Skin burn risk prediction method and intelligent interaction system

Abstract

The invention relates to the technical field of human body thermal protection, in particular to a fireman skin burn risk prediction method and an intelligent interaction system. The method specifically comprises the following steps: step 1, acquiring and inputting multi-source heat and humidity transfer characteristic data; 2, training a burn risk assessment model constrained by physical information; step 3, burn risk assessment is carried out through the intelligent decision-making workflow; and step 4, outputting a prediction result and a decision suggestion through the intelligent interaction platform. According to the method, the technical limitation of a traditional burn prediction method in the aspects of calculation efficiency and interactivity is solved, accurate prediction of the burn risk is achieved, and an intelligent solution is provided for individualized fire-fighting equipment optimization design and human body safety prediction in a fire scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of human thermal protection, in particular to a firefighter skin burn risk prediction method and an intelligent interaction system. BACKGROUND

[0002] When firefighters perform tasks in high-temperature thermal radiation environments, the risk of skin burns is extremely high. As the core protective equipment, the thermal protection performance of the fire suit is directly related to the safety of the firefighters. The traditional skin burn risk prediction method is mainly based on finite element numerical simulation. By establishing a heat transfer model of the environment-clothing-human system, combining the Pennes skin bioheat transfer model and the Henriques skin burn integral model, the skin temperature field distribution is calculated and the burn time is predicted. However, this method has significant limitations: Firstly, the traditional finite element model relies on standardized assumptions and physical equations, and the data generalization ability is insufficient, resulting in deviations between the predicted results and the actual individualized scenarios. Secondly, the numerical simulation method has poor interactivity and cannot directly provide thermal protection recommendations based on the prediction results, which lacks flexibility.

[0003] In addition, although pure data-driven machine learning algorithms (such as support vector machines, random forests, etc.) can improve prediction accuracy, they also face problems such as overfitting, high computational cost, and reliance on large-scale labeled samples. Moreover, it is difficult to obtain experimental data in extreme fire environments, the prediction results may violate the basic laws of thermodynamics, and lack of explainability and reliability. SUMMARY

[0004] The main purpose of the present application is to overcome the deficiencies in the prior art and provide a firefighter skin burn risk prediction method and an intelligent interaction system. It provides a more accurate, efficient and reliable solution for firefighter thermal protection, which has important practical application value.

[0005] The technical scheme adopted by the present application to achieve its technical purpose is: a firefighter skin burn risk prediction method based on a physical information neural network (PINN) model, specifically including the following steps: S1, constructing a multi-dimensional heat and moisture transfer feature dataset, including environmental parameters, clothing parameters and human physiological parameters; S2, training an initial deep learning model for burn risk prediction; S3, verifying the prediction results based on the physical constraint mechanism to obtain an optimized deep learning model; S4, obtaining the burn risk prediction results and outputting them through an intelligent interaction interface.

[0006] S1→S2→S3→S4, forming a complete "data construction→model training→verification optimization→result output" technical chain.

[0007] wherein: The environmental parameters include radiant temperature and heat exposure time; the environmental parameters (radiant temperature, heat exposure time) determine the external thermal load.

[0008] The clothing parameters include the thickness, fiber density, thermal conductivity coefficient, standard moisture regain of the fibers under 65% relative humidity, and emissivity of the outer fabric of the fire-fighting protective clothing; the clothing parameters (thickness, thermal conductivity coefficient, etc.) quantify the thermal insulation performance of the protective clothing.

[0009] The human physiological parameters include the thickness of the skin epidermis layer, dermis layer, and subcutaneous tissue; the human physiological parameters (skin layer thickness): affect the speed of heat transfer to the deep tissue.

