lower than a preset threshold. When the score is lower than the threshold, the edge computing module triggers a protection action through the dynamic suppression module to supplement metal foam particles into the fire arrester body; S4, the edge computing module regularly uploads data packets to the cloud management platform to update the parameters of the digital twin model; and S5, the cloud management platform analyzes the causes of the fault through the digital twin model and optimizes the working parameters.

[0009] Further, in S1, the optical fiber sensor collects temperature and strain data in the flame arrester body at intervals; the acoustic wave detector records the acoustic pressure waveform of the flame arrester body in real time, performs FFT analysis to extract the energy characteristics of a specific frequency range, obtains acoustic spectrum data, and identifies the deflagration characteristics; and the infrared thermal imager captures a frame of thermal distribution image at intervals, and the edge computing module performs noise filtering and feature enhancement on the collected images.

[0010] Further, in S2, the pre-processing process of the edge computing module is to synchronize the temperature data, acoustic spectrum data and thermal distribution image stream through timestamp synchronization technology, and generate a time-space multi-dimensional feature matrix.

[0011] Further, in S2, the inference process of the AI model is to obtain the flame arrester efficiency score through the scoring formula by running the lightweight convolutional neural network on the pre-processed data, and to predict the particulate matter loss. The scoring formula is: ; In the formula, represents the average temperature; represents the peak value of acoustic pressure, represents the defect area identified by thermal imaging; represents the temperature weight coefficient, ; represents the acoustic pressure weight coefficient, ; represents the defect area weight coefficient, .

[0012] Further, in S3, the flame arrester efficiency score formula result ranges between 0 and 100%, and the threshold values include a first response threshold value and a second response threshold value, the first response threshold value is 80%, and the second response threshold value is 75%.

[0013] Further, when the flame arrester efficiency score is lower than the first response threshold value, the triggered protection action is to open the electromagnetic valve to release the metal foam in the storage bin to fill the gap at the top of the flame arrester section to absorb the detonation energy. When the flame arrester efficiency score is lower than the second response threshold value, the triggered protection action is to drive the metal foam in the storage bin into the flame arrester section according to the AI model prediction of the particulate matter loss.

[0014] Further, in S3, the metal foam particles in the storage bin and in the flame arrester section are of different types, and the different types of metal foam particles are placed in the storage bin in the up-down direction.

[0015] The beneficial effects of this invention are as follows: By combining temperature, sound wave, and image data, this invention can comprehensively and effectively monitor the overheating warning, sealing performance, and heat distribution of the flame arrester body. This not only avoids misjudgments from a single data source but also improves the robustness of the monitoring system in complex environments. Furthermore, by using an AI model to fuse and analyze multiple data sources, it significantly improves the reliability and accuracy of flame arrester failure warnings. In this invention, the edge computing module achieves millisecond-level rapid response through a lightweight model and effectively reduces errors caused by traditional mechanical compensation by using particulate matter filling. Simultaneously, it utilizes a digital twin model in the cloud management platform to analyze the causes of failures and effectively optimizes flame arresting performance through data analysis, thereby significantly improving the safety of the flame arrester. Attached Figure Description

[0016] Figure 1 The diagram shown is a schematic representation of the structural principle of the flame arrester of the present invention.

[0017] Figure 2 for Figure 1 The partial structural diagram shows the connection relationship between the dynamic suppression module and the fire arrestor section.

[0018] Figure 3 The diagram shows the overall structure of the real-time fire-retardant performance monitoring system.

[0019] Explanation of reference numerals in the attached diagram: 1. Inlet section; 2. Flame-arresting section; 3. Flame-stopping section; 4. Shrinkage section; 5. Outlet section; 6. Flange; 7. Fiber optic sensor; 8. Acoustic wave detector; 9. Infrared thermal imager; 10. Storage silo; 11. Solenoid valve; 12. Piston; 13. Motor. Detailed Implementation

[0020] This invention discloses a flame arrester, which is described below in conjunction with... Figure 1 and Figure 2 The following is a detailed description.

