Diagnosis System for Hydrogen Gas Facility Based on Optical Sensor

Abstract

The present invention relates to a diagnostic system for hydrogen gas facilities based on optical sensors, and more specifically, to a diagnostic system for hydrogen gas facilities based on optical sensors that integrates a line-type optical vibration and temperature sensor laid along an extended path of a hydrogen gas facility with a point-type optical hydrogen sensor installed at a key location, and applies Edge-AI technology to monitor various abnormal signs such as excavation, impact, and hydrogen leakage in real time, and precisely diagnose and predict potential risks.

Description

Technology Field

[65535] The present invention relates to a diagnostic system for hydrogen gas facilities based on optical sensors, and more specifically, to a diagnostic system for hydrogen gas facilities based on optical sensors that integrates a line-type optical vibration and temperature sensor laid along an extended path of a hydrogen gas facility with a point-type optical hydrogen sensor installed at a key location, and applies Edge-AI technology to monitor various abnormal signs such as excavation, impact, and hydrogen leakage in real time, and precisely diagnose and predict potential risks. Background Technology As the energy transition to address climate change and achieve carbon neutrality accelerates globally, the importance of hydrogen as a clean energy source is growing day by day. Accordingly, the construction of infrastructure for the production, storage, transportation, and utilization of hydrogen is actively underway, and the distribution of high-pressure hydrogen gas facilities is also rapidly increasing. However, despite its utility, hydrogen energy carries inherent risks in handling; therefore, ensuring safety has emerged as the top priority for the expansion of hydrogen infrastructure. Hydrogen gas possesses characteristics such as a very wide flammability range and low minimum ignition energy, meaning that even a minute electrostatic spark can lead to an explosion. Furthermore, hydrogen molecules are very small and can penetrate into metal materials, causing hydrogen embrittlement and leading to brittle fracture. Additionally, due to their extremely rapid diffusion rate, there is a high risk of creating an explosive atmosphere within a short period if a leak occurs in a confined space. Furthermore, since hydrogen is colorless and odorless and the flames from a leak are not easily visible to the naked eye, there are limitations in detecting early warning signs of accidents using conventional sensory inspections or general monitoring systems. Conventional gas safety management systems have primarily relied on point-type electric sensors, such as electrochemical or catalytic combustion sensors. While these electric sensors are useful for measuring gas concentrations at specific points, they have a critical drawback: they create blind spots where leaks occurring in un-sensored areas or along long sections of pipelines cannot be detected. Additionally, electric sensors have the potential to generate sparks during operation, making high-cost explosion-proof equipment mandatory when installed in hazardous areas. Moreover, concerns have been raised regarding their vulnerability to electromagnetic interference (EMI) and corrosive environments, making it difficult to guarantee durability and reliability.To address these issues, distributed sensing technology utilizing optical fibers has been introduced; however, this approach also faces limitations in application due to the inherent physical characteristics of hydrogen gas. In the case of natural gas (LNG) or liquefied petroleum gas (LPG), the Joule-Thomson effect—in which a rapid temperature drop occurs due to adiabatic expansion upon leakage—is distinct, making it easy to detect leaks simply by sensing temperature changes. However, when hydrogen gas leaks at room temperature, the Joule-Thomson coefficient takes on a negative value or the change is very minimal, resulting in almost no temperature change or a tendency for the temperature to rise. Consequently, it is virtually impossible to detect minute hydrogen gas leaks in their early stages using only existing temperature-based optical fiber sensors, which acts as a significant gap in the safety management of hydrogen facilities. Furthermore, hydrogen pipelines are often buried underground in urban areas or industrial complexes, constantly exposed to vibrations and shocks caused by external excavation work or vehicle traffic. Conventional vibration detection technologies have primarily relied on measuring vibration magnitude and triggering an alarm when a threshold is exceeded. However, this approach fails to distinguish between vibrations caused by nearby excavation work and those resulting from actual pipe damage, leading to frequent false alarms. These frequent false alarms cause a "boy who cried wolf" effect, distracting control personnel and hindering rapid response in the event of an actual accident, thereby undermining the overall reliability of the system. Furthermore, existing centralized cloud-based monitoring systems face limitations in network bandwidth and latency issues when transmitting and processing the vast amount of high-frequency sensor data collected from the field in real time. Since emergency situations, such as hydrogen explosions, unfold in seconds, the time delays in the process of collecting data, transmitting it to a server for analysis, and sending control signals back to the site cause the system to miss the critical "golden time" for accident prevention.Therefore, there is an urgent need to introduce edge computing technology that enables immediate data analysis and decision-making on-site. Against this backdrop, the development of a new type of diagnostic system that integrates specialized sensing technology capable of precisely detecting minute hydrogen gas leaks, intelligent analysis technology capable of minimizing false alarms caused by external environmental factors, and high-speed signal processing technology capable of processing large volumes of data without delay has emerged as an urgent task. The problem to be solved The present invention has been devised to improve upon the aforementioned problems. The purpose of the present invention is to provide a diagnostic system for hydrogen gas facilities based on optical sensors that can precisely detect changes in the physical state of hydrogen gas facilities and micro-leaks without blind spots by applying a multimodal sensing method that combines a line-type optical vibration and temperature sensor with a point-type optical hydrogen sensor. Furthermore, according to an embodiment of the present invention, the purpose is to provide an intelligent diagnostic system that can clearly distinguish between simple vibrations caused by external excavation work and abnormal signs caused by hydrogen leakage, and drastically reduce the false alarm rate, by introducing an FPGA module capable of high-speed signal processing and an Edge-AI algorithm to analyze large volumes of sensor data in real-time at the site. In addition, according to an embodiment of the present invention, the purpose is to provide a system that can self-generate the accuracy of a diagnostic model by establishing a systematic data management platform that leads to data collection, processing, labeling, and verification, and by continuously relearning algorithms that reflect field feedback. means of solving the problem To solve the aforementioned problem, the present invention relates to a diagnostic system for a hydrogen gas facility based on optical sensors for real-time monitoring of the physical condition and gas leakage of a facility including piping, valves, and storage tanks of a hydrogen gas facility, comprising: a line-type optical vibration and temperature sensor installed along an extended path of the hydrogen gas facility to detect vibration and temperature changes in a distributed manner, and a point-type optical hydrogen sensor installed at a key point including a valve connection or joint of the hydrogen gas facility to detect localized hydrogen gas leakage, wherein the system includes a sensing unit that detects changes in the physical condition of the hydrogen gas facility as optical signals; a vibration data collection unit connected to the sensing unit via an optical path to collect vibration signals detected from the line-type optical vibration and temperature sensor, a hydrogen concentration data collection unit that collects hydrogen concentration signals detected from the point-type optical hydrogen sensor, and a temperature data collection unit that collects temperature signals detected from the line-type optical vibration and temperature sensor, wherein the optical signals are converted into digital signals through an FPGA module capable of high-speed sampling board control, and each collected data is stored in a time-series database called Influx DB (240). A data collection and analysis system that stores data, performs preprocessing including noise removal, data normalization, and statistical characteristic analysis on the data stored in the Influx DB, and then analyzes the data using an Edge AI algorithm to determine, classify, and output normal state data and abnormal state data in real time;A diagnostic platform for a hydrogen gas facility that receives, stores, and