[0010] Preferably, in S1, the construction of the multi-dimensional heat and moisture transfer characteristic data set comprises: S11, obtaining large-scale "environment-clothing-human" system heat and moisture transfer characteristic data under fire rescue environment through finite element numerical simulation; The finite element numerical simulation is based on a multi-physical field heat and moisture transfer coupling model, which includes a gas phase heat transfer model, a solid-liquid phase heat transfer model, and a water mass conservation model of the "environment-clothing-human" system under fire rescue environment; S12, establishing a mapping relationship of multi-dimensional characteristics of the "environment-clothing-human" heat and moisture coupling transfer process, the mapping relationship is embodied as a quantitative association between input features and output labels; Wherein the input features include environmental parameters, clothing parameters, and human physiological parameters, and the output features include secondary burn time and tertiary burn time; S13, sensitivity evaluation and preprocessing of key parameters in the heat and moisture transfer process under fire environment, the evaluation uses a random forest feature importance algorithm to calculate the contribution of each parameter to the output label.

[0011] S11→S12→S13, sequentially complete data generation, mapping relationship establishment, and parameter optimization.

[0012] S11 generates heat and moisture transfer data of the "environment-clothing-human" system through finite element simulation, coupling gas phase heat transfer, solid-liquid phase heat transfer, and mass conservation model, to ensure the physical authenticity of the data. S12 establishes a quantitative mapping between input (environment / clothing / physiological parameters) and output (burn time), and clearly defines the prediction target. Through the random forest algorithm, key parameters (such as air layer thickness, radiant temperature) are screened, redundant data is removed, and the model efficiency is improved.

[0013] Preferably, in S2, the construction of the deep learning model comprises: S21, Establish a deep learning model architecture for processing thermal-humidity coupling transfer characteristic data, using a Physical Information Neural Network (PINN) model, the model architecture consisting of an artificial neural network, thermodynamic physical equations and a feedback mechanism; S22 defines the heat and moisture coupling transfer problem of the "environment-clothing-human body" system, and introduces the thermodynamic partial differential equation into the loss function, which includes data error term and physical information error term. S23 performs data post-processing and plotting, predicts skin temperature changes, assesses the risk of second- and third-degree burns, and outputs the time of second- and third-degree burns.

[0014] S21→S22→S23, construct the PINN model and complete training and output.

[0015] S21 employs the PINN architecture, combining neural networks with thermodynamic equations to enhance the model's physical interpretability. S22 embeds thermodynamic partial differential equations into the loss function, employing dual constraints of data error terms and physical information error terms during the training process. S23 outputs skin temperature change curves and burn time, directly supporting firefighters' safety decisions.

[0016] Preferably, in S3, the physical constraint mechanism includes: S31 uses the finite element simulation results as prior knowledge for training deep learning models, dividing them into training and testing sets. S32 embeds the laws of thermodynamic physics into the loss function in the form of partial differential equations to constrain the physical rationality of the prediction results; S33, using small-scale physical experimental data to test the accuracy of the burn risk assessment model constrained by physical information, wherein the experimental data is obtained through an environmental chamber and a dummy device.

[0017] S31→S32→S33, verify the physical rationality and accuracy of the model in stages.

[0018] S31 divides the limited metadata into training and testing sets to evaluate the model's generalization ability. S32 constrains the loss function through partial differential equations to prevent predictions from violating physical laws. S33 calibrates the model using environmental chamber and dummy experimental data to ensure that predictions are consistent with real-world scenarios.

[0019] Preferably, in step S4, the implementation of the intelligent interaction interface includes: S41, use the multi-dimensional heat and moisture transfer feature dataset as the historical feature data of the intelligent interaction platform; S42 constructs a knowledge base document from prior thermodynamic physics knowledge and firefighter safety protection recommendations, and embeds it into the intelligent decision-making workflow through large language model and retrieval-enhanced generation technology (RAG); S43 integrates the firefighter safety protection prediction model constrained by the trained physical information into the intelligent decision-making workflow, and outputs real-time prediction data of second- and third-degree burn time.

[0020] S41→S42→S43 enables intelligent interaction between data, knowledge, and models. S41 uses historical data as the foundation of the interaction platform, supporting parameter comparison and trend analysis. S42 uses RAG technology to match user questions with the knowledge base, generating protection suggestions. S43 allows the model to respond to input parameters in real time, outputting prediction results and forming a closed-loop decision chain.

