Intelligent ultraviolet flame detection and early warning system and method suitable for complex industrial scene

Abstract

The invention relates to the technical field of fire detection and early warning, in particular to an intelligent ultraviolet flame detection and early warning system and method suitable for complex industrial scenes. Comprising distributed multispectral ultraviolet detection nodes and a central processing early warning unit. The nodes carry out signal acquisition and preliminary identification through a multispectral sensor, local processing and high-precision time synchronization; and the central unit fuses multi-node data, performs three-dimensional flame positioning by using a TDOA algorithm, performs machine learning recognition and false alarm elimination, and performs trajectory tracking by using a Kalman filter so as to realize graded early warning linkage. According to the intelligent ultraviolet flame detection and early warning system and method suitable for the complex industrial scene, the problems that in the prior art, the false alarm rate is high, flame positioning is not accurate, trajectory tracking is missing, cooperative monitoring is limited and the like are solved or at least relieved, false alarms are effectively reduced, and the positioning precision, the tracking capacity and the early warning reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of fire detection and early warning technology, and in particular to an intelligent ultraviolet flame detection and early warning system and method applicable to complex industrial scenarios. Background Technology

[0002] As modern industrial production processes become increasingly complex and automated, the fire hazards they face are also becoming more diverse and complex. Ultraviolet (UV) flame detection and alarm systems are a crucial line of defense for industrial safety, and their importance is self-evident. UV flame detectors utilize the principles of ultraviolet and visible light to detect UV and visible radiation in flames, achieving non-contact detection. Compared to traditional smoke and heat detectors, they offer advantages such as faster response and adaptability to detecting some smokeless or low-smoke fires. Although this technology has undergone many years of development and made significant progress, it still faces serious challenges, especially in complex industrial environments, including constantly changing environments, significant interference, and centimeter-level positioning requirements.

[0003] Existing technologies have attempted to address the problem of industrial fire detection using various methods. Patent CN113593171B discloses a smoke detection and early warning method and system for cable tunnel spaces. It employs a video image acquisition module and a flame detection module, using red and ultraviolet light for flame detection, and then combining pre- and post-fire image information for early warning. This solution fuses two different modalities of information: optical images and radiation detection at a specific frequency band. Compared to single-mode fire detection, it offers a certain improvement in judgment criteria and provides a relatively intuitive early warning method for completely enclosed environments like cable tunnels where fires are difficult to detect, representing an early manifestation of multimodal information fusion detection. However, the core of this technical solution still relies on a relatively single ultraviolet detection method for flame identification, lacking multispectral analysis capabilities. It is easily misled by non-flame ultraviolet light sources in complex electromagnetic environments and high-temperature industrial operations, such as electric arcs, strong ultraviolet radiation from welding, thermal radiation from high-temperature furnaces, and ultraviolet radiation from certain gas discharges. These non-flame events can lead to system misjudgment and a high false alarm rate. More importantly, the system does not involve three-dimensional spatial positioning of the flame source or flame trajectory tracking. In large industrial plants, tank areas, or multi-story production facilities, relying on detection from a single or few perspectives makes it difficult to quickly and accurately locate the fire source, let alone predict the spread of the flame. This undoubtedly greatly reduces the efficiency and accuracy of fire emergency response.

[0004] Building upon this, another patent document, CN113065421B, discloses a multi-source optical fire detection method and system for oil-filled equipment. By acquiring video, ultraviolet, and infrared signals from the oil-filled equipment, filtering, amplifying, preprocessing, and comparing with historical data, it determines whether a fire has occurred and then issues an alarm. This patent represents a significant breakthrough in addressing the issue of a single information source. By introducing multi-source optical signal processing of infrared and video signals, it improves robustness to fire events through the complementarity and cross-validation of different spectral information, and to some extent reduces the false alarm rate. It focuses on the precise monitoring of specific types of equipment, using a multi-sensor information fusion approach to identify localized fire events, representing an exploration of specialized protection for specific flammable equipment. However, this technical solution has limitations. Its core technological concept remains limited to fire detection of single equipment or localized areas, and the system architecture does not incorporate the concept of distributed detection nodes. In other words, the system cannot integrate multiple independent detection units into a unified whole, thus failing to form a global flame monitoring network in large-scale or heavily obstructed industrial environments to achieve comprehensive flame monitoring and location. In addition, due to the lack of inter-node correlation and time synchronization of spatial data, the system has weak flame tracking capabilities and cannot accurately predict the speed and direction of flame spread. Therefore, it is not very applicable in situations where large-scale, real-time dynamic complex industrial environments are required for flame monitoring and control.

[0005] The shortcomings of the aforementioned existing technologies lie in their inability to simultaneously address the challenges of high fire warning accuracy, low false alarm rate, and poor predictive ability for the dynamic spread of fire flames in complex industrial scenarios, while pursuing rapid response ultraviolet flame warning systems. This is primarily because the rapid response principle of ultraviolet flame warning systems makes them sensitive to non-flame ultraviolet light sources, and the lack of refined spectral resolution prevents them from distinguishing different ultraviolet light sources, leading to a high false alarm rate. Furthermore, even with multi-source optical signals, without establishing a distributed, high-precision time-synchronized detection network and combining it with spatial positioning algorithms, it is impossible to overcome the perspective limitations of single or local detection units, accurately locate the fire source in three-dimensional space, or track the spread of fire flames. The point or surface-based local perspective approach cannot meet the demands for global, dynamic, and high-precision fire monitoring. In complex industrial scenarios, the density of industrial equipment, the multi-layered structure, the dynamic and ever-changing sources of fire hazards, and the rapid changes in environmental parameters make it impossible to meet the increasingly stringent application requirements—from simple fire warnings to timely and comprehensive fire prevention—through simple detection overlay. False alarms reduce resource utilization, and long-term resource waste may lead to a wolf-sheep effect, reducing operators' attention to fire warning signals; missed alarms or inaccurate positioning will delay the best time to extinguish the fire, causing the disaster to escalate and resulting in incalculable losses.

[0006] Therefore, in order to solve the above-mentioned technical problems, those skilled in the art urgently need to provide an intelligent early warning system and method that can avoid interference from non-flame ultraviolet light sources, achieve refined identification of flame types in fire early warning, have high three-dimensional spatial positioning accuracy in fire early warning, and be able to dynamically track the flame spread trend. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent ultraviolet flame detection and early warning system and method applicable to complex industrial scenarios. This system can solve or at least mitigate problems existing in complex industrial environments, such as high false alarm rates due to non-flame ultraviolet source interference, insufficient accuracy in three-dimensional spatial positioning of flame sources, lack of dynamic flame trajectory tracking capabilities, and limitations in distributed system collaborative monitoring. This invention constructs a distributed multispectral ultraviolet detection network, supplemented by a high-precision time synchronization mechanism, advanced signal processing, multi-node data fusion, and three-dimensional spatial positioning and trajectory prediction algorithms, thereby achieving accurate flame identification, rapid location, dynamic tracking, and graded early warning.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent ultraviolet flame detection and early warning system suitable for complex industrial scenarios, comprising: Multiple distributed multispectral ultraviolet detection nodes, each of which integrates: The ultraviolet detection module is equipped with at least three narrowband ultraviolet sensor arrays. The spectral response bands of each narrowband ultraviolet sensor in the array are independent of each other and are used to capture specific narrowband ultraviolet radiation signals from the environment. The narrowband ultraviolet sensors include a first narrowband ultraviolet sensor for detecting characteristic ultraviolet radiation of flames, a second narrowband ultraviolet sensor for detecting wide-range radiation of flames, and a third narrowband ultraviolet sensor for collecting and analyzing environmental background noise to distinguish non-flame interference sources. The local data processing unit is electrically connected to the ultraviolet detection module and is used to receive the signal output by the ultraviolet detection module and perform preliminary flame signal judgment. A high-precision time synchronization module, electrically connected to the local data processing unit, is used to provide a high-resolution, precise timestamp for the initially determined potential flame signal data; and The wireless communication module is electrically connected to the local data processing unit and is used to upload the preliminary determined potential flame signal data. A central processing and early warning unit is used to receive and process data from the plurality of ultraviolet detection nodes, the central processing and early warning unit comprising: The data receiving and timestamp alignment module is used to continuously receive wireless data from the multiple ultraviolet detection nodes, and to precisely align the received data from the multiple different ultraviolet detection nodes according to the precise timestamp embedded in the data packet. The data fusion and spatial positioning module is electrically connected to the data receiving and timestamp alignment module. It is used to receive precisely aligned data and use the pre-stored precise three-dimensional geographical location information of all the ultraviolet detection nodes to calculate the three-dimensional spatial coordinates of the flame source through the Time Difference of Arrival (TDOA) algorithm. A flame identification and early warning module, electrically connected to the data fusion and spatial positioning module, is used to receive local spectral feature parameters from the ultraviolet detection node and the three-dimensional spatial coordinates of the flame source calculated by the data fusion and spatial positioning module. It then employs an intelligent identification algorithm for high-level flame identification, type judgment, and false alarm elimination, triggering corresponding early warning levels based on a preset risk assessment strategy. The dynamic trajectory tracking module is electrically connected to the data fusion and spatial positioning module. It is used to continuously receive the real-time three-dimensional spatial coordinates of the flame source from the data fusion and spatial positioning module, and to use a dynamic filtering algorithm to estimate and predict the motion state of the flame source in order to construct the dynamic spread trajectory of the flame.

