Foam extinguishing agent multi-component proportion mixing method and foam mixing generation device

Abstract

The application relates to the technical field of foam fire extinguishing mixing, and discloses a foam fire extinguishing agent multi-component proportion mixing method and a foam mixing generating device. The foam fire extinguishing agent multi-component proportion mixing method comprises the following steps: acquiring fire scene information in real time through multi-modal sensing, identifying a fire type and an environmental condition by using an edge AI model based on the fire scene information, and generating a structured fire condition label. Through edge AI local reasoning and a straight-through mixing path design, the first dose of foam can be quickly sprayed out after fire confirmation, the golden fire extinguishing time is effectively seized, the method is especially suitable for fast-spreading fire conditions such as lithium batteries, a three-stage composite formula is dynamically called according to the fire condition, high-cost functional components are added only when needed, fire extinguishing efficiency is improved, multi-modal sensing and a closed-loop feedback mechanism are fused, even if there is deviation in initial identification, the components and mixing parameters can be corrected in real time through fire scene feedback, and fire extinguishing failure caused by misjudgment is avoided.

Description

Technical Field

[0001] This invention relates to the field of foam fire extinguishing mixing technology, specifically to a method for mixing multiple components of foam fire extinguishing agents in a specific ratio and a foam mixing and generating device. Background Technology

[0002] Foam extinguishing agents, as an important fire protection product, are widely used in fire fighting scenarios involving flammable liquids such as petroleum, chemical, airport, and warehouse. Their extinguishing mechanism mainly involves forming a continuous and stable foam covering layer on the surface of the burning material, isolating oxygen, inhibiting the volatilization of flammable vapors, and extinguishing the fire through cooling.

[0003] Traditional foam fire extinguishing systems generally use a single premixed foam concentrate with a fixed formula, making it difficult to meet the differentiated performance requirements of foam in different fire scenarios. For example, water-soluble polar solvents require the addition of high-molecular-weight anti-alcohol polymers to form a protective film, lithium battery fires require cooling, insulation, and covering functions, while strong winds or high-altitude environments require foams with stronger stability and anti-evaporation capabilities. Therefore, existing technologies suffer from single components, inability to combine multiple functional additives as needed, risk of premix failure, and the tendency for multiple components to coexist for a long time to cause stratification, precipitation, or chemical deactivation. They also suffer from slow response, difficulty in switching, and the need to shut down the system and change tanks when changing agents, making it impossible to adapt to changes in the fire situation in real time. Summary of the Invention

[0004] This invention provides a method for mixing multiple components of a foam extinguishing agent in a specific ratio and a foam mixing and generating device, which has the beneficial effect of dynamically adapting to the fire situation and adjusting the mixing effect of the foam extinguishing agent.

[0005] This invention provides the following technical solution: a method for mixing multiple components of a foam fire extinguishing agent, comprising the following steps: Real-time fire scene information is obtained through multimodal sensing; Based on the fire scene information, an edge AI model is used to identify the fire type and environmental conditions, and a structured fire label is generated. According to the fire label, the corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database. The three-stage fire extinguishing formula template includes a rapid suppression stage, a precise fire extinguishing stage, and a maintenance assessment stage. Based on the main water flow rate, the target flow rate of each functional component is calculated according to the current component ratio, and multiple independent metering units are controlled to add water in a coordinated manner. Based on the current stage, the mixing mode is dynamically switched. In the rapid suppression stage, a low-mixing-efficiency direct path is used to output the initial mixture. In the precise fire extinguishing stage, a multi-level gradient mixing chamber and ultrasonic homogenization are used to perform high-precision mixing. During the firefighting process, fire scene feedback data is continuously collected, and the component ratio or mixing parameters for the next cycle are dynamically adjusted based on the feedback data. Once the fire is under control, it automatically transitions to the maintenance assessment phase, implementing pulsed micro-spraying and continuously monitoring the risk of reignition until the fire is confirmed to be extinguished.

[0006] As an optional scheme of the multi-component ratio mixing method of the foam fire extinguishing agent described in this invention, wherein: in the three-stage fire extinguishing formula template, different mixing efficiency factors η are defined for each stage; Wherein, the mixing efficiency factor η of the rapid suppression phase is smaller than the mixing efficiency factor η of the precision extinguishing phase; The maintenance assessment phase is used to add trace components that prevent reignition or stabilize foam, and the mixing efficiency is dynamically adjusted according to the pulse frequency. The mixing efficiency factor η is calculated as follows: in, This refers to the actual residence time of the fluid in the effective mixing region. The theoretical minimum time required to achieve 95% uniformity.

[0007] As an optional solution to the multi-component mixing method of the foam fire extinguishing agent described in this invention, the fire scene information includes flame images, thermal imaging data, gas concentrations, and environmental parameters. The method for identifying fire types and environmental conditions using the edge AI model includes: The fire scene information is synchronized and preprocessed in time to generate a multimodal input frame containing visual frames, temperature distribution maps, gas concentration vectors, and environmental parameters. Multimodal input frames are fused and inferred using an edge AI model; The output identifies the flame shape, color, dynamic features, and smoke density based on the visual frame. The output extracts hotspot locations, maximum temperature values, and temperature gradients based on the temperature distribution map. The output identifies characteristic gas combinations based on the gas concentration vector to determine the type of combustible material. Based on the aforementioned environmental parameters, wind speed, ambient temperature, humidity, and altitude information are determined. Based on the fusion inference results, a predefined fire classification standard library is matched to output the fire type and its confidence level, and a structured fire label is generated by integrating environmental parameters. The structured fire label is a machine-readable standardized data structure that includes fire type field, environmental condition field, hazard indication field and initial handling suggestion field. The structured fire tags are transmitted to the fire suppression controller as a basis for decision-making regarding the activation of multi-component foam formulations and the setting of mixing strategies.