[0021] This invention also provides an intelligent interactive system for assessing the risk of skin burns in firefighters, which is used to implement the above-mentioned method, specifically including: S5, the data fusion module, is used to integrate real-time user input data with historical feature data; S6, Prediction Calculation Module, deploys a firefighter safety protection prediction model constrained by the trained physical information; S7, the decision support module, is used to output the distribution of human burn risk and to specially mark high-risk areas of fire-fighting clothing protection performance.

[0022] S5→S6→S7, the system modules work together to complete data integration, prediction and decision support.

[0023] S5 (Data Fusion Module) integrates real-time and historical data to ensure input consistency. S6 (Predictive Calculation Module) deploys the PINN model, the core computing unit. S7 (Decision Support Module) identifies high-risk areas and provides suggestions, directly guiding the design of fire suits.

[0024] Preferably, the data fusion module supports receiving new training data from the user to fine-tune the burn risk assessment model and applying the model to the user-provided data samples. The data fusion module supports model fine-tuning to adapt to new data scenarios.

[0025] Preferably, the prediction and calculation module deploys a burn risk assessment model constrained by physical information to calculate changes in skin temperature and the risk of second- and third-degree burns. The prediction module outputs skin temperature and burn time to quantify the risk level.

[0026] Preferably, the decision support module provides quantitative technical suggestions for the material selection and structural design of fire-fighting clothing. The decision module provides material selection suggestions (such as optimizing the air layer thickness) to address weak points in protection.

[0027] Compared with existing technologies, the beneficial effects of this invention are as follows: This artificial intelligence-based method and intelligent interactive system for predicting the risk of skin burns to firefighters has the following significant advantages: It can achieve high-precision prediction: By integrating physical information neural network (PINN) with thermodynamic physical equations and embedding differential equation constraints into the loss function, the model can capture complex nonlinear relationships and follow physical laws, and the prediction accuracy is significantly better than that of traditional numerical models.

[0028] High interpretability and reliability: By ranking the importance of features and parametric analysis, the influence mechanism of key design parameters on thermal protection performance is clarified, guiding the optimization design of fire suits and enhancing the scientific nature of decision-making recommendations.

[0029] It enables intelligent interaction and efficient decision-making: By integrating predictive models and professional knowledge bases, the system generates burn risk prediction and protection plans in real time after the user inputs environmental or clothing parameters, realizing a closed loop of "data input-prediction-decision" and greatly improving emergency response efficiency.

[0030] It can combine physical constraints with data-driven approaches: by using finite element simulation data as prior knowledge and combining it with small-scale physical experiments to calibrate the model, it solves the overfitting problem of pure data-driven methods, while ensuring that the prediction results conform to thermodynamic laws.

[0031] This invention overcomes the technical limitations of traditional burn prediction methods in terms of computational efficiency and interactivity, achieving accurate prediction of burn risk and providing an intelligent solution for personalized fire-fighting equipment optimization design and human safety prediction in fire scenarios. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Fig. 1 A flowchart for a method and intelligent interactive system for predicting the risk of skin burns to firefighters.

[0034] Fig. 2 A flowchart for a method to predict the risk of skin burns to firefighters. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention. Example 1

[0036] Please see Figs. 1-2 A method for predicting the risk of skin burns to firefighters, specifically including the following steps: Step 1: Acquisition and input of multi-source heat and moisture transfer characteristic data; Step 2: Train a burn risk assessment model constrained by physical information; Step 3: Conduct burn risk assessment through intelligent decision-making workflow; Step 4: Output prediction results and decision suggestions through the intelligent interactive platform.