[0009] To further realize the present invention, the following technical solutions may be preferred: Preferably, the ultraviolet detection module further includes a low-noise amplifier for amplifying the weak current signal output by the ultraviolet detection module; the narrowband ultraviolet sensor uses a photodiode based on a wide bandgap semiconductor material and integrates a filter and an optical collimating lens.

[0010] Preferably, the local data processing unit includes: An analog-to-digital converter is used to convert the analog signal output by the ultraviolet detection module into a digital signal; A digital signal processor is configured to perform real-time digital filtering, denoising, and local spectral feature extraction on the digital signal. The local spectral feature extraction includes calculating the intensity ratio, transient scintillation frequency, and signal rise time among the signals acquired by the different narrowband ultraviolet sensors. The microcontroller is used to compare the extracted spectral feature parameters with a preset flame discrimination model to perform preliminary flame signal judgment. The flame discrimination model includes a preset intensity ratio threshold range, a transient flicker frequency threshold range, and a signal rise time threshold.

[0011] Preferably, the high-precision time synchronization module has a built-in high-stability crystal oscillator to provide a highly stable local clock; the high-precision time synchronization module is periodically calibrated through a global time source to ensure that the timestamps of all detection nodes achieve high precision.

[0012] A method for intelligent ultraviolet flame detection and early warning applicable to complex industrial scenarios includes the following steps: S1: Distributed multispectral ultraviolet signal acquisition and local preliminary identification, including: S101: Deployment of detection nodes and high-precision time synchronization, including strategically deploying multiple ultraviolet detection nodes with known accurate three-dimensional geographic coordinates in the area to be monitored in a complex industrial scenario, and periodically calibrating the high-precision time synchronization module inside each ultraviolet detection node with a global time source. S102: Real-time acquisition and local signal processing of multispectral ultraviolet radiation, including real-time parallel acquisition of radiation intensity signals of different narrowband ultraviolet bands in the environment by the narrowband ultraviolet sensor array in each ultraviolet detection node, and analog-to-digital conversion, digital filtering and noise reduction processing of the acquired raw signals. S103: Local spectral feature extraction and preliminary flame judgment, including the local data processing unit calculating the intensity ratio, transient scintillation frequency and signal rise speed of each band based on the preprocessed multi-band ultraviolet signal, and comparing the extracted spectral feature parameters with the preset flame feature model. If the extracted spectral feature parameters simultaneously meet the flame discrimination threshold, the local data processing unit initially determines it as a potential flame signal, timestamps the potential flame signal data and uploads it to the central processing and early warning unit through the wireless communication module. S2: Centralized collaborative processing, spatial positioning, and integrated early warning, including: S201: Multi-node data reception and precise timestamp alignment, including the data reception and timestamp alignment module of the central processing and early warning unit continuously receiving potential flame signal data packets from multiple different ultraviolet detection nodes, performing data verification, and precisely aligning the received data from multiple different ultraviolet detection nodes according to the high-precision timestamp embedded in the data packet; S202: Three-dimensional spatial positioning of the flame source, including the data fusion and spatial positioning module calculating the time difference of arrival (TDOA) of signals from different ultraviolet detection nodes that have been precisely aligned, and combining the known precise three-dimensional spatial coordinates of all the ultraviolet detection nodes and the calculated TDOA value, and solving the real-time position coordinates of the flame source in three-dimensional space through the TDOA algorithm based on nonlinear optimization algorithm; S203: Fusion identification, dynamic trajectory tracking, and risk assessment, including the flame identification and early warning module combining multispectral feature parameters uploaded by each node and the three-dimensional spatial location information of the flame calculated by the data fusion and spatial positioning module, using an intelligent identification algorithm for final flame identification confirmation; the dynamic trajectory tracking module continuously receives and processes flame source location data from the data fusion and spatial positioning module, using a dynamic filtering algorithm to estimate the current motion state of the flame in real time, and predict its future spread direction and speed; and S204: Graded early warning triggering and external linkage, including the flame recognition and early warning module triggering graded early warning based on flame recognition results, three-dimensional spatial location, dynamic spread trend prediction and risk assessment results, and sending the early warning information to external systems through a standard industrial communication interface.

[0013] Preferably, in step S101, the strategic deployment method ensures that the geometric layout between the distributed multispectral ultraviolet detection nodes is non-collinear and preferably non-coplanar, so as to guarantee the solution accuracy and stability of the TDOA positioning algorithm; the deployment also ensures that each node has a stable wireless communication link; the precise three-dimensional geographic coordinates of each distributed multispectral ultraviolet detection node are obtained and stored in advance through high-precision measurement.

[0014] Preferably, in step S102, the three narrowband ultraviolet sensors in each narrowband ultraviolet sensor array respectively collect radiation intensity signals in a specific ultraviolet band; the signal preprocessing includes digital filtering using a bandpass filter with an adjustable center frequency, and noise reduction processing using an adaptive denoising algorithm.

[0015] Preferably, in step S103, the intensity ratio calculated by the local data processing unit includes the ratio of the intensities of at least two narrowband ultraviolet sensors; the transient scintillation frequency is obtained by extracting the dominant frequency components and their intensities through frequency domain analysis of the signal; the signal rise time is obtained by calculating the slope of the signal within a specific time window; the flame feature model includes a preset intensity ratio threshold range, a transient scintillation frequency threshold range, and a signal rise time threshold.

[0016] Preferably, in step S202, the Time Difference of Arrival (TDOA) calculation adopts a signal cross-correlation analysis method to accurately determine the time difference; the three-dimensional spatial coordinates of the flame source are solved by a nonlinear optimization algorithm; step S202 also includes auxiliary positioning, that is, the data fusion and spatial positioning module can combine the Received Signal Strength Indication (RSSI) information from the wireless communication module, convert the RSSI value into an estimated distance through a pre-established wireless signal propagation attenuation model, and use the data fusion algorithm to fuse the TDOA positioning result with the RSSI auxiliary positioning result to optimize the final position estimation of the flame source.

[0017] Preferably, in step S203, the fusion secondary confirmation is completed through a pre-trained intelligent recognition model. This model takes the multispectral intensity ratio, transient flicker frequency, signal rise time, spatial positioning consistency of the flame source, and time series features of continuous positioning data from all relevant nodes as input. If the confidence level of the model output is higher than a preset threshold, it is confirmed as a real flame event. The dynamic trajectory tracking module adopts a dynamic filtering algorithm, using the position, velocity, and acceleration of the flame source as state variables, to estimate the current motion state of the flame in real time and predict its future spread direction and speed. In step S204, the graded early warning includes multiple levels. The external system linkage sends the early warning information, flame location, type, spread trend, and risk assessment results to the on-site automatic fire extinguishing system, fire dispatch center, industrial control system, or other emergency response platform through a standard industrial communication interface.