manages the normal state data and the abnormal state data identified from the data collection and analysis system, predicts risks and diagnoses abnormal situations of the hydrogen gas facility based on event classification results by the Edge AI algorithm, stores the diagnosed results and metadata in an integrated database including a relational database Maria DB and a time-series database, provides visualized monitoring information to a user through a web server (Web Server, 310), or transmits analysis data to an external upper-level control platform through a Rest API;The above data collection and analysis system includes, by analyzing the correlation of vibration, temperature, and hydrogen concentration data, which are multimodal data collected at the site and stored in the above Influx DB, detects risk factors in real time, including vibration patterns caused by external excavation work, impact patterns caused by changes in internal pipe pressure, concentration change patterns due to hydrogen leakage, and rapid temperature rise patterns due to fire occurrence, and transmits event information regarding the detected risk factors to the diagnostic platform for the hydrogen gas facility so that an alarm is executed. The above line-type optical vibration and temperature sensor has a structure in which an optical fiber, glass fiber reinforced plastic protecting the optical fiber, an aluminum tube, and a polyethylene sleeve are sequentially laminated, and is installed by being buried or attached in a zigzag or spiral shape along the piping of the hydrogen gas facility, and the above point-type optical hydrogen sensor is installed at leakage-vulnerable points including valves, pressure regulators, and joints of the hydrogen gas facility to compensate for the characteristic of hydrogen gas, in which the temperature change due to the Joule-Thomson effect is weak upon leakage due to the characteristics of hydrogen molecules, and the hydrogen at a specific point Detects changes in concentration. The data acquisition and analysis system comprises: a hardware control module including a laser diode driving circuit, a temperature control controller, a pulse driving controller, and a digital-to-analog converter for driving the line-type optical vibration and temperature sensor and the point-type optical hydrogen sensor; and an FPGA chip equipped with a signal processing algorithm that performs phase analysis of an optical signal received from the sensing unit and performs high-speed, high-capacity signal acquisition and processing.It is configured to include, wherein the Edge AI algorithm identifies the normal state of the hydrogen gas facility and the abnormal state including the risk factors through a model trained based on a convolutional neural network or a long short-term memory model using collected time-series data and spatial distribution data as input. The diagnostic platform for the hydrogen gas facility includes a data inspection module that receives the abnormal state data primarily classified by the data collection and analysis system, sets it as data to be inspected, and verifies the consistency of the data through a multi-stage inspection process including acquisition, refinement, labeling, and full inspection; The present invention further includes an analysis algorithm advancement module that performs retraining of the Edge AI algorithm based on data verified through the data inspection module, and updates and distributes the retrained algorithm to the data collection and analysis system to reduce the false alarm rate and improve diagnostic accuracy. The present invention relates to a diagnostic method for a hydrogen gas facility based on optical sensors for real-time monitoring of the physical condition and gas leakage of a facility including piping, valves, and storage tanks of a hydrogen gas facility, comprising the steps of: a sensing unit detecting vibration and temperature changes in a distributed manner through a line-type optical vibration and temperature sensor laid along an extended path of the hydrogen gas facility, and detecting local hydrogen gas leakage through a point-type optical hydrogen sensor installed at a key point of the hydrogen gas facility to generate an optical signal; and a data collection and analysis system receiving the optical signal generated from the sensing unit, and collecting vibration, hydrogen concentration, and temperature data through a vibration data collection unit, a hydrogen concentration data collection unit, and a temperature data collection unit, respectively, and storing them in a time-series database called Influx DB (240).A step in which the data collection and analysis system performs preprocessing, including noise removal, data normalization, and vectorization, on raw data stored in the Influx DB to process it into a data form suitable for Edge AI learning and inference; a step in which the data collection and analysis system inputs the preprocessed data into an Edge AI algorithm for analysis, thereby classifying and determining in real-time risk factor patterns, including external excavation, impact, hydrogen leakage, and fire, and normal state patterns; The diagnostic platform for hydrogen gas facilities receives normal and abnormal state data identified from the data collection and analysis system and stores it in an integrated database, diagnoses dangerous situations based on event classification results, and provides visualized information through a web server or transmits the analysis data to an external upper-level control platform to generate an alarm; wherein the step includes a complex analysis of the correlations of the vibration, temperature, and hydrogen concentration data to exclude false detections by a single sensor, and detects risk factors by distinguishing between micro-leaks based on hydrogen molecule characteristics and noise caused by the external environment. The step includes a step of performing noise filtering based on sensor-specific characteristics on time-series data and spatial distribution data extracted from the influx database; and a step of performing normalization and standardization to unify the scale of the filtered data. and includes the step of vectorizing data to match the input dimensions of the Edge AI algorithm and performing data augmentation to compensate for the lack of training data. Prior to the above step, the method comprises the step of constructing a training dataset by performing labeling by event class, including temperature anomaly signs, normal operation of hydrogen valves, vibration anomaly signs, leak diffusion patterns, external excavation and perforation, hydrogen leakage and abnormal temperature, normal operation of hydrogen pipelines, and micro-hydrogen leakage, using collected historical vibration, temperature, and hydrogen concentration data.The method further includes the step of training a deep learning model based on a convolutional neural network or a recurrent neural network using the constructed training dataset, and loading the trained model into the data collection and analysis system as an Edge AI algorithm. The step distinguishes whether the vibration is caused by simple excavation work or an accident accompanied by a gas leak due to pipe damage by synchronizing and comparing the trend of temperature data change at the corresponding location with the change in hydrogen concentration of the point-type optical hydrogen sensor when the vibration signal detected by the line-type optical vibration and temperature sensor exceeds a preset threshold. Effects of the invention A diagnostic system for hydrogen gas facilities based on optical sensors according to one embodiment of the present invention provides the following effects. First, it offers an accident prevention effect through intelligent risk management technology. By integrating line-type and point-type optical sensors specialized for high-pressure hydrogen facilities, it compensates for the leakage characteristics of hydrogen gas, which exhibits minimal temperature changes, and detects various risk factors ranging from micro-leaks to external impacts in real time, thereby preventing major accidents in advance. Second, it enhances the scientific nature and efficiency of safety management. By collecting and analyzing status data of energy facilities in real time and introducing edge artificial intelligence technology, it transforms safety management—which previously relied on empirical judgment—into data-driven scientific management. This reduces unnecessary shutdowns of high-pressure hydrogen production, storage, and transportation facilities, optimizes component replacement cycles, reduces facility maintenance costs, and maximizes management efficiency. Third, it secures synergy effects and technological competitiveness through the convergence of heterogeneous technologies. By integrating heterogeneous technologies such as optical sensor technology, FPGA-based high-speed signal processing technology, and AI and big data analysis technology, it secures source technology in the technically challenging field of hydrogen facility diagnosis, thereby bridging the technological gap with advanced nations and lowering entry barriers. Fourth, it offers the potential to expand application fields. Multimodal sensor technology capable of simultaneously measuring vibration and temperature, along with an AI-based analysis platform, can be expanded for