[0008] As an optional solution to the multi-component mixing method of the foam fire extinguishing agent described in this invention, wherein: based on structured fire tags, an appropriate three-stage fire extinguishing formula template is automatically invoked, and based on the measured main water flow rate... Real-time calculation of the target dosing flow rate of each functional component This output is then sent to each metering pump / valve actuator to achieve closed-loop precise dosing. The calculation formula is as follows: in, This is the volume percentage defined in the formula; The edge AI model includes: The object detection sub-network of the EfficientDet architecture is used to detect flame and smoke areas from visible light video streams; A temporal analysis subnetwork of ResNet and spatiotemporal attention mechanism is used to analyze the flame flickering frequency and spread trend. 1D convolutional neural network or LSTM subnetwork is used to process multi-gas concentration time series data to identify characteristic combustion products; The multilayer perceptron fusion layer is used to weight and fuse visual features, thermal imaging features, gas features, and environmental parameters to output the final classification result.

[0009] As an optional solution to the multi-component mixing method of the foam fire extinguishing agent described in this invention, the structured fire label adopts JSON format, and the value of its fire type field is limited to a preset enumeration set. When any sensor data is missing or the confidence level is below the threshold, the edge AI model automatically reduces the weight of that modality and relies on the remaining valid modalities to complete the inference, ensuring that valid fire tags are still output even when some sensors fail. To prevent sudden changes in the flow command of metering pumps / valve when switching between different fire suppression stages or adjusting component formulations, the actual output flow command is adjusted via a ramp transition: in, The actual output flow rate command for component i at time t; The current traffic value before the switch. The target flow rate is required for the new phase. For ramp time, The current system time. This is the trigger moment for phase switching.

[0010] As an optional embodiment of the multi-component mixing method for foam fire extinguishing agents described in this invention, the method for dynamically switching mixing modes includes: During the rapid suppression phase, the control fluid pipeline is switched to a straight path, so that the main water flow and the base foam concentrate are initially mixed only through a single static mixer, bypassing the ultrasonic homogenizing unit and multi-stage fine mixing chambers, and outputting the initial mixture with a low mixing efficiency factor η to ensure that the first dose of foam is sprayed out quickly. During the precise fire extinguishing phase, the fluid pipeline is switched to a series path, so that the main water flow passes through a multi-stage gradient mixing chamber in sequence, and an ultrasonic transducer is activated in at least one stage of the mixing chamber for auxiliary dispersion, so as to achieve high-precision mixing of multi-functional components with a high mixing efficiency factor η. The multifunctional component includes at least two of the following: anti-alcohol polymer, foam stabilizer, insulating microspheres, or phase change microcapsules; The switching between the rapid suppression phase and the precise fire extinguishing phase is triggered by structured fire tags, and the target dosage of each component is dynamically calculated based on the real-time main water flow rate.

[0011] As an optional embodiment of the multi-component mixing method for foam fire extinguishing agent described in this invention, it further includes: During the firefighting process, fire scene feedback data is continuously collected at a set sampling period; The fire feedback data includes the flame area change rate, the highest temperature decay rate, the characteristic gas concentration trend, and the foam coverage uniformity. The current fire extinguishing effect is evaluated based on the fire feedback data. If the fire extinguishing effect does not reach the preset threshold within multiple consecutive sampling cycles, the parameters of the next control cycle are dynamically adjusted.

[0012] A mixing device for using a multi-component mixing method for foam extinguishing agents includes: A multimodal sensing device is used to collect fire scene information in real time, including flame images, thermal imaging data, gas concentrations, and environmental parameters. An edge AI controller, which is deployed locally and communicates with the multimodal sensing device, is used to perform fusion reasoning on the fire information, identify the fire type and environmental conditions, and generate structured fire tags. The corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database based on the structured fire label; A main water flow metering unit, which is used to measure the main water flow rate in the water supply pipeline in real time; Multiple independent functional component dosing units store and measure the base foam concentrate and at least two functional additives, the functional additives including any combination of anti-alcohol polymers, foam stabilizers, insulating microspheres or phase change microcapsules, and each dosing unit is controlled by the edge AI controller to do so in a coordinated manner according to the target flow rate. A reconfigurable mixing device includes a through mixing path and a multi-level gradient mixing path. The through mixing path includes a single-stage static mixer, and the multi-level gradient mixing path includes a series of multi-stage mixing cavities and an ultrasonic homogenizing unit integrated in at least one stage mixing cavity. The reconfigurable mixing device is also equipped with a flow path switching mechanism, controlled by the edge AI controller, for opening the direct mixing path during the rapid suppression phase, bypassing the ultrasonic homogenizing unit, and outputting a preliminary mixture. During the precise fire extinguishing phase, the multi-level gradient mixing path is activated, and the ultrasonic homogenizing unit is enabled to achieve high-precision mixing. The feedback sensing module is used to continuously collect fire scene feedback data during the fire extinguishing process. The fire scene feedback data includes flame area change rate, temperature decay rate, characteristic gas concentration trend and foam coverage status. A closed-loop optimization module is set inside the edge AI controller. It evaluates the fire extinguishing effect based on the fire feedback data and dynamically adjusts the component ratio, ultrasonic power, mixing efficiency factor or spraying strategy for the next control cycle when the fire extinguishing effect does not reach the threshold. The foam output interface is connected to the outlet of the reconfigurable mixing device and is used to deliver the mixed foam liquid to the foam generator.