[0018] The beneficial effects of this invention are: This invention significantly improves the ability to identify non-flame ultraviolet sources (such as electric arcs, welding operations, and high-temperature furnace radiation) by integrating at least three sensors optimized for different narrowband ultraviolet bands at each detection node, combined with a local data processing unit for preliminary spectral feature extraction and judgment, thus greatly reducing the false alarm rate. For example, by comparing the intensity ratio and scintillation frequency characteristics of the flame characteristic band (200-240nm) and the background noise band (240-260nm), it can effectively distinguish between real flame radiation and interference sources that only have emission characteristics in a relatively wide ultraviolet spectral range. Furthermore, by deploying multiple distributed multispectral ultraviolet detection nodes and utilizing a high-precision time synchronization module to achieve nanosecond-level time synchronization, combined with the TDOA algorithm of the central processing and early warning unit, this invention overcomes the limitations of existing single or limited-view detection, achieving high-precision three-dimensional spatial positioning of flame sources in complex industrial scenarios, with positioning accuracy reaching sub-meter level. This positioning capability provides crucial information for quickly locating fire sources and guiding firefighting operations.

[0019] Meanwhile, this invention continuously receives and processes spatial positioning data of the flames, and uses advanced algorithms such as Kalman filters or extended Kalman filters to dynamically construct and predict the flame's spread trajectory, direction, and speed. This dynamic trajectory tracking capability enables the early warning system to assess fire risks in real time, providing forward-looking information for personnel evacuation and emergency response decisions, effectively preventing the fire from spreading out of control.

[0020] Furthermore, this invention achieves multi-node data fusion processing by constructing a central processing and early warning unit. This data fusion includes secondary verification of multispectral feature parameters and spatial positioning information from different nodes, and high-level flame identification and false alarm elimination through machine learning algorithms, greatly enhancing the system's robustness and reliability. The system also features tiered early warning and linkage with external fire protection and control systems, enabling rapid and accurate emergency response. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system architecture of the intelligent ultraviolet flame detection and early warning system of the present invention.

[0022] Figure 2 This is a schematic diagram of the structure of the distributed multispectral ultraviolet detection node of the present invention.

[0023] Figure 3 This is a schematic diagram of the structure of the ultraviolet detection module of the present invention.

[0024] Figure 4 This is a schematic diagram of the central processing and early warning unit of the present invention.

[0025] Figure 5 This is a flowchart illustrating the intelligent ultraviolet flame detection and early warning method of the present invention.

[0026] Figure 6 This is a flowchart illustrating the distributed multispectral ultraviolet signal acquisition and local preliminary identification method of the present invention.

[0027] Figure 7 This is a flowchart illustrating the central collaborative processing, spatial positioning, and fusion early warning method of the present invention.

[0028] Figure 8 This is a schematic diagram illustrating the multi-node signal arrival time difference positioning principle of the present invention. Detailed Implementation

[0029] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1 This embodiment discloses an intelligent ultraviolet flame detection and early warning system applicable to complex industrial scenarios. Its overall architecture aims to address technical challenges in complex industrial environments, such as high false alarm rates from non-flame ultraviolet sources, insufficient accuracy in three-dimensional spatial positioning of flame sources, lack of dynamic flame trajectory tracking capabilities, and limitations in distributed system collaborative monitoring. The system constructs a distributed multispectral ultraviolet detection network, supplemented by a high-precision time synchronization mechanism, advanced signal processing technology, multi-node data fusion algorithms, and accurate three-dimensional spatial positioning and trajectory prediction algorithms, thereby achieving rapid flame identification, precise location, dynamic tracking, and tiered early warning.

[0032] Reference Figure 1 The intelligent ultraviolet flame detection and early warning system of this invention mainly includes multiple distributed multispectral ultraviolet detection nodes and a central processing and early warning unit. The ultraviolet detection nodes are responsible for collecting multispectral ultraviolet radiation information in real time at the industrial site and performing preliminary signal processing and judgment; the central processing and early warning unit is responsible for receiving data from each detection node, performing fusion processing, three-dimensional spatial positioning, high-level flame identification, dynamic trajectory tracking, and graded early warning.

[0033] Furthermore, referring to Figure 2 The ultraviolet detection node is a highly integrated smart sensor that includes an ultraviolet detection module, a local data processing unit, a high-precision time synchronization module, a wireless communication module, and a power management module. All these modules are encapsulated in an industrial-grade housing, typically made of corrosion-resistant and explosion-proof materials.

[0034] Specifically, refer to Figure 3The ultraviolet detection module is the core component of the detection node, used to capture specific narrowband ultraviolet radiation signals from the environment. This module includes at least three narrowband ultraviolet sensor arrays: a first narrowband ultraviolet sensor, a second narrowband ultraviolet sensor, and a third narrowband ultraviolet sensor. Each narrowband ultraviolet sensor has an independent and non-overlapping spectral response band.

[0035] The first narrowband ultraviolet sensor has a center wavelength of 200 nm to 220 nm, and its full width at half maximum (FWHM) is strictly controlled to be no greater than 10 nm. This narrowband ultraviolet sensor with its chosen center wavelength and FWHM is selected to highlight a small absorption peak in the ultraviolet absorption spectrum of the flame. The physical basis for this selection is the analysis of the ultraviolet radiation spectral characteristics of the flame combustion process, particularly for hydrocarbon flames. Hydrocarbon radicals have strong emission lines around 205 nm, and carbon-oxygen radicals have strong emission lines around 215 nm. High-sensitivity ultraviolet radiation detection within this narrow band can maximize the extraction of the flame's characteristic absorption lines and enhance the signal-to-noise ratio.

[0036] The second narrowband ultraviolet sensor has a center wavelength of 220 nm to 240 nm and a spectral half-width of no more than 10 nm. This center wavelength and spectral half-width were chosen for the narrowband ultraviolet sensor to detect a wider flame radiation region. It complements the first narrowband ultraviolet sensor and is used to detect flames under different fuel types and combustion conditions. Furthermore, the ratio of the signal intensity of the second band to that of the first band is also an important characteristic for distinguishing flame type and interference.

[0037] The third narrowband ultraviolet sensor has a center wavelength of 240nm to 260nm and a spectral half-width of no more than 10nm. It is primarily used for collecting and analyzing environmental background noise. In this band, the ultraviolet radiation intensity of a real flame is generally weak, while many non-flame ultraviolet interference sources (electric arc discharge, welding arc light, high-temperature furnace radiation, solar ultraviolet background radiation, etc.) can still generate strong signals. By analyzing the signals in this band and comparing them with the signals collected by the first two narrowband ultraviolet sensors, it is possible to distinguish between real flames and non-flame ultraviolet interference sources, thereby reducing the false alarm rate of fire detectors.

[0038] Each narrowband ultraviolet sensor consists of Schottky photodiodes based on wide-bandgap semiconductor materials such as aluminum gallium nitride. This type of material has inherent solar blindness and stability over a wide temperature range, making it particularly suitable for ultraviolet detection. Each photodiode contains a visible light blocking filter to block unwanted visible and near-infrared radiation, such as a Balkan thin-film interferometer filter or a doped glass filter to block all visible and near-infrared radiation, allowing only specific ultraviolet radiation to pass through, thus preventing interference from sunlight, incandescent lamps, or high-temperature objects. Each ultraviolet sensor is equipped with a dedicated optical collimating lens. The material of this lens (such as fused silica or sapphire) should have high ultraviolet transmittance and be precisely calibrated to ensure that its field of view (FOV) is strictly controlled between 30 and 60 degrees. By precisely calibrating the field of view, the sensor's field of view can be focused, thereby suppressing scattered light from outside the field of view and providing good directional indication for subsequent three-dimensional spatial positioning. The ultraviolet detection module uses a low-noise transimpedance amplifier (TIA) to convert the weak current signal (nanoampere to microampere level) generated by each photodiode into a usable voltage signal and performs preliminary amplification of the current signal to ensure that the signal has a sufficient signal-to-noise ratio when it enters the subsequent digital processing link.