safety diagnosis in various industrial infrastructures—such as hydrogen pipelines, other energy transport pipelines, chemical plants, and hydrogen refueling stations—thereby enhancing the overall safety level of the industry. Fifth, it contributes to public safety and the achievement of carbon neutrality. By establishing a reliable risk monitoring system for high-pressure hydrogen facilities, it contributes to public safety by alleviating the public's vague anxieties regarding hydrogen infrastructure and preventing damage caused by natural disasters. Furthermore, it can contribute to achieving the 2050 carbon neutrality goal by accelerating the transition to eco-friendly energy.Finally, through an edge computing-based real-time response system, immediate risk identification and alert issuance are possible at the scene without data transmission delays in the event of an accident, thereby securing the golden time and minimizing the spread of damage. Brief explanation of the drawing FIG. 1 is a schematic diagram showing the overall configuration of a diagnostic system for a hydrogen gas facility based on an optical sensor according to an embodiment of the present invention, as well as the connection relationship between the internal network and the upper platform. FIG. 2 is an exemplary diagram showing a schematic optical fiber sensor packaging for mounting on a hydrogen gas facility according to an embodiment of the present invention. FIG. 3 is a block diagram illustrating a physical hardware photograph of a high-speed, high-capacity signal acquisition FPGA module mounted on a data acquisition and analysis system according to an embodiment of the present invention and an internal firmware control structure. FIG. 4 is a flowchart showing the development process of edge artificial intelligence-based risk prediction and anomaly detection technology according to an embodiment of the present invention in stages. FIG. 5 is a conceptual diagram showing the data collection, processing, purification, labeling, and multi-stage inspection processes performed on a diagnostic platform for a hydrogen gas facility according to an embodiment of the present invention. Specific details for implementing the invention As illustrated in FIG. 1, the present invention comprises a sensing unit (100) that detects changes in the physical state of a hydrogen gas facility as an optical signal, including a line-type optical vibration and temperature sensor (110) laid along an extended path of the hydrogen gas facility to detect vibration and temperature changes in a distributed manner, and a point-type optical hydrogen sensor (120) installed at a key point including a valve connection or joint of the hydrogen gas facility to detect localized hydrogen gas leakage; A data collection and analysis system (200) comprising a vibration data collection unit (210) connected to the sensing unit (100) via an optical path to collect vibration signals detected from the line-type optical vibration and temperature sensor (110), a hydrogen concentration data collection unit (220) to collect hydrogen concentration signals detected from the point-type optical hydrogen sensor (120), and a temperature data collection unit (230) to collect temperature signals detected from the line-type optical vibration and temperature sensor (110); converting the optical signals into digital signals through an FPGA module capable of high-speed sampling board control; storing each collected data in an Influx DB (240), which is a time-series database; performing preprocessing including noise removal, data normalization, and statistical characteristic analysis on the data stored in the Influx DB (240); and analyzing using an Edge AI algorithm to determine, classify, and output normal state data and abnormal state data in real time. and receives, stores, and manages the normal state data and the abnormal state data determined from the data collection and analysis system (200), predicts risks and diagnoses abnormal situations of the hydrogen gas facility based on the event classification results by the Edge AI algorithm, and stores the diagnosed results and metadata in an integrated database (320) including a relational database Maria DB and a time series database (TSDB).The present invention includes a diagnostic platform (300) for hydrogen gas facilities that provides visualized monitoring information to a user through a web server (310) or transmits analysis data to an external upper control platform (400) through a Rest API. The data collection and analysis system (200) analyzes the correlation of vibration, temperature, and hydrogen concentration data, which are multimodal data collected at the site and stored in the Influx DB (240), to detect risk factors in real time, including vibration patterns caused by external excavation work, shock patterns caused by changes in internal pipe pressure, concentration change patterns due to hydrogen leakage, and rapid temperature rise patterns due to fire occurrence, and transmits event information regarding the detected risk factors to the diagnostic platform (300) for hydrogen gas facilities so that an alarm is performed. As illustrated in FIG. 2, the present invention relates to a diagnostic system for hydrogen gas facilities based on optical sensors for monitoring the physical condition of facilities, including pipes, valves, or storage tanks of hydrogen gas facilities, and whether gas is leaking in real time. The system according to the present invention is largely composed of a sensing unit (100), a data collection and analysis system (200), a diagnostic platform for hydrogen gas facilities (300), and an upper control platform (400). Each of the above components is organically combined to enable early detection and response to various risk factors that may occur in hydrogen gas facilities. The sensing unit (100) performs the role of detecting changes in the physical state of the hydrogen gas facilities as optical signals and includes a line-type optical vibration and temperature sensor (110) and a point-type optical hydrogen sensor (120). The line-type optical vibration and temperature sensor (110) is installed along the extended path of the piping or storage tank of the hydrogen gas facilities to detect vibration and temperature changes in a distributed manner. Looking at the specific stacked structure of the line-type optical vibration and temperature sensor (110), the central optical fiber,The structure has a sequentially laminated glass fiber reinforced plastic (GFRP) to protect the optical fiber, an aluminum tube (Al tube) to block external shocks and electromagnetic interference, and a polyethylene sleeve (PE sleeve) to protect the outermost layer. This multi-layer structure safely protects the optical fiber even in the harsh environment of a high-pressure hydrogen facility and helps to collect reliable data over a long period. The line-type optical vibration and temperature sensor (110) may be installed by being embedded or attached in a zigzag or spiral shape on the surface of the hydrogen gas facility to increase detection efficiency. Additionally, the sensing unit (100) includes a point-type optical hydrogen sensor (120). While the temperature of a general gas decreases due to the Joule-Thomson effect upon leakage, hydrogen gas, due to its molecular characteristics, may show only a slight temperature change or even increase upon leakage, making it difficult to detect micro-leaks using only a temperature sensor. Accordingly, the point-type optical hydrogen sensor (120) is installed at key locations with a high probability of leakage, such as valve connections, joints, or pressure regulators of hydrogen gas facilities, to directly detect local changes in hydrogen concentration, thereby compensating for the limitations of the line-type optical vibration and temperature sensor (110). The optical signal detected by the sensing unit (100) is transmitted to the data collection and analysis system (200). The data collection and analysis system (200) is equipped with a vibration data collection unit (210), a hydrogen concentration data collection unit (220), and a temperature data collection unit (230) to collect vibration, hydrogen concentration, and temperature signals, respectively, from the line-type optical vibration and temperature sensor (110) and the point-type optical hydrogen sensor (120). The data collection and analysis system (200) includes an FPGA module capable of high-speed sampling board control, and the FPGA module includes a laser diode (LD) driving circuit, a temperature control (TEC) controller,A light source and a sensor are controlled through a hardware control module including a pulse drive controller and a digital-to-analog converter (DAC). The FPGA chip is equipped with a signal processing algorithm that performs phase analysis of an optical signal received from the sensing unit (100), converting a high-speed analog optical signal into a digital signal in real time and performing large-capacity signal collection and processing. Each digital data collected by the data collection and analysis system (200) is stored in an Influx DB (240), which is a time-series database. The Influx DB (240) is a database optimized for efficiently storing and querying large-capacity sensor data that is continuously generated over time. The data collection and analysis system (200) performs preprocessing on the raw data stored in