[0021] The fire arrester includes a fire arrester body, a multi-modal sensing module, an edge computing module, a dynamic suppression module and a cloud management platform, the fire arrester body includes an inlet section 1, an expanding section 2, a fire blocking section 3, a contraction section 4 and an outlet section 5 connected in sequence, and flanges 6 are fixedly connected to the inlet section 1 and the outlet section 5. The multi-modal sensing module includes a plurality of optical fiber sensors 7, a plurality of acoustic wave detectors 8 and a plurality of infrared thermal imagers 9, each optical fiber sensor 7 is arranged in the inlet section 1, the expanding section 2, the contraction section 4 and the outlet section 5, and each optical fiber sensor 7 is uniformly arranged along the axial direction of the fire arrester body, for real-time detection of the temperature gradient of particulate matter, and micro-strain caused by the change in the tightness of the particulate matter is detected through the Bragg grating (FBG). The temperature measurement accuracy of each optical fiber sensor 7 needs to be less than 0.5℃, and the resolution of each optical fiber sensor 7 to micro-strain is not higher than 1 με. Each acoustic wave detector 8 is fixed on the outer wall of the inlet section 1 and the outlet section 5, for real-time recording of the sound pressure waveform. The plurality of infrared thermal imagers 9 all adopt non-refrigerated embedded infrared thermal imagers 9, and are all provided with high-temperature-resistant quartz windows. Each infrared thermal imager 9 is located adjacent to the flanges 6 on both sides of the fire blocking section 3 body, and is embedded in the inlet section 1 and the outlet section 5. The edge computing module adopts an industrial-grade embedded controller and integrates an FPGA acceleration unit, and the operating temperature range of the edge computing module is between-40℃ and 85℃. The dynamic suppression module includes a motor 13, a piston 12, a storage bin 10 and an electromagnetic valve 11, the piston 12 is arranged in the storage bin 10, and the motor 13 is in transmission connection with the piston 12. A plurality of metal foam particles are arranged in the storage bin 10 and the fire blocking section 3, and the metal foam particles in the storage bin 10 and the metal foam particles in the fire blocking section 3 can be of different types, and the metal foams stored in the storage bin 10 can also be stacked in different types in the up-down direction. The storage bin 10 is communicated with the fire blocking section 3 through a connecting channel, and the electromagnetic valve 11 is arranged in the connecting channel.

[0022] The application discloses an AI-based real-time monitoring method for fire blocking efficiency, which is applied to the fire arrester and will be specifically described below with reference to the drawings.

[0023] As shown in the drawings, Figure 3 The AI-based real-time monitoring method for fire blocking efficiency specifically includes the following steps: first, the multi-modal sensor module monitors various data of the fire arrester body in real time, wherein the optical fiber sensor 7 collects temperature and strain data in the fire arrester body every certain period of time; the acoustic wave detector 8 records the sound pressure waveform of the fire arrester body in real time, performs FFT analysis to extract the energy characteristics of a specific frequency range, obtains sound wave spectrum data, and identifies the deflagration characteristics; the infrared detector captures a frame of thermal distribution image every certain period of time, and performs noise filtering and feature enhancement on the collected image through the edge computing module.

[0024] Second step, the edge computing module pre-processes the data through timestamp synchronization technology, aligns the temperature data, acoustic spectrum data and thermal distribution image stream, and generates a time-space multi-dimensional feature matrix. Then, the pre-processed data is input into the AI model, and the fire resistance performance score is obtained through the running of the lightweight convolutional neural network through the scoring formula, and the particulate matter loss is predicted.

[0025] The scoring formula is: ; In the formula, represents the average temperature; represents the peak sound pressure, represents the defect area identified by thermal imaging; represents the temperature weight coefficient, ; represents the sound pressure weight coefficient, ; represents the defect area weight coefficient, The range of the fire resistance performance score formula result is between 0 and 100%.

[0026] Third step, the edge computing module judges whether the fire resistance performance score is lower than the preset threshold value. When it is lower than the threshold value, the protection action is triggered through the dynamic suppression module. The threshold value includes a first response threshold value of 80% and a second response threshold value of 75%. When the fire resistance performance score is lower than the first response threshold value, the protection action triggered is to open the electromagnetic valve 11 to release the metal foam particles in the storage bin 10 to fill the gap at the top of the fire resistance section 3 for absorbing the detonation capacity. When the fire resistance performance score is lower than the second response threshold value, the protection action triggered is that the motor 13 drives the piston 12 to push the metal foam particles into the fire resistance section 3 for compensation according to the predicted particulate matter loss of the AI model.