[0039] Next, the local data processing unit is electrically connected to the ultraviolet detection module and receives analog voltage signals from it. (Refer to...) Figure 2 The local data processing unit includes a multi-channel analog-to-digital converter (ADC), a digital signal processor (DSP), and a microcontroller (MCU).

[0040] The multi-channel analog-to-digital converter has a resolution of no less than 16 bits to properly quantize weak ultraviolet signals, and a sampling rate of no less than one million samples per second to sample the transient waveform of the flame signal, especially the high-frequency flicker component. Thus, the ADC converts the analog voltage signal of the ultraviolet detection module into a digital signal for digital signal processing.

[0041] The digital signal processor (DSP) performs real-time advanced processing of digital signals. This includes digital filtering, employing a fifth-order Butterworth low-pass filter with a cutoff frequency of 500Hz to effectively filter out common industrial harmonics (50Hz / 60Hz) and other high-frequency noise, while preserving the flicker characteristics of the flame (typically between 0.5Hz and 15Hz). Simultaneously, the DSP performs complex denoising, such as using an adaptive Kalman filter algorithm to dynamically adjust filtering parameters based on the statistical characteristics of the signal and noise, suppressing random noise and sudden interference to the maximum extent without losing signal details. Furthermore, the DSP performs local spectral feature extraction. Feature extraction includes calculating the intensity ratios between signals acquired by different narrowband ultraviolet sensors, such as the ratio of the intensity (I1) of the first narrowband ultraviolet sensor to the intensity (I2) of the second narrowband ultraviolet sensor, R12 = I1 / I2, and the ratio of the intensity (I2) of the second narrowband ultraviolet sensor to the intensity (I3) of the third narrowband ultraviolet sensor, R23 = I2 / I3. These ratios are key criteria for distinguishing real flames from non-flame interference. Simultaneously, the DSP extracts the transient flicker frequency components and their dominant frequencies within the range of 0.5Hz to 15Hz by performing Fast Fourier Transform (FFT) or wavelet transform on the signal. The flicker frequency of a flame is a unique fingerprint of its dynamic combustion characteristics, effectively distinguishing it from stable radiation sources. Furthermore, the DSP calculates the signal rise rate, i.e., the maximum slope of the signal intensity change within a specific time window (e.g., 10ms). A rapidly rising signal often indicates the sudden appearance or intensification of the flame.

[0042] The microcontroller compares the spectral feature parameters extracted by the DSP with the flame discrimination model to make a preliminary judgment of the flame signal. The flame discrimination model consists of multiple threshold logic judgments, such as a predefined intensity ratio threshold range (R12 between 0.8 and 1.5, R23 between 0.6 and 1.2), a transient flicker frequency threshold range (significant transient flicker frequencies exist between 2Hz and 10Hz, with a peak signal-to-noise ratio greater than 5dB), and a signal rise time threshold (100mV / ms). If the extracted spectral feature parameters meet one of the discrimination thresholds in the flame discrimination model, the microcontroller preliminarily judges it as a flame signal and assigns a local preliminary judgment confidence score. The output of the local data unit includes: multispectral intensity data (I1, I2, I3), spectral feature parameters (R12, R23, transient flicker frequency, signal rise time), local preliminary judgment result, and confidence score.

[0043] The high-precision time synchronization module is electrically connected to the local data processing unit. The module includes a oven-controlled crystal oscillator with a long-term frequency stability better than ±5 ppb. This oscillator provides a highly stable local reference clock. To ensure accurate timestamps across all distributed detection nodes, the high-precision time synchronization module periodically synchronizes with a global time source via a GPS timing receiver or a Network Time Protocol (NTP) client within the module. This synchronization guarantees timestamp accuracy for all detection nodes, at the nanosecond level or higher, such as less than 10 ns, which is crucial for TDOA calculations at the backend of the central processing and early warning unit. The synchronization frequency is configured by the administrator, preferably every 10 seconds to compensate for the negligible drift of the oven-controlled crystal oscillator. The high-precision time synchronization module adds a precise timestamp to each acquired UV signal data packet, with picosecond-level accuracy. This timestamp is encapsulated within the data packet and is only used when the central processing and early warning unit performs TDOA calculations.

[0044] The wireless communication module is electrically connected to the local data processing unit. The wireless communication module receives and transmits suspected fire event signals generated by the local data processing unit based on preliminary local fire detection results. These suspected fire event signals are transmitted to the central processing and early warning unit via the wireless communication module in a low-priority and low-reliability manner. The data structure of these suspected fire event signals should be able to carry all necessary information, including: a unique identifier for the detection node (Node ID), a timestamp provided by the high-precision time synchronization module, intensity data from three narrowband ultraviolet sensors, spectral feature parameters extracted by the DSP, a preliminary local confidence score, and a Cyclic Redundancy Check (CRC) code for data integrity verification.

[0045] The power management module provides DC power to all devices at the ultraviolet (UV) detection node. To prevent interruption of the UV detection node during a main power outage, the power management module can have a backup battery. The power management module includes an intelligent charging circuit that can switch power supply modes according to the main power status. During main power restoration, it charges the backup battery to ensure uninterrupted and normal monitoring of the UV detection node.

[0046] Reference Figure 4 The central processing and early warning unit is typically located in the control room or monitoring center. It receives all data from all distributed detection nodes and performs advanced processing such as flame identification, fire location, fire tracking, and fire early warning. The central processing and early warning unit includes a data receiving and timestamp alignment module, a data fusion and spatial positioning module, a flame identification and early warning module, a dynamic trajectory tracking module, a data storage and management module, and a human-machine interface.

[0047] The data reception and timestamp alignment module is the front-end component of the central processing and early warning unit, comprising a multi-channel wireless receiver array and a data buffer and alignment processor. The multi-channel wireless receiver array consists of multiple independent industrial-grade wireless receivers used to simultaneously receive wireless data streams from multiple ultraviolet detection nodes. The data buffer and alignment processor is equipped with a high-bandwidth network interface and high-speed cache. After connecting to the network, it has a high data reception rate and performs verification on the received data packets, such as using a CRC checksum. The data buffer and alignment processor times-aligns the received data from at least three ultraviolet detection nodes, with timestamps within a specific time window (e.g., 200 nanoseconds), according to the high-precision timestamps embedded in the data packets. This time window is crucial; if the window is too small, signals may not originate from the same flame event; if the window is too large, it will include small random delays in wireless transmission. Alignment is achieved by comparing the actual arrival time of the signal from each node to the central processing and early warning unit, taking into account the timestamp of each node, to compensate for small delays in network transmission and ensure the accuracy of subsequent TDOA calculations.

[0048] The data fusion and spatial positioning module is connected to the data receiving and timestamp alignment module. The data receiving and timestamp alignment module provides time-aligned measurement data to the data fusion and spatial positioning module. Using this aligned time measurement data and pre-stored precise 3D geographic location information of all detection nodes, the data fusion and spatial positioning module calculates the 3D spatial coordinates of the flame source using the Time Difference of Arrival (TDOA) algorithm. The TDOA algorithm employs either Chan's algorithm based on nonlinear least squares optimization or Taylor series expansion. This method first calculates the time difference of arrival of the signal between each pair of detection nodes relative to a reference node (usually the node where the signal arrives earliest). This time difference is linearly related to the distance difference between the flame source and the two nodes. By simultaneously solving at least three time difference equations and the precise 3D coordinates of each detection node, the real-time position coordinates (X, Y, Z) of the flame source in 3D space are iteratively calculated. The convergence speed and accuracy of the algorithm can be optimized with appropriate initial values ​​and iteration step sizes. The positioning accuracy can be better than sub-meter level, typically better than 0.5 meters. To improve positioning accuracy or provide auxiliary positioning results in certain situations, the data fusion and spatial positioning module can also receive Signal Strength Indication (RSSI). Using a pre-established wireless signal propagation attenuation model (e.g., a logarithmic distance path loss model), the RSSI value is converted into an estimated distance. Then, a weighted least squares method is used to fuse the TDOA positioning result and the RSSI-based auxiliary positioning result to optimize the final location estimation of the flame source, especially in cases of non-line-of-sight propagation or when signals from certain nodes are blocked. The output of the data fusion and spatial positioning module is the real-time three-dimensional spatial coordinates of the flame source and a positioning accuracy estimate.