the Influx DB (240), including noise removal, data normalization, and statistical characteristic analysis. During this process, noise filtering based on the characteristics of each sensor is performed, standardization is carried out to unify the scales of data with different units, and data augmentation is performed to compensate for the lack of training data. The preprocessed data is analyzed through an Edge AI algorithm. The Edge AI algorithm receives and analyzes collected time-series data and spatial distribution data using a model trained based on a Convolutional Neural Network (CNN) or Long Short-Term Memory (LSTM) model. The Edge AI algorithm comprehensively analyzes the correlations of the vibration, temperature, and hydrogen concentration data to distinguish and classify normal state data and abnormal state data in real time. For example, if the vibration signal detected by the linear optical vibration and temperature sensor (110) exceeds a preset threshold,The data collection and analysis system (200) compares and analyzes the trend of temperature data change at the location and the change in hydrogen concentration of the point-type optical hydrogen sensor (120) by synchronizing the time. Through this, it is possible to clearly distinguish whether the vibration is caused by simple external excavation work or an accident accompanied by gas leakage due to pipe damage. Specifically, it detects various risk factors in real time, including vibration patterns caused by external excavation work, shock patterns caused by changes in internal pipe pressure, concentration change patterns due to hydrogen leakage, and rapid temperature rise patterns due to fire. Normal state data and abnormal state data primarily determined by the data collection and analysis system (200) are transmitted to a diagnostic platform (300) for hydrogen gas facilities. The diagnostic platform (300) for hydrogen gas facilities stores and manages the received data and ultimately diagnoses the risk prediction and abnormal situation of the hydrogen gas facilities based on the event classification results by the Edge AI algorithm. The diagnosed results and related metadata are stored in an integrated database (320) that includes a relational database, Maria DB, and a time-series database (TSDB). The integrated database (320) efficiently manages structured data and time-series data and is utilized for future data analysis and history management. The diagnostic platform (300) for hydrogen gas facilities provides visualized monitoring information to the user through a web server (310). Through the web server (310), the user can intuitively check the real-time status of the facility, the location of abnormal signs, and the trend of changes in sensor data. In addition,The diagnostic platform (300) for the hydrogen gas facility described above can transmit analysis data to an external upper control platform (400) via a Rest API. The upper control platform (400) can perform the role of managing multiple facilities in an integrated manner or sharing information with relevant organizations in the event of a disaster. Additionally, the diagnostic platform (300) for the hydrogen gas facility described above includes an advanced function to continuously improve the accuracy of the analysis algorithm. Abnormal state data classified by the data collection and analysis system (200) is verified for consistency through a multi-stage inspection process including acquisition, refinement, labeling, and full inspection via a data inspection module. Subsequently, when feedback information regarding the results of on-site situation inspections and whether actual events have occurred is received, the labeling of existing training data is modified or new data is updated to reflect this. By using the updated dataset to retrain the Edge AI algorithm and distributing the retrained model back to the data collection and analysis system (200), a virtuous cycle structure is established that lowers the false alarm rate and increases diagnostic accuracy. Overall, the optical sensor-based diagnostic system for hydrogen gas facilities according to the present invention precisely detects minute hydrogen gas leaks and signs of facility abnormalities through the fusion analysis of specially designed optical fiber sensor packaging and multimodal sensor data, and can dramatically improve the safety of hydrogen infrastructure through real-time analysis based on edge computing and platform-based integrated management. As shown in FIG. 3,The optical sensor-based diagnostic system for hydrogen gas facilities according to the present invention adopts a hardware and firmware structure capable of high-speed, high-capacity signal processing to precisely monitor the physical condition of the hydrogen gas facilities and whether there is a gas leak. A data collection and analysis system (200) that performs the core signal processing function of the system collects minute optical signals transmitted from a sensing unit (100) installed at the site, converts them into digital data, and processes them into an analyzable form. The above data collection and analysis system (200) includes an FPGA module capable of high-speed sampling board control as a core component to process a wide range of data received in real time from a line-type optical vibration and temperature sensor (110) and a point-type optical hydrogen sensor (120). Looking specifically at the hardware configuration of the FPGA module mounted on the above data collection and analysis system (200), it is largely composed of an LD driving circuit area that drives and controls the light source, an FPGA chip area responsible for the logic operations of the entire system, a USB interface for external data transmission, and an SMA port for optical signal input and output. The LD driving circuit performs the role of supplying stable power to the laser diode and applying a driving signal, and ensures the stability of the light source output through high-precision current control. The above FPGA chip utilizes a Spartan series or equivalent high-performance programmable logic chip to rapidly convert analog signals received from the sensing unit (100) into digital signals and perform filtering and primary signal processing. The above SMA port is physically connected to the line-type optical vibration and temperature sensor (110) and the point-type optical hydrogen sensor (120) to interface with a photo detector that converts optical signals into electrical signals, orIt provides a path for outputting a trigger signal externally. The internal firmware and logic structure of the FPGA module has a structure in which various sub-controllers are organically connected around the main control logic. The main control logic plays a pivotal role in managing the timing of the entire system and controlling the flow of data, and performs bidirectional communication with the TEC controller, DAC controller, SPI ADC controller, and pulse driving controller. The TEC controller is logic for driving a thermo-electric cooler that controls the temperature of the laser diode. The laser diode has the characteristic that its oscillation wavelength fluctuates slightly depending on the temperature change, and such wavelength fluctuation can have a fatal effect on the measurement precision of the line-type optical vibration and temperature sensor (110). Accordingly, the TEC controller receives feedback on the thermistor value measured in real time from the temperature sensing circuit and controls the TEC driving circuit, thereby maintaining the temperature of the laser diode constant and ensuring wavelength stability of the light source. The DAC controller performs the function of finely adjusting the optical output of the laser diode by converting the digital signal into an analog voltage. Since the data acquisition and analysis system (200) must maintain optimal light intensity according to the measurement environment or the distance of the target equipment, it applies a driving current appropriate to the situation to the laser diode through the DAC controller. In addition, the pulse driving controller precisely controls the pulse width and repetition rate of the optical pulse incident on the line-type optical vibration and temperature sensor (110). This is an important factor in determining spatial resolution and measurement distance in a distributed sensing system,Precise pulse control in nanosecond (ns) units is performed according to the command of the main control logic above. The generated control signal passes through a digital output buffer and is transmitted to an external pump laser diode (Pump LD) via an SMA port to generate an optical pulse. The SPI ADC controller plays the role of converting and collecting analog sensor data received from the sensing unit (100) into digital data. The vibration data collection unit (210), hydrogen concentration data collection unit (220), and temperature data collection unit (230) process signals input from their respective sensors, and in this process, the SPI ADC controller within the FPGA module efficiently reads the data through high-speed serial communication. In particular, since minute signal changes occurring during a hydrogen gas leak or vibration patterns caused by external shocks occur within a very short period of time, the high-speed sampling function of the FPGA module is essential for capturing warning signs of an