[0037] Among them, AI-based methods for predicting the risk of skin burns to firefighters include: S1, construct a multi-dimensional heat and moisture transfer feature dataset, including environmental parameters, clothing parameters, and human physiological parameters; including: S11, obtains large-scale heat and moisture transfer characteristic data of the "environment-clothing-human body" system in the fire rescue environment through finite element numerical simulation; The finite element numerical simulation is based on a multi-physics heat and moisture transfer coupling model, which includes a gas phase heat transfer model, a solid-liquid phase heat transfer model, and a water mass conservation model for the "environment-clothing-human body" system under fire rescue conditions. The multi-physics heat and moisture transfer coupling model is established using nonlinear partial differential equations. The implicit finite difference method is used to discretize the nonlinear partial differential equation system, and the Thomas iterative method is used to solve the nonlinear partial differential equation system to obtain heat and moisture transfer characteristic data. S12, establish the mapping relationship of multi-dimensional feature data of the "environment-clothing-human body" heat and moisture coupling transfer process; The multi-dimensional feature data mapping relationship of the "environment-clothing-human body" heat and moisture coupling transfer process is reflected in the quantitative correlation between input features and output labels. Orthogonal combination design is performed on various input parameters, and the time required for the human body to reach second- and third-degree burns under corresponding parameter conditions is obtained by solving a finite element numerical model. As a preferred implementation, the multi-source heat and moisture transfer feature dataset is generated by finite element numerical simulation, taking into account environmental parameters, clothing parameters, and human physiological parameters.

[0038] The clothing is a fire-fighting protective suit worn by firefighters in high-temperature heat radiation environments, consisting of an outer layer, a waterproof and breathable layer, and a heat insulation layer.

[0039] Specifically, environmental parameters mainly include radiation temperature and heat exposure time; clothing parameters mainly include the thickness of each layer of fabric in the fire-fighting protective clothing, fiber density, thermal conductivity, standard moisture regain of fibers at 65% relative humidity, and emissivity of the outer layer of fabric in the fire-fighting protective clothing; human physiological parameters include the thickness of the epidermis, dermis, and subcutaneous tissue. The range of fabric parameters used for training is shown in Table 1.

[0040] Table 1 Fabric parameter range

[0041] Table 2 Environmental Parameters

[0042] S13, conduct sensitivity assessment and importance ranking of key parameters in the heat and moisture transfer process under fire conditions; In this embodiment, by random forest feature importance analysis, the common features that have the greatest impact on burn time and thermal protection performance are identified: air layer thickness (importance > 0.5), radiation temperature, outer layer thickness, insulation layer thickness, and waterproof and breathable layer thickness.

[0043] Furthermore, in the process of constructing a large dataset through iterative calculations of a numerical model for heat and moisture coupling in multilayer fabrics, the obtained dataset may contain some missing, irrelevant, or duplicate values. If these values ​​are used directly for model building, the model results will be biased.

[0044] Therefore, as a preferred implementation, the collected data needs to be processed before model training to remove outliers and replace missing values ​​with the average value, thereby establishing a mapping relationship between feature parameters and the time of second- and third-degree burns.

[0045] Since the features have different dimensions, it is necessary to eliminate the influence of dimensions on the results. If the original data is used directly for modeling, it cannot simultaneously reflect the importance of each feature. Therefore, it is necessary to perform dimensional analysis on all data based on Buckingham's π theorem and normalize the data in the range [0,1] using the following formula: ; S2, Training the initial deep learning model for burn risk prediction; including: S21, Establish a deep learning model architecture for processing thermal-humid coupling transfer feature data; As a preferred implementation, a physical information neural network architecture is used to process the input multi-source heat and moisture transfer feature dataset, which broadly covers environmental, fabric, and human skin parameters to reflect the transient heat and moisture transfer phenomenon between "high-temperature thermal radiation environment - multi-layer fabric - air layer - skin tissue". The deep learning model architecture consists of an artificial neural network, thermodynamic physical equations, and a feedback mechanism. Key parameters for neural network training are defined, including learning rate, batch size, and number of iterations. S22 defines the heat and moisture coupling and transfer problem of the "environment-clothing-human body" system, and uses physical laws to guide the initial deep learning model training; As a preferred implementation, a physical information neural network architecture is used to simulate the heat and mass transfer process between the environment, clothing, and human body under physical information constraints. First, physical constants and material parameters are set, defining physical constants such as time step, spatial step, and total duration, and material parameters such as skin layer thickness and specific heat capacity of fabric layer. Then, an artificial neural network model is created to predict temperature distribution, inputting a multi-source heat and moisture transfer feature dataset generated by finite element simulation, including environmental parameters, clothing parameters, and human physiological parameters. The output target is set as the time for second- and third-degree burns. The core of the deep learning model training lies in introducing physical laws, represented by thermodynamic partial differential equations, into the loss function. The loss function includes a data error term and a physical information error term. The data error term is the difference between the PINN prediction output and the output data mapped from the heat and moisture coupling transfer feature dataset, while the physical information error term is the residual obtained by substituting the PINN prediction output into the thermodynamic partial differential equations. The initial deep learning model is iteratively trained based on this training method. S23, perform data post-processing and plotting, predict skin temperature changes, and assess the risk of second- and third-degree burns.