[0013] The present invention has the following beneficial effects: 1. The multi-component mixing method and foam mixing and generating device of this foam extinguishing agent, through edge AI local inference and direct mixing path design, can quickly spray the first foam after the fire is confirmed, effectively seizing the golden time for fire extinguishing. It is especially suitable for rapidly spreading fires such as those involving lithium batteries. The three-stage composite formula is dynamically called according to the fire situation, and high-cost functional components are added only when needed to improve fire extinguishing efficiency. It integrates multi-modal perception and closed-loop feedback mechanism. Even if there is a deviation in the initial identification, the components and mixing parameters can be corrected in real time through fire scene feedback to avoid fire extinguishing failure due to misjudgment.

[0014] 2. The multi-component mixing method and foam mixing and generating device of the foam extinguishing agent achieve a dynamic balance between speed and precision through reconfigurable flow path and ultrasonic homogenization technology. It prioritizes speed and safety in the initial stage, precision and thoroughness in the middle stage, and stability and prevention of reignition in the later stage, thus comprehensively covering the needs of the entire life cycle of fire extinguishing. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the fire perception and tag generation process of the present invention.

[0016] Figure 2 This is a schematic diagram of the three-stage formula calling and mixing spraying process of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1 Please see Figures 1-2 The method for mixing the multi-component components of foam extinguishing agents includes the following steps: Real-time fire scene information is obtained through multimodal sensing; Based on fire scene information, edge AI models are used to identify fire types and environmental conditions, and generate structured fire labels. Based on the fire label, the corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database. The three-stage fire extinguishing formula template includes the rapid suppression stage, the precise fire extinguishing stage, and the maintenance assessment stage. Based on the main water flow rate, the target flow rate of each functional component is calculated according to the current component ratio, and multiple independent metering units are controlled to add water in a coordinated manner. Based on the current stage, the mixing mode is dynamically switched. In the rapid suppression stage, a low-mixing-efficiency direct path is used to output the initial mixture. In the precise fire extinguishing stage, a multi-level gradient mixing chamber and ultrasonic homogenization are used to perform high-precision mixing. During the firefighting process, fire scene feedback data is continuously collected, and the component ratio or mixing parameters for the next cycle are dynamically adjusted based on the feedback data. Once the fire is under control, it automatically transitions to the maintenance assessment phase, implementing pulsed micro-spraying and continuously monitoring the risk of reignition until the fire is confirmed to be extinguished.

[0019] Upon activation or receipt of a fire alarm signal, a multimodal sensing device deployed at the front end of the fire-fighting equipment first collects real-time fire scene information. This device includes a visible light camera, an infrared thermal imager, multiple gas sensors such as CO, VOCs, and HF, as well as environmental parameter sensors to measure temperature, humidity, wind speed, and air pressure. It can simultaneously acquire multi-dimensional data such as flame morphology, heat distribution, characteristic combustion products, and on-site environmental conditions. After time alignment and preprocessing, this raw data forms a unified multimodal input frame.

[0020] Subsequently, a lightweight AI model deployed on local edge computing devices, such as embedded controllers with NPU acceleration units, performs fusion inference on the aforementioned multimodal input frames. This model employs a multi-branch architecture: the vision branch utilizes an improved EfficientDet network to identify flame regions, colors, and dynamic features; the thermal imaging branch extracts hotspot locations, maximum temperatures, and temperature gradients; the gas analysis branch uses 1D-CNN or LSTM to identify characteristic gas combinations to determine the type of combustion product (e.g., lithium battery thermal runaway releases HF, and polar solvent combustion produces specific VOCs); and the environmental branch analyzes conditions such as wind speed, low temperature, and high altitude.

[0021] After attention-weighted fusion of features from each branch, the data is matched against a predefined fire classification standard library to output a structured fire label. This label is a machine-readable, standardized data structure that explicitly includes fields such as fire type (e.g., lithium battery thermal runaway), environmental conditions (e.g., wind speed 8.2 m / s, temperature -5°C), hazard level, and initial response recommendations. This data is transmitted in real-time to the fire suppression controller for decision-making. Based on this structured fire label, the controller automatically retrieves the matching three-stage fire suppression formula template from a pre-set formula database.

[0022] This template divides the entire fire suppression process into three logical stages: rapid suppression, precise fire suppression, and maintenance assessment. Each stage defines specific component ratios, mixing efficiency factors η, duration strategies, and spray patterns. For example, for lithium battery fires in high-wind-speed environments, the system might utilize a composite formulation containing anti-alcohol polymers, ceramic microspheres, and phase change microcapsules, prioritizing the addition of polymers during the rapid suppression stage to form a pre-coating film. After determining the current stage, the target flow rate of each functional component is calculated based on the measured flow rate of the main water flow, obtained in real-time by an electromagnetic flowmeter, according to the corresponding volume percentage for that stage. For example, if the main water flow is 100 L / min and a component ratio is 1.2%, the target flow rate is 1.2 L / min. Multiple independent metering units, such as high-precision diaphragm pumps or screw pumps, work collaboratively to add the base foam concentrate and functional additives, such as anti-alcohol polymers, foam stabilizers, insulating microspheres, and phase change microcapsules. For high-viscosity components, the system implements early pump start-up and initial flow compensation; for trace components (<0.1%), a weighing sensor is introduced for cumulative mass feedback correction to ensure dosing accuracy within ±0.5%. Simultaneously, the system dynamically switches mixing modes according to the current stage. In the rapid suppression stage, the electric three-way valve switches the flow path to a straight path, allowing the main water flow and basic components to undergo preliminary mixing only through a single static mixer, bypassing the ultrasonic homogenizing unit and subsequent multi-stage mixing chambers. This results in a low-mixing-efficiency (η≈0.4) preliminary mixture output within 3 seconds, achieving the tactical objective of suppressing the fire initially. Once the precision extinguishing stage begins, the valve switches to a series path, with the fluid flowing sequentially through three gradient mixing chambers. Simultaneously, a 28kHz ultrasonic transducer integrated in the second-stage chamber is activated for cavitation dispersion, ensuring thorough homogenization of the multi-components and achieving high mixing efficiency (η≥0.85), ensuring dense foam coverage and strong resistance to reignition. The mixing efficiency factor η is determined by both residence time and ultrasonic power, and is indirectly set through flow path configuration.