accident without missing them. The data collection and analysis system (200) transmits the data, which has been converted into digital and processed in the first stage through the FPGA module, to the Influx DB (240). The Influx DB (240) is a time-series database that stores and manages large volumes of data acquired from the vibration data collection unit (210), hydrogen concentration data collection unit (220), and temperature data collection unit (230) in chronological order. Data transmission between the FPGA module and the Influx DB (240) is carried out via a USB interface or Ethernet communication, thereby converting analog physical quantities at the site into a digital database without delay. The stored data is subsequently subjected to noise removal,The data undergoes preprocessing steps such as data normalization and statistical characteristic analysis, and is utilized as input data for an Edge AI algorithm. The diagnostic platform (300) for the hydrogen gas facility receives normal state data and abnormal state data analyzed by the data collection and analysis system (200) and stores them in an integrated database (320). High-quality data obtained through precise control of the FPGA module serves as a foundation for maximizing the learning and inference performance of the Edge AI algorithm. For example, data for phase analysis of the line-type optical vibration and temperature sensor (110) is demodulated in real time through the high-speed signal processing capability of the FPGA, which enables the diagnostic platform (300) for the hydrogen gas facility to accurately identify minute deformations or leak locations in the piping. In addition, since the data of the point-type optical hydrogen sensor (120) is also collected with noise minimized through the FPGA module, the diagnostic platform (300) for the hydrogen gas facility can quickly determine whether there is a hydrogen gas leak without false alarms. The FPGA module of the data collection and analysis system (200) performs not only the function of simply collecting data but also the function of self-diagnosis and status monitoring of the system. The main control logic monitors the operating status of each sub-controller in real time and reports to the diagnostic platform (300) for the hydrogen gas facility if signs of aging of the laser diode or abnormality in the temperature control circuit are detected. Such hardware-level status diagnostic information is provided to the user through the web server (Web Server, 310) or transmitted to an upper control platform (400) via a Rest API to be used for establishing a maintenance plan for the equipment. Overall, the FPGA-based high-speed, high-capacity signal acquisition and control module applied in the present invention enables precise control of a laser light source, high-speed data sampling,Furthermore, real-time signal processing is performed integrally within a single hardware. This provides a technical foundation for the diagnostic platform (300) for hydrogen gas facilities to detect risk factors of hydrogen gas facilities early and accurately diagnose them by processing a vast amount of multimodal data (vibration, temperature, hydrogen concentration) acquired from the sensing unit (100) without delay and transmitting it to the influx DB (240) and the integrated database (320). As illustrated in FIG. 4, the optical sensor-based diagnostic system for hydrogen gas facilities according to the present invention introduces Edge-AI technology to perform an advanced process that predicts dangerous situations of hydrogen gas facilities early and precisely detects abnormal signs. The process is systematically carried out by broadly dividing it into a field demonstration data collection stage, a data preprocessing and abnormal pattern detection stage, a data processing and event labeling stage, a complex analysis algorithm optimization stage, and an inspection process stage for the advancement of the analysis algorithm. This series of processes is performed through the organic linkage of a data collection and analysis system (200) and a diagnostic platform (300) for hydrogen gas facilities, and constitutes a core process in which physical data acquired from a sensing unit (100) is converted into actual safety diagnostic information. First, regarding the field verification data collection stage, the sensing unit (100) detects the physical condition of the site in real time through line-type optical vibration and temperature sensors (110) laid across the entire length of the hydrogen gas facility and point-type optical hydrogen sensors (120) installed at key locations. At this time, the vibration data collection unit (210), hydrogen concentration data collection unit (220), and temperature data collection unit (230) within the data collection and analysis system (200) each collect raw data regarding temperature distribution, vibration distribution, and hydrogen exposure. The collected data includes both time-series characteristics and spatial distribution characteristics, andThe data collection and analysis system (200) stores this in a time-series database, the Influx DB (240), to build a large dataset for subsequent analysis. Next, in the noise removal, data normalization, statistical analysis, and anomaly pattern detection steps, preprocessing is performed to improve the quality of the raw data stored in the Influx DB (240). Since various environmental noises, such as mechanical vibrations or changes in ambient temperature, can be mixed into the optical signal in the field environment, the data collection and analysis system (200) applies filtering technology to extract only valid signal components. In addition, since the vibration data, temperature data, and hydrogen concentration data have different physical units and scales, they undergo normalization and standardization processes so that an artificial intelligence model can learn them effectively. The data collection and analysis system (200) converts preprocessed time-series data and spatial distribution data into a two-dimensional or three-dimensional matrix form to form visual patterns, and primarily detects abnormal patterns that deviate from the average range through statistical analysis. Subsequently, data processing, feature extraction, and event labeling steps are performed. The data collection and analysis system (200) extracts unique feature values ​​from the preprocessed data that can represent the state of the equipment. The extracted feature values ​​are classified into similar patterns through a clustering algorithm, and specific event labeling is performed based on this. The event classes defined in the present invention are subdivided in various ways, including signs of temperature anomalies, normal operation of hydrogen valves, signs of vibration anomalies, leak diffusion patterns, external excavation and perforation, hydrogen leakage and abnormal vibration, normal operation of hydrogen tanks, hydrogen leakage and abnormal temperature, normal operation of hydrogen piping, and micro-hydrogen leakage. Such detailed labeling goes beyond simple anomaly detection,It enables a clear distinction as to whether the anomaly is caused by excavation work or by an actual gas leak. In the complex analysis algorithm optimization stage, the training and optimization of a deep learning model are performed based on the extracted feature values ​​and labeled data. The data collection and analysis system (200) utilizes high-performance learning models such as Transformer, LSTM, or Convolutional Neural Network (CNN). The input layer of the learning model receives not only temperature distribution data, vibration distribution data, and hydrogen sensor data, but also metadata including the location or installation time of the equipment. The learning model comprehensively analyzes the correlations between these multimodal data to output classification results and classification reliability for each event class. For example, when a strong impact pattern is detected in the vibration distribution data, if no increase in concentration is simultaneously observed in the hydrogen sensor data, it is classified as external excavation and perforation; however, if it is accompanied by an increase in concentration, it is classified as hydrogen leakage and abnormal vibration, thereby maximizing the accuracy of the diagnosis. Finally, the inspection process step for the advancement of the analysis algorithm is a virtuous feedback process performed centered on the diagnostic platform (300) for hydrogen gas facilities. The diagnostic platform (300) for hydrogen gas facilities reviews the learning results through the Confusion Matrix, which is the result of the learned model. If a false alarm or a undetected event occurs at the site, the actual event is verified through a site situation inspection, and the labeling data is modified or updated based on this. The verified data accumulated in the integrated database (320) is utilized again as training data to contribute to algorithm optimization. That is, the diagnostic platform (300) for hydrogen gas facilities continuously collects site feedback and retrains the edge artificial intelligence algorithm of the data collection and analysis system (200),An evolutionary safety management system is established in which the diagnostic performance of the system improves and the false alarm rate decreases over time. As illustrated in FIG. 5, the