[0046] The risk assessment for second- and third-degree burns is expressed by the time it takes for a person to develop second- and third-degree burns.

[0047] S3 verifies the prediction results based on physical constraint mechanisms, thereby obtaining an optimized deep learning model; including: S31 uses the finite element simulation results as prior knowledge for model training; The finite element simulation results are used as prior knowledge for training the deep learning model. These results serve as the original dataset, which is divided into a training set and a test set during deep learning model training to evaluate the model's generalization ability. In this embodiment, 80% of the original dataset is allocated to the training set for model training, and 20% is allocated to the test set for evaluating the model's generalization ability. During the training phase, the model adjusts its internal data using known training data. Evaluating the model's performance on the training data helps determine whether the model has learned the patterns in the data. During the testing phase, unseen test data is used for evaluation to test the model's generalization ability on new data, reflecting the model's true performance and effectiveness in practical applications. The coefficient of determination is used. Mean squared error (RMSE) is used as an evaluation metric for model performance. Characterizes the model's fit to data variations, ranging from 0 to 1. The closer to 1, the better the model fit; RMSE is the positive square root of MSE. When comparing errors, RMSE is more intuitive than MSE. The smaller the value, the better the model performance. The formulas for calculating RMSE are as follows: ; ; Where N is the total number of samples, These are experimental values. It is a predicted value. It is the average value of the experiment.

[0048] S32 embeds the thermodynamic equation into the loss function to constrain the physical rationality of the prediction results of the prediction model; The thermodynamic physical laws are embedded into the loss function to quantify the difference between the model prediction and the residual of the physical equation, thereby constraining the physical rationality of the deep learning model's prediction results. The thermodynamic physical laws are embedded into the loss function in the form of partial differential equations to ensure that the prediction results follow known physical laws and improve data utilization efficiency.

[0049] S33 uses small-scale physical experiment data to test the accuracy of the burn risk assessment model constrained by physical information.

[0050] The model's accuracy is further validated using real experimental data to promptly identify and adjust any potential issues. The small-scale physical experiment data, acquired through an environmental chamber and dummy devices, is embedded into the model's loss function as a data-driven loss function to further calibrate the deep learning model, resulting in an optimized model. Specifically, the steps involve quantifying the difference between the model's predictions and the small-scale physical experiment data, and then embedding this data into the model's loss function as the data-driven loss function.

[0051] S4 obtains burn risk prediction results and outputs them through an intelligent interactive interface; including: S41, use the multi-source heat and moisture transfer feature dataset as the historical feature data of the intelligent interaction platform; The multi-dimensional heat and moisture transfer feature dataset was used as historical feature data for an AI-based intelligent interactive platform for predicting firefighter safety protection. As a preferred implementation, when users need to predict burn time or evaluate fabric parameters, the large language model organizes and converts the user's input parameters to maintain dimensional consistency with the embedded intelligent multi-source heat and moisture transfer feature dataset. This step ensures the standardization and consistency of parameters, providing an accurate data foundation for subsequent formula calculations or equation solving, and avoiding calculation errors or result deviations caused by parameter issues.