[0023] Throughout the firefighting process, fire feedback data is continuously collected at fixed sampling intervals (e.g., every 2 seconds), including the rate of change of flame area (calculated by a visual algorithm), the rate of decay of the highest temperature (from thermal imaging), the trend of characteristic gas concentrations (e.g., whether CO decreases), and the uniformity of foam coverage (assessed through image segmentation). The closed-loop optimization module constructs a comprehensive firefighting effectiveness index based on this data. If the effectiveness fails to reach a preset threshold (adaptively set according to the fire type) within 2-3 consecutive cycles, the parameters for the next control cycle are automatically adjusted. This may involve increasing the foam stabilizer ratio by 10%-30%, enhancing ultrasonic power, extending mixing time, or activating backup spray branches to expand coverage, achieving adaptive optimization through continuous adjustment.

[0024] Once the fire is determined to be under control (e.g., temperature remains below 100°C and no visible flames for more than 10 seconds), the system automatically transitions to the maintenance assessment phase. This phase employs a pulsed micro-injection mode, intermittently adding a low-concentration mixture containing anti-reignition components every 20–60 seconds, while continuously monitoring for reignition risk (e.g., temperature rise, HF concentration rebound). Injection is only terminated and a pipeline self-flushing procedure is initiated only after confirming there is no possibility of reignition (e.g., temperature remains consistently below 80°C for more than 60 seconds) to prevent residue blockage.

[0025] In summary, through edge AI local inference and direct hybrid path design, the first-dose foam can be rapidly sprayed after fire confirmation, effectively seizing the golden fire extinguishing time. It is especially suitable for rapidly spreading fires such as those involving lithium batteries. The three-stage composite formula is dynamically called according to the fire situation, and high-cost functional components are added only when needed to improve fire extinguishing efficiency. By integrating multimodal perception and closed-loop feedback mechanisms, even if there are deviations in the initial identification, the composition and mixing parameters can be corrected in real time through fire scene feedback to avoid fire extinguishing failure due to misjudgment. Through reconfigurable flow path and ultrasonic homogenization technology, a dynamic balance is achieved between speed and precision. In the initial stage, speed ensures safety; in the middle stage, precision ensures thoroughness; and in the later stage, stability prevents reignition, comprehensively covering the needs of the entire fire extinguishing life cycle.

[0026] Example 2 This embodiment is an improvement upon embodiment 1. For details, please refer to [link / reference]. Figures 1-2 In the three-stage fire extinguishing formula template, different mixing efficiency factors η are defined for each stage; Among them, the mixing efficiency factor η of the rapid suppression stage is smaller than that of the precision extinguishing stage. During the maintenance and evaluation phase, trace components that prevent reignition or stabilize foam are added, and the mixing efficiency is dynamically adjusted according to the pulse frequency. The mixing efficiency factor η is calculated as follows: in, This refers to the actual residence time of the fluid in the effective mixing region. The theoretical minimum time required to achieve 95% uniformity.

[0027] The rapid suppression phase involves delivering foam liquid with basic fire-extinguishing capabilities in the shortest possible time to suppress the spread of fire. With the direct mixing path enabled, the fluid flows only through a single-stage Kenics static mixer, bypassing the ultrasonic homogenizing unit and multiple stages of fine mixing chambers; The length of stay is =0.6-1.0s; The theoretical time is due to the simple composition, usually consisting of only basic foam stock solution and a small amount of polymer. =2.0s; The mixing efficiency factor η is calculated based on this, and the foam expansion ratio is output as slightly lower and the coverage is incomplete, but it is sufficient to form a preliminary covering layer and block the oxygen supply.

[0028] The precise fire suppression stage is used to completely extinguish the flames and prevent deep reignition, requiring the foam liquid to be highly uniform and the functional components to be fully dispersed. The mixing strategy involves switching to a multi-stage gradient mixing path, where the fluid sequentially passes through three series mixing chambers (total volume approximately 2.0L). A 28kHz ultrasonic transducer (power 100–300W) is activated at the outlet of the second-stage chamber to generate a cavitation effect, promoting the stable suspension and dispersion of micron / nano-scale additives (such as phase change microcapsules and insulating microspheres). During the assessment phase, monitor and suppress potential reignition with minimal agent consumption, and avoid over-spraying; Combined strategy: Use pulsed micro-dosing, adding only anti-reignition agents (such as slow-release polymers) or foam stabilizers (such as silicones), with a flow rate typically 0.1%–0.5% of the main water flow; Dynamic adjustment of the blending mode: If only water-soluble foam stabilizer is added, a single-stage static mixer (η≈0.5) should be used. If nanoparticles or microcapsules are present, ultrasound is used briefly (η≈0.7–0.8). The mixing efficiency factor η is dynamically adjusted according to the pulse frequency: For short pulse intervals (e.g., 20s) requiring rapid mixing, η should be set relatively low (≈0.5). A long pulse interval (e.g., 60s) allows for thorough mixing, and η is set relatively high (≈0.8). Based on the coverage effect and temperature trend of the previous pulse, the system automatically selects whether to enable ultrasound, achieving on-demand mixing.