optical sensor-based diagnostic system for hydrogen gas facilities according to the present invention includes a diagnostic platform (300) for hydrogen gas facilities that systematizes a series of processes ranging from data collection to processing, purification, labeling, and final inspection to ensure the quality of training data for an artificial intelligence model and maximize diagnostic accuracy. The platform (300) establishes a cyclical data management system that goes beyond simply collecting physical data from the site to verifying data consistency and providing feedback on errors to continuously advance the analysis model. This data processing process is broadly divided into a data collection stage, a data processing and purification stage, and a data inspection and feedback stage. First, regarding the data collection stage, the line-type optical vibration and temperature sensor (110) and the point-type optical hydrogen sensor (120) constituting the sensing unit (100) detect physical signals generated in real time from the piping and key points of the hydrogen gas facilities. The above-described line-type optical vibration and temperature sensor (110) detects vibration and temperature changes distributed along the extended path of the hydrogen pipeline in the form of optical signals, and the above-described point-type optical hydrogen sensor (120) precisely measures changes in hydrogen concentration at local points such as valves or joints. The analog optical signals detected in this way are transmitted to the data collection and analysis system (200), and the vibration data collection unit (210), hydrogen concentration data collection unit (220), and temperature data collection unit (230) within the system (200) convert each signal into digital data and collect it. The data collected at this time goes beyond simple numerical data, and includes time and distance, as shown in the drawing,Alternatively, it can be visualized as multidimensional image data in the form of a spectrogram or heatmap containing frequency components. The data collection and analysis system (200) establishes a raw data repository by storing this large volume of unstructured data in an Influx DB (240), which is a time-series database. Next, in the data processing, refinement, and labeling stage, a process is performed to convert the raw data accumulated in the Influx DB (240) into a form suitable for artificial intelligence learning. The data collection and analysis system (200) removes noise and performs normalization from the collected image-type signal data, and then performs a labeling operation to identify and tag the type of event that the data represents. The data generated in this process takes on a structured file format such as JSON, and each file includes meta-information such as the sensor identification number, time of occurrence, type of signal (normal, excavation vibration, gas leak, etc.), and signal intensity. This structured data functions as a standardized input dataset that the Edge AI algorithm can learn from. In particular, the present invention features a multi-stage data inspection process to ensure data reliability. The data inspection steps performed on the diagnostic platform (300) for hydrogen gas facilities are subdivided into a first inspection (acquisition), a second inspection (refining), a third inspection (labeling), and a fourth inspection (full inspection). In the first inspection stage, the acquisition inspection stage, it is verified whether the data collected from the data collection and analysis system (200) has been acquired normally without omission, and whether the file format or size meets the specifications. If data loss or format errors are detected at this stage,The data is classified as non-compliant and rejected. In the subsequent second inspection, the purification inspection stage, it is determined whether environmental noise or outliers included in the data have been properly removed. Since hydrogen gas facility sites are exposed to various external factors, there is a high possibility that background noise unrelated to actual events will be mixed into the signal. Therefore, the diagnostic platform (300) for the hydrogen gas facility analyzes the signal-to-noise ratio (SNR), etc., to filter out data whose purification quality does not meet the standards, and feeds this back to the purification process. In the third inspection, the labeling inspection stage, an expert or verification algorithm reviews the accuracy of the label assigned to the data. For example, it checks whether a vibration pattern caused by simple vehicle traffic has been incorrectly labeled as an impact pattern caused by pipe perforation, and corrects it if there is an error. Finally, in the fourth inspection stage, the full inspection stage, the consistency of the data is finally approved through random sampling or a full inspection of the entire final dataset that has passed the preceding inspection process. The analysis results and characteristics of the error data derived from this inspection process are transmitted to the data collection and analysis system (200) through a feedback loop. The diagnostic platform (300) for hydrogen gas facilities collects information on misclassification cases or new patterns discovered during the inspection process to build a casebook, and based on this, adjusts the parameters of the Edge AI algorithm or performs retraining. In other words, the inspection stage acts not only as a filter to simply screen out defective data, but also as a teacher that continuously makes the AI ​​model smarter. Finally, the high-quality data that has completed the inspection is stored and managed in the integrated database (320). The integrated database (320) is composed of a relational database, Maria DB, and a time-series database (TSDB).It efficiently stores standardized inspection history information and large-capacity sensor data. The diagnostic platform (300) for hydrogen gas facilities visually provides the user with statistics on completed inspection data, trends in event occurrence, and the status of data inspection through the web server (310). In addition, the data stored in the integrated database (320) is linked with the upper control platform (400) and can be utilized as basic data for evaluating the safety of hydrogen facilities in a national disaster management system or a wide-area control center. According to one embodiment of the present invention, a specific process for distinguishing between a micro-leakage signal caused by hydrogen molecule characteristics and noise caused by external environmental factors is described. First, the sensing unit (100) detects physical data in real time through a line-type optical vibration and temperature sensor (110) and a point-type optical hydrogen sensor (120) installed at the site of a hydrogen gas facility. The line-type optical vibration and temperature sensor (110) detects vibrations caused by external impacts, such as excavation work or vehicle traffic around the pipe, and simultaneously measures changes in ambient temperature, while the point-type optical hydrogen sensor (120) detects localized changes in hydrogen concentration around pipe joints or valves. At this time, the data collection and analysis system (200) collects and digitizes the optical signal transmitted from the sensing unit (100) through the vibration data collection unit (210), the hydrogen concentration data collection unit (220), and the temperature data collection unit (230), and stores it in the time-series database, the Influx DB (240). The data collection and analysis system (200) performs an analysis to distinguish between noise and actual leakage based on the data stored in the Influx DB (240). Specifically, hydrogen gas has the characteristic that the temperature change due to the Joule-Thomson effect is weak or has a negative value when leaking, making it difficult to detect with a temperature sensor alone, and there is a problem that external excavation work may be mistaken for pipe damage with a simple vibration sensor alone. To resolve this, the data collection and analysis system (200) analyzes the rate of change of data of the point-type optical hydrogen sensor (120) at the same time, using this as a trigger when a vibration pattern exceeding a threshold is detected in the line-type optical vibration and temperature sensor (110). If a vibration signal is generated strongly but there is no change in the hydrogen concentration collected through the hydrogen concentration data collection unit (220), the data collection and analysis system (200) classifies this as non-hazardous noise caused by an external environment (excavation, vehicles, etc.).On the other hand, if a pattern is detected in which the concentration level of the point-type optical hydrogen sensor (120) rises along with minute vibrations or temperature changes, this is determined to be a minute leak of hydrogen molecules. The result determined in this way is transmitted to a diagnostic platform (300) for hydrogen gas facilities, stored in an integrated database (320), and managed as a history. If the determined event is an actual dangerous situation, the diagnostic platform (300) for hydrogen gas facilities immediately displays a visualized alarm to the manager via a web server (Web Server, 310), and if necessary, transmits data to a higher-level control platform (400) to enable a wide-area