[0052] S42 embeds prior physical knowledge and professional protection advice into the intelligent decision-making workflow through large language models and retrieval enhancement generation technology; Prior thermodynamic physics knowledge and firefighter safety protection recommendations are constructed into a knowledge base document, which is then embedded into the intelligent decision-making workflow using a large language model and retrieval-enhanced generation (RAG) technology. The matching of the knowledge base document with the user's question is accomplished using the BM25 algorithm. The framework of the intelligent decision-making workflow includes a user question input node, a user intent recognition module, an IF selector, input parameter processing, user query rewriting, firefighter safety protection prediction, retrieval of the knowledge base to summarize firefighter safety protection recommendations, and output results. As a preferred implementation, a knowledge base on the thermal protection of fire suits is first embedded. This knowledge base is a crucial supporting component of the system, containing a rich library of fire suit materials, a library of thermal protection standards, and numerous academic articles related to the thermal protection performance of fire suits. These resources provide strong data support for semantic retrieval, ensuring that the system can quickly extract valuable content from massive amounts of information.

[0053] Subsequently, user queries were rewritten during the interaction. To address users' needs for answers regarding basic knowledge of the thermal protection performance of fire suits, the system rewrote the user's question. This process ensured a better match between the question's wording and the external knowledge base, thereby improving the efficiency and accuracy of subsequent retrieval, generation, and summarization, and providing users with answers that better met their needs.

[0054] Furthermore, the knowledge base utilizes retrieval enhancement generation technology, employing RAG technology and combining the BM25 algorithm with a large language model to match user questions with knowledge base documents, generating natural language answers. This process achieves efficient transformation from knowledge acquisition to answer generation, providing users with more accurate, professional, and targeted solutions.

[0055] S43 integrates the burn risk assessment model constrained by the trained physical information into the intelligent decision-making workflow.

[0056] The firefighter safety protection prediction model, constrained by the trained physical information, is integrated into the intelligent decision-making workflow. The activation process of the firefighter safety protection prediction model includes artificial intelligence recognizing user input, performing fuzzy queries on the knowledge base, organizing, converting and correcting input parameters, and the firefighter safety protection prediction model completing real-time calculations and responses, outputting corresponding second- and third-degree burn time prediction data.

[0057] Specifically, the intelligent decision-making workflow involves first receiving user text input at the input node, triggering the intent recognition module. The user inputs a question through a dialog interface, and the system receives the text information as the basis for subsequent processing.

[0058] Next, intent recognition is performed, with the user's input question first entering the "intent recognition" module. This module uses a large language model to analyze the semantics of the text, accurately identifying the intent behind the question and categorizing it into three types: basic question answering, burn risk prediction, and fabric parameter evaluation, laying the foundation for subsequent targeted processing. An IF selector serves as a key node connecting multiple downstream branches, determining which branch to run based on predefined conditions, thus achieving differentiated processing of different question types and ensuring the system can efficiently and systematically handle various complex problems.

[0059] The burn risk assessment model, which is constrained by the physical information obtained from training, is embedded in the workflow for burn time prediction and fabric parameter evaluation. Example 2

[0060] Based on the above embodiments, this invention also provides an intelligent interactive system for firefighters' skin burn risk, which is used to implement the method described in Embodiment 1, specifically including: S5, the data fusion module, is used to integrate real-time user input data with historical feature data; it includes: It supports fine-tuning the burn risk assessment model by receiving new training data from users; specifically, it uses incremental learning technology to dynamically update model parameters, avoiding repeated training with full data.

[0061] The burn risk assessment model, constrained by the trained physical information, is applied to user-provided data samples. Specifically, the construction of the intelligent interaction system includes automatically identifying the format of user input data (such as Excel, JSON, or sensor stream data), performing feature alignment, matching the input dimension of the model, and performing secondary verification of outliers.

[0062] When the user inputs no missing values ​​for the basic fabric parameters and environmental parameters, and requests a "predicted time of burn occurrence," the intelligent interactive system calculates the time when a firefighter will suffer second- or third-degree burns based on the workflow, providing decision-making suggestions for the firefighter's safe rescue time.