[0027] Fourth step, the edge computing module regularly uploads the data packet to the cloud management platform to update the parameters of the digital twin model.

[0028] Fifth step, the cloud management platform analyzes the failure cause through the digital twin model and proposes suggestions to the staff to optimize the working parameters.

[0029] The following describes the first embodiment in combination with the actual scenario of the gas phase communication pipeline of the petroleum chemical storage tank.

[0030] 1. Scene parameters: In a certain tank area, one DN300 carbon steel pipeline containing oil and gas, wall thickness 12mm, working pressure 0.8MPa, medium is light crude oil volatile gas (methane accounts for 60%, ethane 30%). Fire barrier 3 design: length 2m, shell material 316L stainless steel, filled with alumina particles (particle size 3mm, density 1.2g / cm³), temperature resistance range between -20℃ to 300℃. Environmental conditions: open-air installation, environmental temperature range between -30℃ to 50℃, relative humidity ≤95%.

[0031] 2. Equipment selection: multi-modal sensing module: optical fiber sensor 7 uses distributed optical fiber temperature measurement technology, temperature measurement accuracy ±0.5℃, strain resolution 1με, node spacing 10cm. Optical fiber sensor 7 collects internal temperature and strain data of fire barrier 3 every 0.1 seconds, and transmits it to the edge computing module through the RS485 bus. Sound detector 8: uses a wideband piezoelectric sensor, frequency response range between 10Hz and 50kHz, sensitivity 50mV / Pa, IP67 protection level. Wideband piezoelectric sensor is used to record sound pressure waveform in real time, and fast Fourier transform (FFT) is used to extract energy features in the frequency range of 1kHz to 50kHz to identify deflagration characteristics. Infrared thermal imager 9: uses an infrared thermal imager with a high-temperature-resistant quartz window (temperature resistance 1200℃), captures a frame of thermal distribution image every 20ms, and transmits it to the edge module after H.265 encoding compression.

[0032] Edge computing module: uses an industrial-grade embedded controller with an integrated FPGA acceleration unit, operating temperature range between -40℃ and 85℃. Through timestamp synchronization technology, the optical fiber temperature data, sound spectrum data and thermal imaging image stream are aligned to generate a time-space multi-dimensional feature matrix. The AI model inference process runs a lightweight convolutional neural network (CNN), the input layer includes average temperature, received thermal imaging image and sound spectrum image, and the output layer outputs the fire barrier efficiency score (error ≤2%) through the fire barrier efficiency score formula.

[0033] Dynamic suppression module: electromagnetic valve 11 response time ≤2ms, making the storage bin 10 and the fire barrier 3 quickly connected. When receiving the edge computing module instruction, the electromagnetic valve 11 opens, releasing the metal foam particles (porosity 90%, pore size 0.5mm) to fill the top gap of the fire barrier 3, absorbing the detonation energy. Linear motor 13 cooperates with high-precision encoder to dynamically adjust the spring pre-tightening force according to the AI model predicted particle loss (error ±3g), linear motor 13 drives piston 12 to push the metal foam particles into the fire barrier 3 for compensation, with a compensation accuracy of ±0.5mm.

[0034] 3、Data acquisition: Data acquisition and synchronization: Fiber optic sensor 7 collects temperature data every 0.1 seconds and transmits it to the edge computing module through the RS485 bus. The broadband piezoelectric sensor records the sound pressure waveform in real time, and the data acquisition card performs FFT analysis to extract energy features in the frequency range of 1 kHz to 50 kHz. The infrared thermal imager captures a frame of image every 20 ms, which is compressed by H.265 encoding and transmitted to the edge computing module.

[0035] 4、Defect identification and decision: The AI model in the edge computing module detects that the temperature gradient in the flame arrestor section 3 increases to 250℃ / m (normal range <100℃ / m), the sound pressure peak value is 12MPa (threshold value 10MPa), and the thermal imaging shows a local crack of 3mm×5mm. The fiber optic sensor 7 detects local temperature anomalies, the sound wave detection unit captures high-frequency sound pressure fluctuations, and the infrared thermal imager displays the crack location. The edge computing module analyzes the AI model and determines that the flame arrestor efficiency score drops to 58%, triggering the primary response threshold and the secondary response threshold. The edge computing module triggers the suppression command within 5ms.