response. This multimodal sensor fusion and cross-verification method significantly reduces false alarms that occur in single-sensor methods and provides the effect of precisely diagnosing minute leaks unique to hydrogen gas. The specific implementation details of a diagnostic platform (300) for hydrogen gas facilities and a database structure constituting the same, according to one embodiment of the present invention, are described. The present invention presents an optimized data modeling to solve the problems of reduced processing speed of large-capacity sensor data and compromise of data integrity when performing safety monitoring of hydrogen gas facilities. The entire system according to an embodiment of the present invention comprises a sensing unit (100) that detects physical changes in the field, a data collection and analysis system (200) that processes signals obtained from the sensing unit (100), and a diagnostic platform (300) for hydrogen gas facilities that stores and manages analyzed data. The sensing unit (100) includes a line-type optical vibration and temperature sensor (110) that detects minute vibrations and temperature changes in the piping, and a point-type optical hydrogen sensor (120) that detects whether hydrogen gas is leaking. The above data collection and analysis system (200) includes a vibration data collection unit (210) that converts an analog signal received from the sensing unit (100) into digital data, a hydrogen concentration data collection unit (220), and a temperature data collection unit (230), and large volumes of raw data generated in a time series are stored in the influx database (240) to process high-speed input / output. Meanwhile, the diagnostic platform (300) for hydrogen gas facilities includes a web server (310) that provides visualized information to the user and an integrated database (320) that stores system operation data and analysis results in a relational structure. The integrated database (320) has a schema structure logically separated into a master data area, an immutable transaction area, a variable transaction area, and a system state snapshot area according to the nature of the data, which is intended to optimize query performance and prevent data locking when the web server (310) provides information to the upper control platform (400) or the user.To explain the structure of the integrated database (320) in detail, first, the master data area includes a Zones table that manages physical zone information and a Sensors table that manages hardware information of the sensing unit (100). The Zones table stores location coordinates containing location information of each facility and zone names, and has a unique identifier ID as the primary key. The Zones table forms a one-to-many relationship with the Sensors table to structure the placement of multiple sensors within a single zone. The Sensors table stores the unique hardware serial number of the line-type optical vibration and temperature sensor (110) or the point-type optical hydrogen sensor (120) as a Unique Key to fundamentally prevent duplicate registration. Additionally, the Sensors table includes a Sensor Type that distinguishes the type of sensor, and a Threshold Min and Threshold Max to prevent false alarms. The threshold settings stored in the sensor table are used as reference data for the data collection and analysis system (200) to determine events, and when an administrator changes the settings through the web server (310), the update time is recorded in the last status change date (Updated At) field. The sensor table specifies whether the equipment is active or under maintenance through the current status field, thereby providing a function to filter data generated from sensors under maintenance so that it does not trigger unnecessary alarms. One of the key features of the integrated database (320) is that it is designed to separate the artificial intelligence detection results from the operational workflow.The AI ​​Detections table located in the aforementioned immutable transaction area serves as a repository for recording all events detected by the data collection and analysis system (200), and follows an 'Insert Only' policy that prevents data modification. The AI ​​Detections table assigns a unique ID of type BigInt to each detection case and identifies which sensor the event occurred from by referencing the ID of the sensor table as a foreign key. The AI ​​Detections table includes a Detection Type, such as leakage or excavation, a Confidence Score calculated by the AI ​​model, and a Detected At, which is the exact time the event occurred. In particular, a database index is set on the Detected At field to ensure search speed when the web server (310) queries the history of a specific period. Additionally, the AI ​​Detections table includes a Raw Data Snapshot field in JSON format. The above raw data snapshot field stores a summary of sensor waveform data before and after the time of event occurrence, thereby reducing the load on the user, who would otherwise have to access the above Influx DB (240) every time to query large amounts of data for detailed analysis. This structure enables high-speed storage without data loss by preventing update transactions resulting from administrator verification from locking the table while the data collection and analysis system (200) continuously inserts new detection results. The Alert Workflow table, which is connected in a one-to-one relationship with the above artificial intelligence detection table, corresponds to a variable transaction area and tracks the administrator's action process regarding the detected event.The aforementioned alarm workflow table maintains strong integrity between the two tables by using the ID of the AI ​​detection table as both a primary and foreign key. The alarm workflow table includes a Workflow Status field indicating progress statuses such as New, Under Investigation, Action Completed, or False Detection, which is frequently updated by administrator operations. Additionally, the alarm workflow table includes the Actual Severity Level designated by the administrator considering the field situation, separate from the reliability determined by the AI; the Acknowledged At time, which is the time the administrator recognized the alarm; and the Handler ID responsible for processing, which is Acknowledged By. The specific details of the actions taken by the administrator at the site are stored in text form in the Admin Note field. By separating workflow information, which undergoes such frequent modifications, into a separate table, the integrated database (320) logically and physically isolates the process of recording a large amount of sensing data from the manager's operation process, thereby ensuring the stability of the entire system. Finally, the integrated database (320) includes a Device Current Status table as a system status snapshot area. The Device Current Status table has a one-to-one relationship with the sensor table and has a structure that maintains (upserts) only the status value at the current point in time, rather than accumulating all past history. The Device Current Status table stores the Last Heartbeat and Connection Status, which indicate whether communication between the sensing unit (100) and the data collection and analysis system (200) is being conducted normally.In addition, it includes an FPGA status (Is FPGA OK) field indicating whether the FPGA, which is an internal computing device of the line-type optical vibration and temperature sensor (110), is operating normally, and an LD status (Is LD OK) field indicating the health of the laser diode that generates the optical signal. When configuring the dashboard screen, the web server (310) can immediately visualize the health of the entire system and transmit it to the upper control platform (400) without scanning a history table containing millions of accumulated logs by performing a single query of only the equipment current status table. This database design enables real-time monitoring performance and data management efficiency to be achieved simultaneously in environments requiring high reliability, such as hydrogen gas facilities, through a separate storage and access strategy based on the nature of the data. Explanation of the symbols 100: Sensing Unit 110: Line-type optical vibration and temperature sensor 120: Point-type optical hydrogen sensor 200: Data acquisition and analysis system 210: Vibration data acquisition unit 220: Hydrogen concentration data acquisition unit 230: Temperature data acquisition unit 240: Influx DB 300: Diagnostic platform for hydrogen gas facilities 310: Web Server 320: Integrated database 400: Upper-level control platform

Claims