[0063] When a user inputs missing values ​​for basic fabric and environmental parameters and requests that "no burns occur," the intelligent interactive system calculates the missing parameter values ​​based on the workflow and provides decision-making suggestions for the selection of clothing materials by searching the knowledge base.

[0064] S6, the prediction and calculation module, deploys a burn risk assessment model constrained by the trained physical information; including: Deploy a burn risk assessment model constrained by physical information and process user-provided data samples; In this embodiment, the user input parameters utilize a multi-layer fabric system. The study range and baseline values ​​for each parameter are shown in Table 4-1. During the parametric analysis, the ambient radiation temperature was maintained at 458 °C, and other basic performance parameters of different fabric layers remained unchanged, as shown in Table 3.

[0065] Table 3. Range of variation and baseline values ​​of various parameters in the model parameter study.

[0066] Table 4 Basic performance parameters of fabrics with different layers

[0067] Calculate changes in skin temperature and the risk of second- or third-degree burns; In the embodiments, the differences in burn time under different parameter values ​​were investigated. With the increase of air layer thickness, both the second-degree burn time and the third-degree burn time showed a trend of first increasing, then decreasing, and finally stabilizing. When the air layer thickness increased to approximately 3 mm, both the second-degree burn time and the third-degree burn time reached their highest values, indicating the best thermal protection performance of the fabric system. This is because the thermal conductivity of the air layer is lower than that of most insulation materials, and compared to conditions without an air layer, increasing the air layer thickness effectively reduces the heat conduction through the air layer. Furthermore, with the increase of air layer thickness, the burn time decreased slightly and remained stable. This suggests that in fabric systems tested under high-temperature thermal radiation conditions, there is an optimal air layer thickness of 3 mm. This may be because although increasing the air layer thickness enhances the insulation capacity of the fabric system, it also increases the buoyancy of the air layer, leading to natural convection heat transfer and thus increasing heat transfer.

[0068] S7, the decision support module, conducts digital evaluations of the thermal protection performance of fire suits, generates material selection recommendations for fire suits, and provides interactive support via natural language. This includes: Output the distribution of human burn risk; wherein the burn risk distribution is characterized by the following quantitative indicators: burn severity classification of skin tissue; critical time to reach second-degree burn; critical time to reach third-degree burn.

[0069] High-risk areas of fire-fighting suits are specially marked; the criteria for determining high-risk areas are: skin areas predicted to suffer burns of level two or higher; and areas where the critical time threshold for level two burns is lower than the user-preset safety standard. This process is used to automatically detect weak areas in the thermal protection performance of fire-fighting suits and to prominently mark high-risk areas in a visual interface.

[0070] This provides quantitative technical recommendations for the selection of materials and structural design of firefighting clothing. Specifically, these recommendations involve suggesting ranges for thermophysical parameters based on missing values ​​of user-provided input parameters, and offering material and structural optimization solutions based on burn risk distribution predictions.

[0071] The solution in this embodiment can be selectively combined with solutions in other embodiments.

[0072] It should be noted that although the above embodiments have been described herein, this does not limit the scope of patent protection of this invention. Therefore, any changes and modifications made to the embodiments described herein based on the innovative concept of this invention, or equivalent structural, procedural, or functional transformations made using the description and drawings of this invention, directly or indirectly applying the above technical solutions to other related technical fields, are all included within the scope of protection of this invention.

Claims

1. A method for predicting the risk of skin burns to firefighters, implemented based on a physical information neural network model, characterized in that, Specifically, the following steps are included: S1, construct a multi-dimensional heat and moisture transfer feature dataset, including environmental parameters, clothing parameters and human physiological parameters; S2, train the initial deep learning model for burn risk prediction; S3 verifies the prediction results based on physical constraint mechanisms, thereby obtaining an optimized deep learning model; S4 obtains burn risk prediction results and outputs them through an intelligent interactive interface.