[0049] The flame recognition and early warning module is electrically connected to the data fusion and spatial positioning module and the ultraviolet detection node. The ultraviolet detection node transmits the local spectral information it has collected; the data fusion and spatial positioning module transmits the calculated three-dimensional spatial coordinate parameters of the flame source. The flame recognition and early warning module uses machine learning classifiers such as Support Vector Machine (SVM) or Multilayer Perceptron (MLP) neural networks for higher-level flame recognition, type matching, and false alarm elimination. The feature vector input to the machine learning classifier includes: the relative multispectral intensity ratios (R12, R23) from all detection nodes participating in the flame event recognition, the transient flicker frequency of the flame source, the signal rise time, and the spatial positioning consistency of the flame source (i.e., whether multiple nodes simultaneously detect events located in the same spatial location, and the convergence of the calculated results for these locations). The machine learning classifier is trained offline using a large dataset of real flames (e.g., gasoline flames, natural gas flames, hydrogen flames, and magnesium flames) and common non-flame ultraviolet sources (e.g., electric arc discharge, welding arc light, high-temperature furnace radiation, lightning, X-ray fluorescence), to build a decision model to distinguish between them. For example, electric arcs typically exhibit broad-spectrum ultraviolet emission without a specific scintillation frequency, while real hydrocarbon flames usually possess specific narrow-band ultraviolet characteristic lines, such as those at 205 nm and 215 nm, and exhibit periodic scintillation (2-10 Hz). By fusing the aforementioned multispectral and multi-parameter methods, along with machine learning, the robustness and accuracy of the system's flame identification can be significantly improved, while effectively suppressing false alarms. Once the classifier confirms a flame event—for example, if the neural network output has a confidence level higher than a certain threshold (e.g., 0.9)—the corresponding warning level is triggered according to a predefined risk assessment strategy.

[0050] The dynamic trajectory tracking module is connected to the data fusion and spatial positioning module. The data fusion and spatial positioning module continuously sends the real-time three-dimensional spatial coordinates of the flame source to the dynamic trajectory tracking module. The dynamic trajectory tracking module uses a Kalman filter or an extended Kalman filter to estimate and predict the motion of the flame source. The filter's state vector consists of the flame source's position (x, y, z), velocity (vx, vy, vz), and acceleration (ax, ay, az) in three-dimensional space. The dynamic trajectory tracking module continuously updates the estimated position of the flame with new measured position information (provided by the positioning module) and uses flame motion models (uniform velocity model for initial spread, uniform acceleration model for accelerated spread) to estimate the current motion state of the flame in real time and predict the spread direction and speed over a future period. The dynamic trajectory tracking module connects a series of continuous flame location points, forming a dynamic flame spread trajectory on the human-computer interface. Based on the predicted spread trend and pre-built industrial scene topology information (such as area boundaries, locations of important equipment, escape routes, etc.), it predicts the potential fire area and spread speed, providing suggestions for firefighters' emergency response.

[0051] The data storage and management module is connected to the flame identification and early warning module and the dynamic trajectory tracking module. The data storage and management module includes an industrial-grade solid-state drive array and a database management system. This module stores all data, including but not limited to the aforementioned raw detection data, local processing results, flame identification results, early warning logs, historical location data, dynamic trajectory data, and other data. The database management system has standard functions such as data query, retrieval, and historical data playback. It can query data based on time, date, timestamp, location, event type, etc., and has backup and recovery functions to ensure the security, reliability, and integrity of the data. All data is backed up and timestamped for playback.

[0052] The human-machine interface (HMI) provides users with intuitive real-time monitoring, early warning display, historical playback, and system configuration functions through a graphical user interface (GUI). The HMI includes: a real-time 3D scene map, overlaid with the projected locations of detection nodes, the current location of the flame source, its dynamic spread trajectory, and predicted location, typically using GIS (Geographic Information System) technology; an early warning information panel displaying the current activity warning level, time, location, flame type, and recommended measures; a historical event query function, allowing users to match time, location, and warning level to replay historical fire events and reconstruct the fire's development; and system parameter configuration, configuring detection thresholds, warning levels, communication parameters, and sensor calibration parameters. The HMI exchanges information with other modules in the central processing and early warning unit via standard network protocols, supporting both local display and remote monitoring.

[0053] Example 2 This embodiment discloses an intelligent ultraviolet flame detection and early warning method applicable to complex industrial scenarios, which operates in conjunction with the aforementioned system. (Refer to...) Figure 5 The method mainly includes two stages: S1: distributed multispectral ultraviolet signal acquisition and local preliminary identification, and S2: central collaborative processing, spatial positioning and fusion early warning.

[0054] Specifically, refer to Figure 6 Phase S1 involves distributed multispectral ultraviolet signal acquisition and preliminary local identification. This phase is primarily completed independently by each ultraviolet detection node, aiming to quickly capture multispectral information and perform preliminary local assessments.

[0055] S101 is for the deployment of probe nodes and high-precision time synchronization.

[0056] S1011 is a strategic deployment of nodes: In the monitored area of ​​a complex industrial scenario, such as a large chemical storage tank area, at least three ultraviolet detection nodes with known precise 3D geographic coordinates are strategically deployed based on site characteristics, potential fire source distribution, and the TDOA positioning algorithm's requirements for node geometric arrangement. The deployment method ensures that the geometric layout between nodes is non-collinear, and preferably non-coplanar, thereby guaranteeing the solution accuracy and stability of the TDOA positioning algorithm. For example, four nodes can be deployed at the vertices of a regular tetrahedron to provide optimal 3D positioning geometric accuracy. The deployment also ensures a stable wireless communication link between nodes to guarantee data transmission reliability. The precise 3D geographic coordinates (X, Y, Z) of each node are measured using an RTK-GPS system and stored in the central processing and early warning unit, with an accuracy typically better than 5 centimeters.

[0057] S1012 is for global time calibration: The high-precision time synchronization module inside each ultraviolet detector node periodically calibrates with a global time source. The global time source can be a time receiver that receives GPS satellite signals, providing a PPS signal with nanosecond-level accuracy, or time information obtained from a high-precision NTP server via the NTP protocol. Calibration ensures that the timestamp accuracy of all detector nodes reaches the nanosecond level, providing an accurate time reference for subsequent Time Difference of Arrival (TDOA) calculations. The calibration frequency is configurable, preferably every 10 seconds, to compensate for minor drift of the local OCXO clock.

[0058] S102 is a real-time acquisition device for multispectral ultraviolet radiation.

[0059] S1021 is a parallel multi-band acquisition system: the narrowband ultraviolet sensor arrays in each ultraviolet detection node acquire radiation intensity signals of different narrowband ultraviolet bands in the environment in real time and in parallel.

[0060] S1022 is for signal preprocessing: The local data processing unit performs analog-to-digital conversion on the acquired raw analog signal, converting it into a 16-bit digital signal, and then performs digital filtering and noise reduction. Digital filtering uses a bandpass filter with an adjustable center frequency to filter out specific environmental harmonic interference or background noise within a specific frequency range. For example, for 50Hz or 60Hz power frequency interference, a notch filter can be configured to avoid these frequencies. Noise reduction employs a wavelet transform-based threshold denoising algorithm, such as the Daubechies wavelet or Symlets wavelet. By thresholding the coefficients in the wavelet domain, the characteristics of the flame signal can be preserved to the maximum extent while suppressing random noise, smoothing the signal, and removing spikes.

[0061] S103 is for local spectral feature extraction and preliminary flame identification.

[0062] S1031 is for spectral feature parameter extraction: The local data processing unit calculates the intensity ratios between each band based on the preprocessed multi-band ultraviolet signal. For example, the ratio of the intensity (I1) of the first narrowband ultraviolet sensor to the intensity (I2) of the second narrowband ultraviolet sensor is R12 = I1 / I2, and the ratio of the intensity (I2) of the second narrowband ultraviolet sensor to the intensity (I3) of the third narrowband ultraviolet sensor is R23 = I2 / I3. Furthermore, by performing a short-time Fourier transform (STFT) on the signal, the dominant frequency components and their intensities in the range of 0.5Hz to 15Hz are extracted to identify the transient flicker frequencies unique to flames. Simultaneously, the signal rise time is calculated, and by performing a differential operation on the signal and finding the maximum slope within a sliding window, the rapid change characteristics of the signal are characterized.