Claim 1 A diagnostic system for a hydrogen gas facility based on optical sensors for real-time monitoring of the physical condition and gas leakage of a facility including piping, valves, and storage tanks of a hydrogen gas facility, comprising: a line-type optical vibration and temperature sensor (110) installed along an extended path of the hydrogen gas facility to detect vibration and temperature changes in a distributed manner, and a point-type optical hydrogen sensor (120) installed at a key point including a valve connection or joint of the hydrogen gas facility to detect localized hydrogen gas leakage, thereby detecting changes in the physical condition of the hydrogen gas facility as an optical signal; a sensing unit (100) that detects changes in the physical condition of the hydrogen gas facility as an optical signal, and a vibration data collection unit (210) connected to the sensing unit (100) via an optical path to collect vibration signals detected from the line-type optical vibration and temperature sensor (110), a hydrogen concentration data collection unit (220) that collects hydrogen concentration signals detected from the point-type optical hydrogen sensor (120), and a temperature data collection unit (230) that collects temperature signals detected from the line-type optical vibration and temperature sensor (110), and capable of high-speed sampling board control. A data collection and analysis system (200) that converts the optical signal into a digital signal through an FPGA module, stores each collected data in a time-series database called an Influx DB (240), performs preprocessing including noise removal, data normalization, and statistical characteristic analysis on the data stored in the Influx DB (240), and then analyzes the data using an Edge AI algorithm to determine, classify, and output normal state data and abnormal state data in real time;A diagnostic system for a hydrogen gas facility based on an optical sensor, comprising: a data collection and analysis system (200), wherein, in order to compensate for the characteristic of hydrogen molecules having a weak temperature change due to the Joule-Thomson effect when hydrogen gas leaks, when a vibration pattern exceeding a preset threshold is detected in the line-type optical vibration and temperature sensor (110), the system performs a cross-verification logic that time-synchronizes and compares the rate of change of hydrogen concentration data of the point-type optical hydrogen sensor (120) at the same time using the vibration pattern as a trigger; wherein if there is no change in the hydrogen concentration when the vibration pattern occurs, it is classified as non-hazardous noise caused by the external environment, and if a pattern in which the hydrogen concentration value rises along with the vibration pattern or temperature change is detected, it is determined as a micro-leakage accident of hydrogen molecules. Claim 2 A diagnostic platform (300) for a hydrogen gas facility based on an optical sensor, further comprising: receiving, storing, and managing the normal state data and the abnormal state data determined from the data collection and analysis system (200); predicting risks and diagnosing abnormal situations of the hydrogen gas facility based on event classification results by the Edge AI algorithm; storing the diagnosed results and metadata in an integrated database (320) including a relational database Maria DB and a time series database (TSDB); providing visualized monitoring information to a user through a web server (310); or transmitting integrated diagnostic data including the diagnosed results and the event classification results to an external upper control platform (400) through a Rest API. Claim 3 In claim 1, the data collection and analysis system (200) analyzes the correlation of vibration, temperature, and hydrogen concentration data, which are multimodal data collected at the site and stored in the influx DB (240), to detect risk factors in real time, including vibration patterns caused by external excavation work, shock patterns caused by changes in internal pipe pressure, concentration change patterns caused by hydrogen leakage, and rapid temperature rise patterns caused by fire, and transmits event information regarding the detected risk factors to the diagnostic platform (300) for hydrogen gas facilities so that an alarm is performed. This characterizes an optical sensor-based diagnostic system for hydrogen gas facilities. Claim 4 In paragraph 2, the diagnostic platform (300) for hydrogen gas facilities further comprises: a data inspection module that receives the abnormal state data primarily classified by the data collection and analysis system (200), sets it as data to be inspected, and verifies the consistency of the data through a multi-stage inspection process including acquisition, purification, labeling, and full inspection; and an analysis algorithm advancement module that performs retraining of the Edge AI algorithm based on the data verified through the data inspection module, and updates and distributes the retrained algorithm to the data collection and analysis system (200) to reduce the false alarm rate and improve diagnostic accuracy; characterized in that it is a diagnostic system for hydrogen gas facilities based on optical sensors. Claim 5 In the method for diagnosing a hydrogen gas facility based on an optical sensor-based diagnostic system for a hydrogen gas facility according to claim 1, the sensing unit (100) detects vibration and temperature changes in a distributed manner through a line-type optical vibration and temperature sensor (110) laid along an extended path of the hydrogen gas facility, and detects whether there is a local hydrogen gas leak through a point-type optical hydrogen sensor (120) installed at a major point of the hydrogen gas facility, thereby generating an optical signal (S100); the data collection and analysis system (200) receives the optical signal generated from the sensing unit (100), and collects vibration, hydrogen concentration, and temperature data through a vibration data collection unit (210), a hydrogen concentration data collection unit (220), and a temperature data collection unit (230), respectively, and stores them in an Influx DB (240), which is a time-series database (S200); the data collection and analysis system (200) performs noise removal, data normalization, and on the raw data stored in the Influx DB (240). A step (S300) of processing the data into a form for Edge AI learning and inference by performing preprocessing including vectorization; a step (S400) in which the data collection and analysis system (200) inputs the preprocessed data into an Edge AI algorithm for analysis, thereby classifying and determining in real-time risk factor patterns including external excavation, impact, hydrogen leakage, and fire, and normal state patterns; and a step (S500) in which a diagnostic platform (300) for hydrogen gas facilities receives normal and abnormal state data determined from the data collection and analysis system (200), stores it in an integrated database (320), diagnoses a dangerous situation based on the event classification results in which the Edge AI classifies risk factor patterns including external excavation, impact, hydrogen leakage, and fire, and provides visualized information through a web server (310) or transmits the analysis data to an external upper control platform (400) to generate an alarm.A diagnostic method for an optical sensor-based hydrogen gas facility, comprising: a step (S400) which analyzes the correlation of vibration, temperature, and hydrogen concentration data to exclude false detections occurring when using only one of a line-type optical vibration and temperature sensor or a point-type optical hydrogen sensor, and time-synchronizes the rate of change in hydrogen concentration using a vibration pattern as a trigger to distinguish between micro-leakage due to hydrogen molecule characteristics and noise caused by the external environment to detect risk factors. Claim 6 In claim 5, the above step (S300) comprises: a step of performing noise filtering according to sensor-specific characteristics on time series data and spatial distribution data extracted from the influx DB (240); a step of performing normalization and standardization to unify the scale of the filtered noise data; and a step of vectorizing the noise filtered data to match the input dimension of the Edge AI algorithm and performing data augmentation to compensate for the lack of training data; characterized in that it is a diagnostic method for a hydrogen gas facility based on an optical sensor. Claim 7 In claim 5, prior to the above step (S400), a step of constructing a training dataset by performing labeling by event class including abnormal temperature signs, normal operation of hydrogen valves, abnormal vibration signs, leak diffusion patterns, external excavation and perforation, hydrogen leakage and abnormal temperature, normal operation of hydrogen piping and micro hydrogen leakage by applying a time-synchronized complex correlation analysis to the collected vibration, temperature and hydrogen concentration data, using the exceedance of a threshold of the vibration signal as a trigger according to cross-validation logic; and a step of training a deep learning model based on a Convolutional Neural Network (CNN) or Recurrent Neural Network (RNN) using the constructed training dataset, and loading the trained model as an Edge AI algorithm into the data collection and analysis system (200). Claim 8 In claim 5, the above step (S400) is characterized by determining whether the vibration is caused by simple excavation work or by distinguishing between an accident accompanied by gas leakage due to pipe damage by time-synchronizing and comparing the temperature data change trend at the corresponding location and the hydrogen concentration change of the point-type optical hydrogen sensor (120) when the vibration signal detected by the line-type optical vibration and temperature sensor (110) exceeds a preset threshold.

Images