2. The method according to claim 1, characterized in that, In step S1, the construction of the multi-dimensional heat and moisture transfer feature dataset includes: S11, obtains large-scale heat and moisture transfer characteristic data of the "environment-clothing-human body" system in the fire rescue environment through finite element numerical simulation; The finite element numerical simulation is based on a multi-physics field heat and moisture transfer coupling model, which includes a gas phase heat transfer model, a solid-liquid phase heat transfer model, and a water mass conservation model for the "environment-clothing-human body" system in a fire rescue environment. S12, establish a mapping relationship of multi-dimensional feature data of the "environment-clothing-human body" heat and moisture coupling transfer process, wherein the mapping relationship is reflected as a quantitative correlation between input features and output labels; The input features include environmental parameters, clothing parameters, and human physiological parameters, while the output features include the time of second-degree burns and the time of third-degree burns. S13, perform sensitivity assessment and preprocessing on key parameters in the heat and moisture transfer process under fire conditions. The assessment uses the random forest feature importance algorithm to calculate the contribution of each parameter to the output label.

3. The method according to claim 1, characterized in that, In S2, the construction of the deep learning model includes: S21, Establish a deep learning model architecture for processing thermal-humid coupling transfer characteristic data, using a physical information neural network model, the model architecture consisting of an artificial neural network, thermodynamic physical equations and a feedback mechanism; S22 defines the heat and moisture coupling transfer problem of the "environment-clothing-human body" system, and introduces the thermodynamic partial differential equation into the loss function, which includes data error term and physical information error term. S23 performs data post-processing and plotting, predicts skin temperature changes, assesses the risk of second- and third-degree burns, and outputs the time of second- and third-degree burns.

4. The method according to claim 1, characterized in that, In S3, the physical constraint mechanism includes: S31 uses the finite element simulation results as prior knowledge for training deep learning models, dividing them into training and testing sets. S32 embeds the laws of thermodynamic physics into the loss function in the form of partial differential equations to constrain the physical rationality of the prediction results; S33, using small-scale physical experimental data to test the accuracy of the burn risk assessment model constrained by physical information, wherein the experimental data is obtained through an environmental chamber and a dummy device.

5. The method according to claim 1, characterized in that, In S4, the implementation of the intelligent interaction interface includes: S41, use the multi-dimensional heat and moisture transfer feature dataset as the historical feature data of the intelligent interaction platform; S42 constructs a knowledge base document by combining prior thermodynamic physics knowledge with firefighter safety protection recommendations, and embeds it into the intelligent decision-making workflow through large language models and retrieval-enhanced generation technology. S43 integrates the firefighter safety protection prediction model constrained by the trained physical information into the intelligent decision-making workflow, and outputs real-time prediction data of second- and third-degree burn time.

6. The method according to claim 1, characterized in that, in: The environmental parameters include radiation temperature and heat exposure time; The clothing parameters include the thickness, fiber density, thermal conductivity, standard moisture regain of the fibers at 65% relative humidity, and emissivity of the outer fabric of the fire-fighting protective suit. The human physiological parameters include the thickness of the epidermis, dermis, and subcutaneous tissue.

7. An intelligent interactive system for implementing the method of any one of claims 1-6, characterized in that, include: S5, the data fusion module, is used to integrate real-time user input data with historical feature data; S6, Prediction Calculation Module, deploys a firefighter safety protection prediction model constrained by the trained physical information; S7, the decision support module, is used to output the distribution of human burn risk and to specially mark high-risk areas of fire-fighting clothing protection performance.

8. The system according to claim 7, characterized in that, The data fusion module supports receiving new training data from users to fine-tune the burn risk assessment model and applying the model to user-provided data samples.

9. The system according to claim 7, characterized in that, The prediction and calculation module deploys a burn risk assessment model constrained by physical information to calculate changes in skin temperature and the risk of second- and third-degree burns.

10. The system according to claim 7, characterized in that, The decision support module provides quantitative technical advice on the selection of materials and structural design for firefighting clothing.

Images