[0063] S1032 is the local preliminary judgment: The local data processing unit compares the extracted spectral feature parameters (R12, R23, flicker frequency, rise time) with a preset flame feature model. The flame feature model includes: a preset intensity ratio threshold range (e.g., R12 between 0.8 and 1.5, R23 between 0.6 and 1.2), a transient flicker frequency threshold range (e.g., there is a significant flicker frequency between 2Hz and 10Hz, and its intensity is 3dB higher than the ambient noise baseline), and a signal rise time threshold (e.g., greater than 100mV / ms). If the extracted spectral feature parameters simultaneously meet the flame discrimination threshold, the local data processing unit preliminarily determines it as a potential flame signal and generates a preliminary confidence score.

[0064] S1033 timestamps and uploads data: The preliminary assessment of potential flame signal data (raw intensity data from three narrowband ultraviolet sensors, extracted spectral characteristic parameters, and precise timestamps added by the high-precision time synchronization module) is encapsulated into a data packet and uploaded to the central processing and early warning unit via the wireless communication module. The data packet transmission time should be controlled within 50 milliseconds to ensure timeliness.

[0065] Next, refer to Figure 7 Phase S2 involves central collaborative processing, spatial positioning, and fusion-based early warning. In this phase, the central processing and early warning unit receives and fuses data from multiple detection nodes to perform precise spatial positioning, secondary identification, and early warning.

[0066] S201 enables multi-node data reception and precise timestamp alignment.

[0067] S2011 is for data reception and verification: The data reception and timestamp alignment module of the central processing and early warning unit continuously receives potential flame signal data packets from different ultraviolet detection nodes and performs data integrity verification, such as by checking the cyclic redundancy check (CRC) code to ensure the accuracy of the received data.

[0068] S2012 is for timestamp alignment: The data receiving and timestamp alignment module precisely aligns received data from at least three different nodes with highly similar timestamps, based on the high-precision timestamps embedded in the data packets. The alignment process compensates for minute random delays in wireless transmission by setting a very small time window (e.g., 200 nanoseconds) and treating all node signals arriving within this window as originating from the same flame event. The accuracy of this time window directly affects the accuracy of subsequent TDOA positioning.

[0069] S202 is the three-dimensional spatial positioning of the flame source.

[0070] S2021 is for calculating the Time Difference of Arrival (TDOA): (Refer to...) Figure 8 The data fusion and spatial positioning module calculates the time difference between signals arriving at the central processing and early warning unit from different nodes, given that the signals have been precisely aligned. Since each detection node is synchronized via a high-precision time synchronization module, the time difference directly reflects the difference in distance from the flame source to each node. The calculation employs the cross-correlation function method, precisely determining the time difference by calculating the peak cross-correlation values ​​of the signals received by different nodes. This method exhibits good robustness against noise.

[0071] S2022 is for 3D spatial coordinate calculation: The data fusion and spatial positioning module combines the precise 3D spatial coordinates of all known detection nodes with the calculated TDOA value, and uses the Chan's algorithm based on a nonlinear optimization algorithm to calculate the real-time position coordinates (X, Y, Z) of the flame source in 3D space. The algorithm constructs a cost function based on the residual between the measured TDOA value and the theoretical TDOA value calculated based on the estimated position, and uses iterative least squares method for optimization, ultimately converging to the optimal 3D position of the flame source.

[0072] S2023 is for assisted positioning: If the wireless communication module simultaneously transmits Received Signal Strength Indication (RSSI) information, the data fusion and spatial positioning module can use the RSSI information as an auxiliary parameter. Based on a pre-established wireless signal propagation model (e.g., a logarithmic distance path loss model obtained through environmental calibration), the RSSI values ​​are converted into estimated distances from the flame source to each node. The estimated distances are then fused with the TDOA positioning results using a weighted least squares fusion algorithm, where the TDOA results are typically given higher weights to further optimize positioning accuracy or provide a coarse location estimate when insufficient signal strength at certain nodes leads to inaccurate TDOA positioning.

[0073] S203 integrates identification, dynamic trajectory tracking, and risk assessment.

[0074] S2031 is a secondary confirmation process: The flame recognition and early warning module combines the multispectral feature parameters (R12, R23, flicker frequency, rise time) uploaded by each node with the three-dimensional spatial location information of the flame calculated by the data fusion and spatial positioning module to perform final flame recognition and confirmation. The confirmation process is completed through a trained multilayer perceptron neural network. This network takes feature parameters from all relevant nodes, the consistency of positioning (i.e., whether the flame positions calculated by different nodes converge to the same small area), and the time-series features of continuous positioning data as input. If the confidence level of the neural network output is higher than a preset threshold (e.g., 0.9), it is confirmed as a real flame event, thereby eliminating possible misjudgments from a single node and improving the robustness of the recognition.

[0075] S2032 is for dynamic trajectory construction and prediction: The dynamic trajectory tracking module continuously receives and processes new flame source location data from the data fusion and spatial positioning module, updating the spatial position of the flame in real time. The dynamic trajectory tracking module uses an extended Kalman filter, with the flame source's position, velocity, and acceleration as state variables. Through prediction and update steps, it estimates the current motion state of the flame in real time and predicts its spread direction and velocity over the next 5 to 30 seconds. Continuous flame location points are connected to form the dynamic spread trajectory of the flame. The trajectory information is updated in real time and displayed on the human-machine interface.

[0076] S2033 is for initial flame type identification: When multispectral feature information is sufficiently rich, the flame identification and early warning module can attempt to use finer-grained spectral fingerprints to make an initial judgment on the fire type. For example, based on the relative intensity ratios of different spectral bands and the flicker frequency pattern, it can determine whether it is a combustible gas flame (e.g., hydrogen flames have strong emission in the UV-C band, while hydrocarbon flames have strong emission in the UV-B band) or a solid combustion flame. The initial judgment results are used for subsequent risk assessment and fire extinguishing selection.

[0077] S204 is a graded early warning trigger and external linkage.

[0078] S2041 is a graded warning trigger: The flame recognition and warning module triggers a graded warning based on the initial flame recognition results, precise three-dimensional spatial location, dynamic spread trend prediction, and risk assessment results.

[0079] Warning levels include: Level 1 Warning (Attention Level): When a fire is initially identified as a flame by the system, but the confidence level is below the confirmation threshold, or only a few nodes detect it, the system prompts monitoring personnel to pay attention and marks the area of ​​concern with a flashing yellow icon on the human-machine interface, and automatically increases the detection frequency and analysis granularity.

[0080] Level 2 Warning (Local Confirmation): When the system confirms a fire as a real flame, but the flame area is small or in its early stages (e.g., less than 1 meter in diameter), and the spread is under control, the system triggers a local audible and visual alarm, displays the precise location and preliminary type of the flame in red on the human-machine interface, and issues a medium-level audible alarm.

[0081] Level 3 Warning (Emergency Response Level): When the system confirms a fire as a spreading flame, or when the size, scale, or spread speed of the flame exceeds a preset safety threshold (e.g., diameter greater than 2 meters, speed greater than 0.5 m / s), posing a threat to personnel and equipment, the system triggers a full-area emergency audible and visual alarm and sends an emergency signal to an external system via a wireless communication module or wired interface. Simultaneously, the system displays the flame's spread trajectory and predicted area in red animation on the human-machine interface and issues the highest-level emergency audible alarm.

[0082] S2042 enables external system linkage: The central processing and early warning unit transmits Level 3 early warning information, precise flame location, type, spread trend, and risk assessment results to on-site automatic fire suppression systems (e.g., activating local or area sprinkler systems or gas extinguishing systems), fire dispatch centers, industrial control systems, or building automation systems via standard industrial communication interfaces (e.g., Modbus TCP, OPC UA, or MQTT protocols), achieving rapid and precise emergency linkage response. Linkage triggering can be completed in a very short time, preferably within 1 second, thereby minimizing fire damage.