[0036] 5、Dynamic suppression and repair: The electromagnetic valve 11 opens within 5ms to release alumina metal foam particles to fill the crack area in the flame arrestor section 3 and absorb the detonation energy. Then, according to the loss predicted by the AI model, the motor 13 drives the piston 12 to move and compensate for the alumina metal foam particles in the flame arrestor section 3, restoring the density of alumina metal foam particles in the flame arrestor section 3.

[0037] 6、Repair effect verification: After repair, the temperature gradient drops to 80℃ / m, the sound pressure peak value falls to 8MPa, and the flame arrestor efficiency score rises to 92%. The system automatically generates a fault report, records the fault location, processing process and repair effect, and uploads it to the cloud management platform.

[0038] 7、Optimization of working parameters: The cloud management platform analyzes the fault cause through the digital twin model and determines that it is caused by long-term high temperature creep leading to metal fatigue. The cloud management platform suggests optimizing the filling density of metal foam particles in the flame arrestor section 3 to 1.25g / cm³ and increasing the density of local temperature monitoring nodes to prevent similar faults from occurring again.

[0039] The following describes embodiment two in combination with the actual city gas transmission pipeline scene.

[0040] 1. Scene parameters: A flammable gas pipeline is laid in an urban underground pipe gallery. The pipeline specification is DN200 polyethylene pipeline, the working pressure is 0.4 MPa, and the conveying medium is methane / hydrogen mixed gas (methane accounts for 50%~90%, hydrogen fluctuates ±30%). The design of the flame arrestor 3: bidirectional reversible structure, the outer shell material is corrosion-resistant stainless steel (wall thickness 8 mm), filled with silicon carbide particles (particle size 1.5 mm, density 1.5 g / cm³), temperature resistance range between -30°C and 200°C, suitable for underground burial depth of 1.5 m and humidity ≤80% environment. Environmental conditions: containing trace amounts of hydrogen sulfide (H2S ≤10 ppm), environmental temperature range between -20°C and 40°C.

[0041] 2. Equipment selection: multi-modal sensing module: optical fiber sensor 7 uses distributed optical fiber temperature measurement technology, temperature measurement accuracy ±0.3°C, strain resolution ±2με, node spacing 5 cm, externally wrapped with anti-hydrogen sulfide corrosion coating. The sensor collects temperature and strain data every 0.05 seconds, transmits to the edge computing module through the industrial communication protocol (such as OPC UA), and generates a three-dimensional thermal map of the flame arrestor 3 in real time. The sound wave detector 8 uses a four-channel microphone array, with a frequency response range of 20 Hz to 100 kHz, a dynamic range of 140 dB, and a deflagration source positioning accuracy of ±1 mm. The unit analyzes the sound pressure waveform through beamforming algorithm, and when the sound pressure peak value is detected to be >8 MPa, the deflagration source is automatically mapped to the coordinates of the flame arrestor 3 (such as X:1.2 m, Y:0.3 m). The infrared thermal imager 9 uses a multispectral infrared thermal imager, which needs to be equipped with a methane / hydrogen special filter (wavelength 3.3 μm / 2.1 μm), with a resolution of not less than 1024×768 pixels and a frame rate of ≥100 fps. The multispectral infrared thermal imager dynamically captures the flame shape, and distinguishes gas components through multispectral fusion technology, such as real-time monitoring of methane concentration from 70% to 50%.

[0042] 3. Edge computing and dynamic suppression process: edge computing module: integrated with high-performance processors and AI acceleration units, supporting real-time analysis of gas components (such as gas chromatography-mass spectrometry interface). The edge computing module runs the optimized deep learning AI model, inputs the thermal imaging map (normalized to 512×512 pixels), sound source coordinates and gas concentration data, and outputs the flame arrestor efficiency score: ; wherein, represents the average temperature; represents the sound pressure peak value, represents the defect area identified by thermal imaging; represents the temperature weight coefficient, ; represents the sound pressure weight coefficient, ; a defect area weight coefficient, .

[0043] In the dynamic suppression module, the lower layer of the storage bin 10 stores zirconia ceramic metal foam particles, and the upper layer stores silicon carbide particles.