[0029] In summary, by uniformly characterizing the mixing performance under different hardware configurations using η, fast mixing and fine mixing become comparable and controllable. The η value, as an important feedback parameter of the controller, can be used to determine whether to extend the mixing time or increase the ultrasonic power, avoiding the waste of ultrasonic energy in stages where high homogeneity is not required, ensuring mixing quality in critical stages, improving the overall energy efficiency ratio of the system, ensuring that high-value functional components (such as phase change materials) are fully dispersed in the precise stage to maximize their effectiveness, and preventing fire extinguishing failure due to poor mixing.

[0030] Example 3 This embodiment is an improvement upon embodiment 2. For details, please refer to [link / reference]. Figures 1-2 Fire scene information includes flame images, thermal imaging data, gas concentrations, and environmental parameters; The methods for edge AI models to identify fire types and environmental conditions include: Fire scene information is synchronized and preprocessed in time to generate multimodal input frames containing visual frames, temperature distribution maps, gas concentration vectors, and environmental parameters; Multimodal input frames are fused and inferred using an edge AI model; The output is based on visual frame recognition of flame shape, color, dynamic features, and smoke density; The output is based on the temperature distribution map, which extracts hotspot locations, maximum temperature values, and temperature gradients. The output identifies characteristic gas combinations based on gas concentration vectors to determine the type of combustible material. The wind speed, ambient temperature, humidity, and altitude information are determined by combining environmental parameters. Based on the fusion inference results, a predefined fire classification standard library is matched to output the fire type and its confidence level. Structured fire labels are generated by integrating environmental parameters. The structured fire labels are standardized data structures that are machine readable and include fields for fire type, environmental conditions, hazard indication, and initial response suggestions. Structured fire tags are transmitted to the fire suppression controller as a basis for decision-making regarding the activation of multi-component foam formulations and the setting of mixing strategies.

[0031] To achieve efficient fusion, the raw fire scene information is first synchronized and preprocessed. Specifically, based on the master control clock, the data from each sensor are interpolated and aligned or held at zero order to construct a unified time window (e.g., a sensing cycle every 2 seconds). Subsequently, various data were standardized: the visible light images were normalized in size and scaled to the [0, 1] range in pixel value; Thermal imaging data is converted into a temperature matrix and grayscale distribution is extracted; The gas concentration vector is normalized according to the sensor range and a moving average filter is applied to suppress noise; environmental parameters are standardized using Z-score based on historical statistical characteristics. A structured multimodal input frame is ultimately generated, containing four core components: a visual frame (RGB image), a temperature distribution map (thermograph form), a gas concentration vector (multidimensional temporal features), and an environmental parameter vector (scalar set). This multimodal input frame is fed into a lightweight multi-task edge AI model deployed on an edge computing device (such as an embedded AI controller with NPU acceleration capabilities) for fusion inference. This model employs a multi-branch parallel architecture, with each branch extracting features for a specific modality. The visual branch, based on an improved EfficientDet or YOLOv8 network, performs object detection and semantic analysis on the visual frame, outputting flame morphology (e.g., jet fire, pool fire, smoldering), flame color distribution (to distinguish between oil-based yellow flames and lithium battery white smoke), dynamic features (e.g., flicker frequency, spread speed), and smoke density level (0–3). The thermal imaging branch uses image processing algorithms to extract hotspot location coordinates, maximum temperature values, and temperature gradient fields from temperature distribution maps to determine thermal runaway areas and fire development trends. The gas analysis branch uses 1D convolutional neural networks or long short-term memory networks to perform time-series modeling of gas concentration vectors and identify characteristic gas combination patterns (e.g., a significant increase in HF concentration accompanied by an increase in CO can be identified as thermal runaway of lithium batteries; alcohol fires exhibit high VOCs and low HF characteristics), thereby inferring the type of combustible material. The environmental context branch directly parses the environmental parameter vector to determine wind speed (affecting foam coverage stability), ambient temperature (determining whether to use the antifreeze formula), relative humidity, and altitude (which, through air pressure conversion, affects the foaming ratio setting).

[0032] The feature vectors output from each of the above branches are fed into a multilayer perceptron (MLP) fusion layer, which introduces a Squeeze-and-Excitation (SE) attention mechanism to automatically learn the contribution weights of different modalities in the current fire scene (e.g., reducing visual weights and increasing gas and thermal imaging weights when dense smoke obscures vision), thus achieving robust cross-modal feature fusion.

[0033] The fused high-dimensional feature vector is input into the classification head and matched against a predefined fire classification standard library. This standard library is built based on national standards and industry experience, covering typical types such as Class A solid fires, Class B non-water-soluble liquid fires, polar solvent fires, lithium battery thermal runaway fires, kitchen grease fires, and metal dust fires. The model outputs the probability distribution of fire types and their corresponding confidence scores (e.g., lithium battery thermal runaway: 0.92). Based on this, a structured fire label is automatically generated by integrating fire type, environmental conditions, and hazard indicators (e.g., HF detection, high-temperature gradient). This label is a machine-readable standardized data structure (preferably JSON or Protocol Buffers format) and contains at least the following fields: Fire type field: enumerated values, taken from the standard library; Environmental conditions field: includes values ​​for wind speed, temperature, humidity, altitude, etc. Hazard indication field: Boolean or numerical type, indicating the presence of risks such as HF, high CO, strong winds, etc.; Initial treatment suggestion field: recommended initial mixing ratio, whether to activate anti-alcohol components, whether to enhance foam stabilization strategies, etc.

[0034] Ultimately, the structured fire tag is transmitted in real time to the fire suppression controller via a local bus (such as CAN or Modbus TCP), serving as the core decision-making basis for subsequent calls to multi-component foam formulation templates, setting the mixing efficiency factor η, configuring flow path modes, and activating metering units. The entire identification process is completed at the edge, with an end-to-end latency of no more than 500 milliseconds, without relying on the cloud, ensuring reliable operation even in communication interruptions or high-interference environments.