[0083] Example 3 In this embodiment, the system of the present invention is applied to a large petrochemical tank area, which is a rectangular area 200m long and 150m wide, containing multiple tanks with a diameter of 50m, as well as supporting pipelines and pumping stations. Four distributed multispectral ultraviolet detection nodes are arranged at four key locations in the southeast, southwest, northwest, and northeast of the tank area, named Node A, Node B, Node C, and Node D, respectively. These four nodes are roughly arranged in a rectangle, and their absolute three-dimensional coordinates are as follows (with the southwest corner as the origin): Node A: (0m, 50m, 5m) Node B: (150m, 50m, 5m) Node C: (150m, 100m, 5m) Node D: (0m, 100m, 5m) All nodes are 5 meters high to ensure a good field of view and communication range. Each node is equipped with the aforementioned three-band ultraviolet sensors, with center wavelengths of 210nm (UV1), 230nm (UV2), and 250nm (UV3), and a half-width at half-maximum (HWHM) of 8nm. The sampling rate is 2MSPS, and the time synchronization accuracy is better than 5ns. The central processing and early warning unit is located in a control room 200m away.

[0084] In a simulation experiment, a leak occurred in one of the tanks in the tank area, which caused a small flame of propane gas. The source of the fire was located slightly north of the center of the tank area.

[0085] 1. Signal Acquisition and Local Preliminary Identification: At a certain moment, the flame is ignited.

[0086] Fifty milliseconds later, Nodes A, B, C, and D simultaneously detected a significant increase in the ultraviolet signal.

[0087] The Node A local data processing unit detected the following values: UV1 = 800mV, UV2 = 650mV, UV3 = 200mV, R12 = 1.23, R23 = 3.25. The dominant peak of the transient flicker frequency is 8Hz, indicating a strong signal. The rise time is 150mV / ms. All of these parameters are determined to exceed the corresponding flame detection model threshold. The local preliminary assessment is a potential flame with a confidence level of 0.92.

[0088] Nodes B, C, and D also perform similar local assessments simultaneously, with confidence levels of 0.88, 0.90, and 0.91, respectively. After 80 milliseconds, all nodes have completed packet encapsulation, adding nanosecond-level timestamps, and sending the packets to the central processing and early warning unit via the wireless network.

[0089] 2. Centralized Coordination and Spatial Positioning: The central processing and early warning unit received data packets from all nodes between 110 and 120 milliseconds. The data reception and timestamp alignment module detected that the timestamps of the four nodes were within a time window of approximately 250 ns, and determined that they were the same event.

[0090] The data fusion and spatial positioning module performs TDOA calculations on these timestamps. Taking Node A as an example, the calculated time differences are Δt_BA = tB - tA = 222 ns, Δt_CA = tC - tA = 87 ns, and Δt_DA = tD - tA = 245 ns. Combining the precise three-dimensional coordinates of the four nodes, the Chan's algorithm is used to calculate the real-time three-dimensional coordinates of the flame source as (75.2 m, 78.5 m, 2.1 m). The positioning accuracy is estimated to be 0.38 m.

[0091] 3. Integration of recognition, dynamic trajectory tracking, and early warning: The flame recognition and early warning module obtains the multispectral feature parameters and flame location of each node. The trained multilayer perceptron neural network calculates a flame confidence score of 0.98, determining that this is a genuine propane gas flame event, excluding interference from electric arcs and other factors.

[0092] The dynamic trajectory tracking module began continuously receiving updates on the flame's position. Within 10 seconds, the flame's position was updated 200 times. The extended Kalman filter estimated that the flame was spreading northeast at a speed of approximately 0.2 m / s. It was predicted that in the next 15 seconds, the fire might spread to the nearby pipe area.

[0093] The system immediately triggered a Level 3 warning: all audible and visual alarms activated throughout the area, and the human-machine interface in the control room displayed the flame location, type (propane), current flame spread rate, and predicted spread path. It also sent a signal via Modbus TCP protocol to the automatic foam extinguishing system in the area to activate fire suppression, and simultaneously sent a fire alarm message (fire flame location and flame spread trend) to the fire dispatch center, enabling personnel at the fire dispatch center to promptly grasp the fire situation. The time from the start of the test flame ignition to the end of the system's Level 3 warning linkage was approximately 1.2 seconds.

[0094] This invention utilizes distributed multispectral ultraviolet detection, high-precision time synchronization, TDOA spatial positioning, multi-feature fusion, and machine recognition to achieve precise flame detection, location, dynamic tracking, and early warning in complex industrial environments. This significantly reduces the false alarm rate and overcomes the shortcomings of existing technologies using single ultraviolet detectors, which are prone to false alarms or missed alarms. It also enables more intelligent flame monitoring and early warning. These advantages translate into faster emergency warning and response in industrial settings, and more precise fire suppression, thereby protecting human life and property.

[0095] Example 4 The ultraviolet detection module further integrates an infrared (IR) sensor module, which is used to collect infrared radiation intensity data of the target area in real time. The local data processing unit further preprocesses and extracts features from the infrared radiation intensity data, and performs multimodal fusion of ultraviolet and infrared feature data locally to enhance the accuracy of the preliminary flame signal judgment. In step S103, the local data processing unit further fuses the multimodal features extracted from the ultraviolet and infrared radiation data and applies a multimodal fusion algorithm to perform the preliminary flame signal judgment.

[0096] Step S1 also includes periodically performing node self-calibration and health monitoring. Self-calibration includes activating the node's embedded reference light source to detect and compensate for sensitivity drift of the ultraviolet detection module. Health monitoring includes real-time evaluation of the performance of each component and generating a node health report when an anomaly occurs. The health report is uploaded to the central processing and early warning unit along with the potential flame signal data.

[0097] Step S2 also includes: periodically evaluating the network connectivity and health status of the distributed multispectral ultraviolet detection nodes; and dynamically adjusting the positioning weights and parameters of the TDOA algorithm when some nodes are detected to be faulty, offline, or have degraded signal quality, so as to maximize the accuracy and reliability of the calculation of the three-dimensional spatial coordinates of the flame source.

[0098] If a node fails to send data packets or its data packet loss rate exceeds a preset threshold (e.g., 20%) for multiple consecutive evaluation periods, the node is considered offline. If the node health report indicates critical issues such as sensor malfunction or power failure, it is considered faulty. If the RSSI of the node's transmitted data remains below the preset minimum signal quality threshold, or if the node health report indicates severe drift in detection sensitivity that cannot be effectively compensated for, it is considered to be in a state of degraded signal quality.

[0099] During normal operation, all nodes participating in the localization process are assigned preset initial weights in the TDOA nonlinear optimization algorithm. When a node is determined to have degraded signal quality, the network topology adaptive module dynamically reduces its weight in the TDOA localization algorithm. The reduction can be adjusted linearly or nonlinearly depending on the severity of the signal quality degradation. For example, if the RSSI drops by 10dB, the weight is reduced by 50%. When a node is determined to be "offline" or "faulty," its weight in the TDOA localization algorithm is set to zero, meaning it is temporarily excluded from the localization calculation.

[0100] In the TDOA algorithm, a reference node is typically selected to calculate the relative time difference. When the original reference node becomes abnormal, the system automatically selects the next healthy node with the best signal quality as the new reference node. If the number of available nodes decreases from at least four (providing redundancy for optimal TDOA positioning) to only three (the minimum requirement for TDOA positioning), the system will adjust the parameters of the positioning algorithm, such as using a more robust but computationally more computationally demanding least squares or iterative algorithm, and may issue a warning of decreased positioning accuracy. If fewer than three nodes are available, TDOA positioning will not be possible, and the system will attempt to switch to other auxiliary positioning modes (such as RSSI positioning if available) or only issue a local warning.