[0044] Dynamic suppression trigger condition: when the hydrogen mixture ratio suddenly increases to 35%, the model detects that the fireproofing efficiency score decreases from 85% to 72%, triggering a two-level response threshold: primary response threshold (fireproofing efficiency score < 80%): high-frequency electromagnetic valve 11 (response time ≤ 1 ms) is opened to fill the gap with zirconia ceramic metal foam particles (porosity 95%, temperature resistance 2000℃), absorbing the energy peak 12MPa of the detonation wave. Secondary response threshold (fireproofing efficiency score < 75%): high-thrust linear motor 13 (thrust ≥ 1000N) combined with laser displacement sensor (resolution 0.1μm) drives piston 12, accurately compensates 80g of silicon carbide metal foam particles, restores the density of the fireproofing layer to 1.55g / cm³.

[0045] 4. Cloud collaboration and maintenance optimization: data transmission and model training: every 10 minutes, upload the running data (including temperature distribution, sound pressure spectrum, particle filling rate) to the cloud management platform. The cloud management platform builds a digital twin model based on historical data to predict the service life decay curve (fitting degree R 2 ≥ 0.98) of the fireproofing section 3, and identifies abnormal wear rate (such as > 0.1mm / month). Dynamic parameter adjustment: when local wear is detected to be excessive, the cloud management platform automatically issues instructions to adjust the filling metal foam particle density to 1.6g / cm³, and synchronizes to the edge module through industrial protocols (such as Modbus TCP), ensuring that the fireproofing efficiency is ≥ 98.5%.

[0046] Fault handling case: once the system alarm shows that the fireproofing efficiency score of the front end of the fireproofing section 3 drops to 65%, the infrared thermal imaging shows a 2mm×4mm crack. The edge computing module triggers metal foam particle filling and particle compensation (60g) through the dynamic suppression module, and the cloud analysis determines that the construction vibration causes structural fatigue, and suggests adding damping supports. After repair, the score is restored to 92%, and the data is stable for the next 3 months.

[0047] 5、Effect verification and technical indicators: Flame resistance: The flame resistance efficiency is greater than or equal to 98.5% and the detonation wave energy attenuation rate is greater than or equal to 88% under the working condition of a hydrogen mixture ratio of 35% through EN ISO 16852:2016 standard testing. Positioning accuracy: The positioning error of the sound wave detection unit for a 5mm micro crack detonation source is less than or equal to ±0.8mm, which is better than the industry standard (±2mm). Environmental adaptability: In a corrosive environment with H2S less than or equal to 10ppm, the service life of the sensor and the flame resistance element is greater than or equal to 5 years without significant performance attenuation. Economy: The annual maintenance cost is reduced by 55% and the manual inspection frequency is reduced from once a week to once a month.

[0048] 6、Fault handling process: On a certain day, the system alarm showed that the front-end efficiency score of the flame resistance section 3 dropped to 65%, and the thermal imaging showed a 2mm x 4mm crack. The processing process is as follows: the edge module triggers the zirconium oxide foam filling, compensating for 60g of particulate matter. The cloud management platform analyzes the causes of the crack and determines that the construction machinery vibration caused structural fatigue, and suggests adding a damping support. After repair, the score returned to 92% and the monitoring data was stable for the next 3 months.

[0049] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present application should also be within the scope of the present application.

Claims

1. A flame arrester, characterized in that: It includes a flame arrester body, a multimodal sensing module, an edge computing module, a dynamic suppression module, and a cloud management platform. The multimodal sensing module and the dynamic suppression module are set on the flame arrester body, and the edge computing module is connected to the multimodal sensing module, the dynamic suppression module, and the cloud management platform respectively. The dynamic suppression module includes a motor (13), a piston (12), a storage bin (10), and a solenoid valve (11). The piston (12) is located inside the storage bin (10), and the motor (13) is connected to the piston (12) via a drive. Several metal foam particles are provided inside the storage bin (10) and the flame arrestor section (3) of the flame arrestor body. The storage bin (10) is connected to the flame arrestor section (3) through a connecting channel, and the solenoid valve (11) is located inside the connecting channel.