[0035] Example 4 This embodiment is an improvement upon embodiment 3. For details, please refer to [link / reference]. Figures 1-2 The structured fire label uses JSON format, and the value of its fire type field is limited to a preset enumeration set; When any sensor data is missing or the confidence level is below the threshold, the edge AI model automatically reduces the weight of that modality and relies on the remaining valid modalities to complete the inference, ensuring that valid fire tags are still output even when some sensors fail.

[0036] To achieve standardized representation of fire information and seamless integration with downstream control systems, the structured fire tags generated by the system are encapsulated in JSON format. This data structure offers advantages such as lightweight design, high readability, and cross-platform compatibility, facilitating efficient communication between edge controllers and fire suppression execution units. The fire type field (e.g., fire_type) is strictly limited to a predefined enumeration set, which is predefined based on national standards and typical industrial fire scenarios. This enumeration constraint effectively prevents illegal or fuzzy classification results from entering subsequent control processes, improving system robustness and decision reliability.

[0037] Considering the complex and variable nature of actual fire scene environments, some sensors may experience data loss or inference confidence levels below preset thresholds (e.g., visual recognition confidence level <0.6, gas sensor drift exceeding limits) due to obstruction, contamination, power outages, or signal interference. Therefore, the edge AI model of this invention incorporates a dynamic modal weight adjustment mechanism. Specifically, during the multimodal fusion stage, the model uses an attention module or a gated fusion network to evaluate the reliability of each input modality in real time. If a modality's data is missing or its feature confidence level is below a set threshold, the weight coefficient of that modality in the fusion process will be automatically reduced (or even set to zero), while the contribution ratio of other effective modalities (such as thermal imaging, gas, and environmental parameters) will be increased.

[0038] For example, when the visible light camera is completely obscured by dense smoke, the system will primarily rely on the hotspot distribution of infrared thermal imaging and the trend of HF gas concentration to make a joint judgment, and can still output a high-confidence lithium battery thermal runaway tag. This mechanism ensures that even under extreme conditions where some sensors fail, effective and usable structured fire tags can still be generated, avoiding the interruption of the entire firefighting process due to a single point of failure, and significantly improving the system's fault tolerance and practical adaptability.

[0039] Furthermore, to prevent abrupt changes in flow commands received by metering pumps or control valves during transitions between different fire suppression stages, such as from rapid suppression to precise fire suppression or dynamic adjustment of component formulations, which could lead to problems like water hammer in pipelines, pump overload, or turbulent mixing flow fields, this invention introduces a flow ramp transition control strategy. This strategy smoothly adjusts the actual output flow command of the components through linear interpolation, and its mathematical expression is as follows: in, The actual output flow rate command for component i at time t; The current traffic value before the switch. The target flow rate is required for the new phase. For ramp time, The current system time. This is the trigger moment for the phase transition; Through this linear ramp transition, the flow command is... The transition from the old value to the new value is continuous and smooth within a short period of time, effectively suppressing the mechanical shock of the actuator and the dynamic disturbance of the fluid system.

[0040] Example 5 This embodiment is an improvement upon embodiment 4. For details, please refer to [link / reference]. Figures 1-2 It also includes: During the firefighting process, fire scene feedback data is continuously collected at a set sampling period; Fire feedback data includes flame area change rate, maximum temperature decay rate, characteristic gas concentration trend, and foam coverage uniformity; The current fire extinguishing effect is evaluated based on fire feedback data. If the fire extinguishing effect fails to reach the preset threshold within several consecutive sampling cycles, the parameters for the next control cycle are dynamically adjusted.

[0041] Specifically, during firefighting operations, the system continuously collects multi-dimensional fire scene feedback data at a set sampling period (e.g., every 2 seconds or every 5 seconds) to form a basis for real-time performance evaluation. The fire scene feedback data includes, but is not limited to, the following four key indicators: Flame area change rate: The flame pixel area is calculated frame by frame using a flame region segmentation algorithm in the visible light video stream (such as a semantic segmentation model based on U-Net or MaskR-CNN), and the relative change rate (% / s) per unit time is obtained to characterize the trend of fire spread or contraction. Maximum temperature decay rate: Extracting the highest temperature point in the scene based on infrared thermal imaging data. Calculate its slope over time. A larger negative value indicates a more significant cooling effect; Characteristic gas concentration trends: Monitor the concentration time-series curves of key combustion products (such as CO, VOCs, HF, etc.) to determine whether they show a continuous downward trend; for example, a stable decrease in CO concentration indicates that the combustion reaction is effectively suppressed. Foam coverage uniformity: Image analysis technology is used to evaluate the texture and color consistency of the foam spraying area, and the coverage (%) and coefficient of variation (CV) are calculated to reflect the density and integrity of the foam distribution.

[0042] Based on the above feedback data, the system constructs a comprehensive fire extinguishing effectiveness index E. This index can be calculated using a weighted fusion method, for example: in, This represents the change in flame area. The initial area, Characteristic gas concentration, Scoring for foam coverage uniformity - These are weighting coefficients that are dynamically adjusted based on the type of fire. Subsequently, the system will combine the comprehensive fire extinguishing efficiency index E with the preset threshold. A comparison is made. This threshold is not a fixed value, but is adaptively set according to the current fire type, environmental conditions, and stage objectives (for example, the threshold is lower for the rapid suppression stage and higher for the precision extinguishing stage). If the extinguishing effectiveness index E does not reach the preset threshold within N consecutive sampling periods (N is a positive integer, usually 2 or 3), the comparison is made accordingly. If the current fire extinguishing strategy is deemed insufficient, the system will automatically trigger a dynamic parameter adjustment mechanism to optimize at least one of the following parameters for the next control cycle: Increase the volume ratio of foam stabilizer or anti-reignition additive by 10% to 30% to enhance the strength and thermal stability of the foam film; Increase the output power of the ultrasonic transducer (e.g., from 200W to 280W) or extend the ultrasonic treatment time to improve the uniformity of difficult-to-disperse components (e.g., microcapsules, nanoparticles). Adjust the mixing efficiency factor η to a higher value, and improve the mixing quality by extending the mixing residence time or enabling more mixing chamber stages; Activate backup spray lines or adjust nozzle angles to expand foam coverage and compensate for blind spots caused by wind speed or obstacles.