[0101] Through the aforementioned dynamic weight and parameter adjustments, the system can minimize positioning errors or complete failures caused by anomalies in some nodes. Even in degraded operation mode, it can provide accurate three-dimensional spatial coordinates of the flame as much as possible, and simultaneously display the system's current health status and positioning accuracy level on the human-machine interface (e.g., a human-machine interface), ensuring that operators have a clear understanding of the system's performance.

[0102] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent ultraviolet flame detection and early warning system suitable for complex industrial scenarios, characterized in that, include: Multiple distributed multispectral ultraviolet detection nodes, each of which integrates an ultraviolet radiation signal ultraviolet detection module, the ultraviolet detection module being equipped with at least three narrowband ultraviolet sensor arrays, the spectral response bands of each narrowband ultraviolet sensor in the narrowband ultraviolet sensor array being independent of each other, used to capture specific narrowband ultraviolet radiation signals from the environment; A central processing and early warning unit is used to receive and process data from the plurality of ultraviolet detection nodes, the central processing and early warning unit comprising: The data receiving and timestamp alignment module is used to accurately align data from multiple ultraviolet detection nodes; The data fusion and spatial positioning module is used to calculate the three-dimensional spatial coordinates of the flame source based on the aligned data and the precise three-dimensional geographical location information of the ultraviolet detection nodes, using the signal time difference of arrival algorithm. The flame recognition and early warning module is used to receive local spectral feature parameters and three-dimensional spatial coordinates of the flame source from the ultraviolet detection node, perform flame recognition, type judgment and false alarm elimination, and trigger an early warning. The dynamic trajectory tracking module is used to continuously receive the three-dimensional spatial coordinates of the flame source, estimate and predict the flame movement state, and construct a dynamic spread trajectory.

2. The intelligent ultraviolet flame detection and early warning system according to claim 1, characterized in that, The narrowband ultraviolet sensor uses a photodiode based on a wide bandgap semiconductor material and integrates a filter and an optical collimating lens. The narrowband ultraviolet sensor includes a first narrowband ultraviolet sensor for detecting characteristic ultraviolet radiation of a flame, a second narrowband ultraviolet sensor for detecting wide-range radiation of a flame, and a third narrowband ultraviolet sensor for collecting and analyzing ambient background noise to distinguish non-flame interference sources.

3. The intelligent ultraviolet flame detection and early warning system according to claim 1, characterized in that, The local data processing unit includes: An analog-to-digital converter is used to convert the analog signal output by the ultraviolet detection module into a digital signal; A digital signal processor is configured to perform real-time digital filtering, denoising, and local spectral feature extraction on the digital signal. The local spectral feature extraction includes calculating the intensity ratio, transient scintillation frequency, and signal rise time among the signals acquired by the different narrowband ultraviolet sensors. The microcontroller is used to compare the extracted spectral feature parameters with a preset flame discrimination model to perform preliminary flame signal judgment. The flame discrimination model includes a preset intensity ratio threshold range, a transient flicker frequency threshold range, and a signal rise time threshold.

4. A method for intelligent ultraviolet flame detection and early warning applicable to complex industrial scenarios, applicable to the intelligent ultraviolet flame detection and early warning system as described in any one of claims 1-3, characterized in that, Includes the following steps: S1: Distributed multispectral ultraviolet signal acquisition and local preliminary identification Each ultraviolet detection node acquires radiation signals of different narrowband ultraviolet bands in real time and in parallel through a narrowband ultraviolet sensor array, processes the signals locally, extracts spectral features, uses a flame feature model to preliminarily determine potential flame signals, and adds a precise timestamp to the potential flame signal data and uploads it. S2: Centralized collaborative processing, spatial positioning, and integrated early warning. It receives potential flame signal data packets from multiple ultraviolet detection nodes and aligns them precisely according to timestamps; based on the aligned signals and the precise three-dimensional coordinates of the nodes, it calculates the three-dimensional spatial coordinates of the flame source using a signal arrival time difference algorithm; combining local spectral features and the three-dimensional spatial location of the flame source, it uses an intelligent recognition algorithm to identify the flame, determine its type, and eliminate false alarms; simultaneously, it uses a dynamic filtering algorithm to estimate and predict the flame's motion state to construct a dynamic spread trajectory; and based on the identification, location, tracking, and prediction results, it triggers graded early warnings and sends warning information.

5. The intelligent ultraviolet flame detection and early warning method according to claim 4, characterized in that, Step S1 specifically includes: S101: Deployment of detection nodes and high-precision time synchronization: Deploy multiple ultraviolet detection nodes with known three-dimensional geographic coordinates, and periodically calibrate each of the ultraviolet detection nodes with the global time source; S102: Real-time acquisition and local signal processing of multispectral ultraviolet radiation. The narrowband ultraviolet sensor arrays in each ultraviolet detection node acquire radiation intensity signals of different narrowband ultraviolet bands in the environment in parallel, and perform analog-to-digital conversion, digital filtering and noise reduction on the acquired raw signals. S103: Local spectral feature extraction and preliminary flame judgment. Based on the preprocessed multi-band ultraviolet signal, calculate the intensity ratio, transient scintillation frequency and signal rise speed between each band, and compare the extracted spectral feature parameters with the preset flame feature model. If the extracted spectral feature parameters simultaneously meet the flame discrimination threshold, it is initially determined to be a potential flame signal, and the potential flame signal data is timestamped and uploaded.

6. The intelligent ultraviolet flame detection and early warning method according to claim 5, characterized in that, In step S101, the geometric layout between the ultraviolet detection nodes is non-collinear and preferably non-coplanar to ensure the accuracy and stability of the signal arrival time difference algorithm.

7. The intelligent ultraviolet flame detection and early warning method according to claim 5, characterized in that, In step S103, the calculated intensity ratio includes the ratio of the intensities of at least two narrowband ultraviolet sensors; the transient scintillation frequency is obtained by extracting the dominant frequency components and their intensities through frequency domain analysis of the signal. The signal rise rate is obtained by calculating the slope of the signal within a specific time window; the flame feature model includes a preset intensity ratio threshold range, a transient flicker frequency threshold range, and a signal rise rate threshold.

8. The intelligent ultraviolet flame detection and early warning method according to claim 4, characterized in that, Step S2 specifically includes: S201: Multi-node data reception and precise timestamp alignment. Receives potential flame signal data packets from multiple different ultraviolet detection nodes and precisely aligns the received data according to the high-precision timestamp embedded in the data packet. S202: Three-dimensional spatial positioning of the flame source. For the aligned signal, calculate its signal arrival time difference, and combine the known three-dimensional spatial coordinates of all ultraviolet detection nodes with the calculated signal arrival time difference to solve the real-time position coordinates of the flame source in three-dimensional space. S203: Fusion identification, dynamic trajectory tracking and risk assessment. Combining the multispectral feature parameters uploaded by each ultraviolet detection node and the three-dimensional spatial location information of the flame, flame fusion identification is performed, flame source location data is processed, the current motion state of the flame is estimated in real time, and its future spread direction and speed are predicted. S204: Tiered early warning triggering and external linkage. Based on flame identification results, three-dimensional spatial location, dynamic spread trend prediction and risk assessment results, tiered early warning is triggered and the warning information is sent to external systems.

9. The intelligent ultraviolet flame detection and early warning method according to claim 8, characterized in that, Step S202 further includes assisted positioning, which combines the received signal strength indication information, converts the received signal strength indication value into an estimated distance through a pre-established wireless signal propagation attenuation model, and fuses the signal arrival time difference positioning result with the received signal strength indication assisted positioning result to optimize the final location estimation of the flame source.

10. The intelligent ultraviolet flame detection and early warning method according to claim 8, characterized in that, In step S203, the fusion recognition is completed through a pre-trained intelligent recognition model. This model takes the multispectral intensity ratio, transient flicker frequency, signal rise speed, spatial positioning consistency of the flame source, and time series features of continuous positioning data from all relevant nodes as input. If the confidence level of the model output is higher than a preset threshold, it is confirmed as a real flame event. The dynamic trajectory tracking module adopts a dynamic filtering algorithm, using the position, velocity, and acceleration of the flame source as state variables, to estimate the current motion state of the flame in real time and predict its future spread direction and speed.

Images