2. A flame arrester according to claim 1, characterized in that: The multimodal sensing module includes a fiber optic sensor (7), an acoustic detector (8), and an infrared thermal imager (9). The end of the fiber optic sensor (7) is embedded in the flame arrester body. The acoustic detector (8) is set on the outer wall of the flame arrester. The infrared thermal imager (9) is adjacent to the flange (6) and embedded in the flame arrester body. The flame arrester body also includes an inlet section (1), a flared section (2), a contraction section (4) and an outlet section (5), with flanges (6) fixed to both the inlet section (1) and the outlet section (5); Several fiber optic sensors (7) are respectively installed in the inlet section (1), the flaring section (2), the contraction section (4) and the outlet section (5), and each fiber optic sensor (7) is evenly arranged along the axial direction of the flame arrester body. Several acoustic detectors (8) are fixed on the outer walls of the inlet section (1) and the outlet section (5), respectively; Several infrared thermal imagers (9) are embedded in the inlet section (1) and the outlet section (5) respectively, and each infrared thermal imager (9) is equipped with a high-temperature resistant quartz window.

3. A flame arrester according to claim 1, characterized in that: The edge computing module uses an industrial-grade embedded controller and integrates an FPGA acceleration unit. The operating temperature range of the edge computing module is between -40℃ and 85℃.

4. A method for real-time monitoring of flame arresting performance based on AI, applied to a flame arrester as described in any one of claims 1 to 3, characterized in that, Includes the following steps: S1, the multimodal sensing module monitors the temperature, sound waves and heat distribution image data of the flame arrester body in real time, and transmits the data to the edge computing module; S2, the edge computing module preprocesses the data and inputs the preprocessed data into the AI ​​model, which then outputs a fire-retardant performance score. S3, the edge computing module determines whether the flame arresting performance score is lower than the preset threshold. When it is lower than the threshold, the dynamic suppression module triggers a protection action to replenish metal foam particles into the flame arrester body. S4, the edge computing module periodically uploads data packets to the cloud management platform to update the parameters of the digital twin model; The S5 cloud management platform analyzes the causes of failures and optimizes operating parameters through digital twin models.

5. The AI-based real-time monitoring method for fire-retardant performance according to claim 4, characterized in that, In S1, the fiber optic sensor (7) collects temperature and strain data inside the flame arrester body at regular intervals; The acoustic wave detector (8) records the acoustic pressure waveform of the flame arrester body in real time, performs FFT analysis to extract the energy characteristics of a specific frequency band, obtains acoustic wave spectrum data, and identifies deflagration characteristics; The infrared thermal imager (9) captures a frame of thermal distribution image at regular intervals and performs noise filtering and feature enhancement on the acquired image through the edge computing module.

6. The AI-based real-time monitoring method for fire-retardant performance according to claim 4, characterized in that, In S2, the preprocessing process of the edge computing module is as follows: by using timestamp synchronization technology, temperature data, acoustic spectrum data and thermal distribution image stream are aligned to generate a temporal and spatial multidimensional feature matrix.

7. The method for real-time monitoring of fire-retardant performance based on AI according to claim 4, characterized in that, In S2, the reasoning process of the AI ​​model is as follows: by running a lightweight convolutional neural network, the preprocessed input data is used to obtain the fire-retardant performance score through a scoring formula, and the particulate matter loss is predicted. The scoring formula is: ； In the formula, Indicates average temperature; Indicates the peak sound pressure level. This represents the area of ​​the defect identified by thermal imaging. This represents the temperature weighting coefficient. ; This represents the sound pressure weighting coefficient. ; This represents the defect area weighting coefficient. .

8. A method for real-time monitoring of fire-retardant performance based on AI, as described in claim 7, characterized in that, In S3, the fire-retardant performance rating formula results range from 0 to 100%, with thresholds including a first-level response threshold and a second-level response threshold. The first-level response threshold is 80%, and the second-level response threshold is 75%.

9. The method for real-time monitoring of fire-retardant performance based on AI according to claim 8, characterized in that, in, When the fire-arresting performance score is lower than the first-level response threshold, the triggered protection action is: to open the solenoid valve (11) and release the metal foam located in the storage bin (10) to fill the gap at the top of the fire-arresting section (3) to absorb the detonation energy. When the fire-retardant performance score is lower than the secondary response threshold, the triggered protection action is: based on the particulate matter loss predicted by the AI ​​model, the motor (13) drives the piston (12) to push the metal foam in the storage bin (10) into the fire-retardant section (3) for compensation.

10. The method for real-time monitoring of fire-retardant performance based on AI according to claim 4, characterized in that, In S3, the metal foam particles located in the storage bin (10) are of different types than those in the fire-resistant section (3), and different types of metal foam particles are placed in the storage bin (10) along the vertical direction.

Images