[0043] All adjustments are transmitted in real time to the metering unit, flow path switching valve, and ultrasonic drive module via the edge controller. Even if there are deviations in the initial fire identification or sudden changes in fire conditions (such as sudden changes in wind direction or secondary fuel leakage), the strategy can be corrected in a timely manner through feedback data to avoid fire extinguishing failure. By introducing a closed-loop control architecture that integrates multi-dimensional fire scene feedback acquisition, quantitative performance evaluation, and dynamic parameter optimization, the system achieves a shift from open-loop spraying to a spraying effect that is adjusted while firing and becomes more accurate over time.

[0044] Example 6 A mixing device for using a multi-component mixing method for foam extinguishing agents includes: A multimodal sensing device is used to collect fire scene information in real time, including flame images, thermal imaging data, gas concentrations, and environmental parameters. Edge AI Controller: The edge AI controller is deployed locally and communicates with multimodal sensing devices to perform fusion reasoning on fire scene information, identify fire type and environmental conditions, and generate structured fire labels. The corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database based on the structured fire label; Main water flow metering unit, used to measure the main water flow rate in the water supply pipeline in real time; Multiple independent functional component dosing units store and measure the base foam concentrate and at least two functional additives, including any combination of anti-alcohol polymers, foam stabilizers, insulating microspheres or phase change microcapsules. Each dosing unit is controlled by an edge AI controller to do so collaboratively according to the target flow rate. A reconfigurable mixing device includes a direct mixing path and a multi-level gradient mixing path. The direct mixing path includes a single-stage static mixer, and the multi-level gradient mixing path includes a series of multi-stage mixing cavities and an ultrasonic homogenizing unit integrated in at least one stage mixing cavity. The reconfigurable mixing unit is also equipped with a flow path switching mechanism controlled by an edge AI controller, which is used to open a direct mixing path during the rapid suppression phase, bypassing the ultrasonic homogenizing unit, and outputting a preliminary mixture. During the precise fire suppression phase, a multi-level gradient mixing path is activated, and an ultrasonic homogenizing unit is employed to achieve high-precision mixing. The feedback sensing module is used to continuously collect fire feedback data during the fire extinguishing process. The fire feedback data includes the rate of change of flame area, the rate of temperature decay, the trend of characteristic gas concentration, and the foam coverage status. The closed-loop optimization module is located inside the edge AI controller. It evaluates the fire extinguishing effect based on fire feedback data and dynamically adjusts the component ratio, ultrasonic power, mixing efficiency factor or spraying strategy for the next control cycle when the fire extinguishing effect does not reach the threshold. The foam output interface connects to the outlet of the reconfigurable mixing device and is used to deliver the mixed foam liquid to the foam generator.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for mixing multiple components of a foam fire extinguishing agent in a specific ratio, characterized in that, Includes the following steps: Real-time fire scene information is obtained through multimodal sensing; Based on the fire scene information, an edge AI model is used to identify the fire type and environmental conditions, and a structured fire label is generated. According to the fire label, the corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database. The three-stage fire extinguishing formula template includes a rapid suppression stage, a precise fire extinguishing stage, and a maintenance assessment stage. Based on the main water flow rate, the target flow rate of each functional component is calculated according to the current component ratio, and multiple independent metering units are controlled to add water in a coordinated manner. Based on the current stage, the mixing mode is dynamically switched. In the rapid suppression stage, a low-mixing-efficiency direct path is used to output the initial mixture. In the precise fire extinguishing stage, a multi-level gradient mixing chamber and ultrasonic homogenization are used to perform high-precision mixing. During the firefighting process, fire scene feedback data is continuously collected, and the component ratio or mixing parameters for the next cycle are dynamically adjusted based on the feedback data. Once the fire is under control, it automatically transitions to the maintenance assessment phase, implementing pulsed micro-spraying and continuously monitoring the risk of reignition until the fire is confirmed to be extinguished.

2. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 1, characterized in that: In the three-stage fire extinguishing formula template, different mixing efficiency factors η are defined for each stage; Wherein, the mixing efficiency factor η of the rapid suppression phase is smaller than the mixing efficiency factor η of the precision extinguishing phase; The maintenance assessment phase is used to add trace components that prevent reignition or stabilize foam, and the mixing efficiency is dynamically adjusted according to the pulse frequency. The mixing efficiency factor η is calculated as follows: in, This refers to the actual residence time of the fluid in the effective mixing region. The theoretical minimum time required to achieve 95% uniformity.

3. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 2, characterized in that: The fire scene information includes flame images, thermal imaging data, gas concentrations, and environmental parameters; The method for identifying fire types and environmental conditions using the edge AI model includes: The fire scene information is synchronized and preprocessed in time to generate a multimodal input frame containing visual frames, temperature distribution maps, gas concentration vectors, and environmental parameters. Multimodal input frames are fused and inferred using an edge AI model; The output identifies the flame shape, color, dynamic features, and smoke density based on the visual frame. The output extracts hotspot locations, maximum temperature values, and temperature gradients based on the temperature distribution map. The output identifies characteristic gas combinations based on the gas concentration vector to determine the type of combustible material. Based on the aforementioned environmental parameters, wind speed, ambient temperature, humidity, and altitude information are determined. Based on the fusion inference results, a predefined fire classification standard library is matched to output the fire type and its confidence level, and a structured fire label is generated by integrating environmental parameters. The structured fire label is a machine-readable standardized data structure that includes fire type field, environmental condition field, hazard indication field and initial handling suggestion field. The structured fire tags are transmitted to the fire suppression controller as a basis for decision-making regarding the activation of multi-component foam formulations and the setting of mixing strategies.

4. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 3, characterized in that: Based on structured fire tags, the system automatically calls up an appropriate three-stage fire extinguishing formula template, based on measured main water flow rate. Real-time calculation of the target dosing flow rate of each functional component This output is then sent to each metering pump / valve actuator to achieve closed-loop precise dosing. The calculation formula is as follows: in, This is the volume percentage defined in the formula; The edge AI model includes: The object detection sub-network of the EfficientDet architecture is used to detect flame and smoke areas from visible light video streams; A temporal analysis subnetwork of ResNet and spatiotemporal attention mechanism is used to analyze the flame flickering frequency and spread trend. 1D convolutional neural network or LSTM subnetwork is used to process multi-gas concentration time series data to identify characteristic combustion products; The multilayer perceptron fusion layer is used to weight and fuse visual features, thermal imaging features, gas features, and environmental parameters to output the final classification result.

5. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 4, characterized in that: The structured fire label uses JSON format, and the value of its fire type field is limited to a preset enumeration set; When any sensor data is missing or the confidence level is below the threshold, the edge AI model automatically reduces the weight of that modality and relies on the remaining valid modalities to complete the inference, ensuring that valid fire tags are still output even when some sensors fail. To prevent sudden changes in the flow command of metering pumps / valve when switching between different fire suppression stages or adjusting component formulations, the actual output flow command is adjusted via a ramp transition: in, The actual output flow rate command for component i at time t; The current traffic value before the switch. The target flow rate is required for the new phase. For ramp time, The current system time. This is the trigger moment for phase switching.

6. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 1, characterized in that, The method for dynamically switching hybrid modes includes: During the rapid suppression phase, the control fluid pipeline is switched to a straight path, so that the main water flow and the base foam concentrate are initially mixed only through a single static mixer, bypassing the ultrasonic homogenizing unit and multi-stage fine mixing chambers, and outputting the initial mixture with a low mixing efficiency factor η to ensure that the first dose of foam is sprayed out quickly. During the precise fire extinguishing phase, the fluid pipeline is switched to a series path, so that the main water flow passes through a multi-stage gradient mixing chamber in sequence, and an ultrasonic transducer is activated in at least one stage of the mixing chamber for auxiliary dispersion, so as to achieve high-precision mixing of multi-functional components with a high mixing efficiency factor η. The multifunctional component includes at least two of the following: anti-alcohol polymer, foam stabilizer, insulating microspheres, or phase change microcapsules; The switching between the rapid suppression phase and the precise fire extinguishing phase is triggered by structured fire tags, and the target dosage of each component is dynamically calculated based on the real-time main water flow rate.

7. The method for mixing the multi-component components of a foam fire extinguishing agent according to claim 6, characterized in that, Also includes: During the firefighting process, fire scene feedback data is continuously collected at a set sampling period; The fire feedback data includes the flame area change rate, the highest temperature decay rate, the characteristic gas concentration trend, and the foam coverage uniformity. The current fire extinguishing effect is evaluated based on the fire feedback data. If the fire extinguishing effect does not reach the preset threshold within multiple consecutive sampling cycles, the parameters of the next control cycle are dynamically adjusted.

8. The mixing apparatus for the multi-component mixing method of foam extinguishing agent according to any one of claims 1-7, characterized in that, include: A multimodal sensing device is used to collect fire scene information in real time, including flame images, thermal imaging data, gas concentrations, and environmental parameters. An edge AI controller, which is deployed locally and communicates with the multimodal sensing device, is used to perform fusion reasoning on the fire information, identify the fire type and environmental conditions, and generate structured fire tags. The corresponding three-stage fire extinguishing formula template is retrieved from the preset formula database based on the structured fire label; A main water flow metering unit, which is used to measure the main water flow rate in the water supply pipeline in real time; Multiple independent functional component dosing units store and measure the base foam concentrate and at least two functional additives, the functional additives including any combination of anti-alcohol polymers, foam stabilizers, insulating microspheres or phase change microcapsules, and each dosing unit is controlled by the edge AI controller to do so in a coordinated manner according to the target flow rate. A reconfigurable mixing device includes a through mixing path and a multi-level gradient mixing path. The through mixing path includes a single-stage static mixer, and the multi-level gradient mixing path includes a series of multi-stage mixing cavities and an ultrasonic homogenizing unit integrated in at least one stage mixing cavity. The reconfigurable mixing device is also equipped with a flow path switching mechanism, controlled by the edge AI controller, for opening the direct mixing path during the rapid suppression phase, bypassing the ultrasonic homogenizing unit, and outputting a preliminary mixture. During the precise fire extinguishing phase, the multi-level gradient mixing path is activated, and the ultrasonic homogenizing unit is enabled to achieve high-precision mixing. The feedback sensing module is used to continuously collect fire scene feedback data during the fire extinguishing process. The fire scene feedback data includes flame area change rate, temperature decay rate, characteristic gas concentration trend and foam coverage status. A closed-loop optimization module is set inside the edge AI controller. It evaluates the fire extinguishing effect based on the fire feedback data and dynamically adjusts the component ratio, ultrasonic power, mixing efficiency factor or spraying strategy for the next control cycle when the fire extinguishing effect does not reach the threshold. The foam output interface is connected to the outlet of the reconfigurable mixing device and is used to deliver the mixed foam liquid to the foam generator.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images