Method for predicting flame retardant property and thermal stability of composite material based on thermogravimetric analysis and infrared spectrum data fusion

Abstract

The invention relates to the technical field of composite material performance prediction, and discloses a composite material flame retardant property and thermal stability prediction method based on thermogravimetric analysis and infrared spectrum data fusion. The invention aims to solve the problem that it is difficult to effectively fuse multi-modal data to realize quantitative prediction in the prior art. The method comprises the following steps: firstly, collecting thermogravimetric and infrared spectrum combined data, calculating gas transmission lag time, and carrying out forward correction on infrared spectrum data to realize time sequence alignment; then solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors are extracted, and a fusion feature matrix is generated; extracting solid-phase and gas-phase characteristics respectively by using a dual-channel LSTM network, and performing weighted fusion based on a self-attention mechanism; and finally, synchronously outputting a flame retardant index predicted value and a flame retardant grade classification result through numerical mapping and classification judgment. According to the method, multi-dimensional feature mining of the whole pyrolysis process of the material is realized, and a rapid and accurate digital evaluation means is provided for screening of the flame-retardant material.

Description

A method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data Technical Field

[0001] This invention relates to the field of composite material performance prediction technology, and in particular to a method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data. Background Technology

[0002] Composite materials, due to their excellent physicochemical properties, are widely used in aerospace, electronics, electrical engineering, construction, and transportation. In these applications, the flame retardancy and thermal stability of materials are key performance indicators for ensuring safety. Thermogravimetric analysis (TGA) coupled with Fourier transform infrared spectroscopy (FTIR) (TG-FTIR) is a standard analytical method for characterizing the thermal degradation behavior and volatile product composition of materials. In conventional applications, the initial decomposition temperature, maximum decomposition rate, and char residue of a material are typically obtained from thermogravimetric curves to evaluate its thermal stability; simultaneously, by analyzing the infrared spectra at the corresponding temperatures, the gaseous products released during pyrolysis are qualitatively identified, thereby inferring its combustion process and flame retardant mechanism. This analytical method provides an effective basis for the preliminary assessment of material properties.

[0003] However, existing technologies still face challenges in using TG-FTIR data for in-depth analysis. Thermogravimetric analysis reflects the mass changes of the solid phase of a material, while infrared spectroscopy analyzes the gaseous products generated after decomposition; both are different dimensions of data describing the same physicochemical process. How to effectively integrate these two types of time-series data—which have different sources, different physical meanings, and complex data structures—and extract deep correlations that can accurately and quantitatively predict the final flame-retardant properties of materials (such as limiting oxygen index and UL-94 rating) is a substantial technical problem that urgently needs to be solved in the field of rapid material screening and digital performance evaluation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, in order to solve the problem that existing technologies are unable to effectively integrate solid-phase thermogravimetric analysis and gas-phase infrared multimodal data, thus failing to achieve quantitative prediction of the flame retardant properties of materials, this invention provides a method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectral data.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The present invention provides a method for predicting the flame retardant performance and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data, which includes the following steps: S1, according to a preset heating rate and gas environment, thermogravimetric data and infrared spectroscopy data of composite material samples are collected using a thermogravimetric and infrared spectroscopy combined device; S2, based on the time difference between the peak occurrences of the thermogravimetric data and the infrared spectroscopy data, the gas transport lag time of pyrolysis products is determined, and the time axis of the infrared spectroscopy data is shifted forward using the gas transport lag time of pyrolysis products to obtain a time sequence pair. S3. Perform feature extraction on the thermogravimetric data and the time-aligned infrared spectral data respectively, construct solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors, and splice them in the time dimension to generate a fused feature matrix; S4. Perform channel-specific processing on the fused feature matrix, extract solid mass change features and gas release features respectively, fuse the extracted features and weight them according to feature importance weights to generate comprehensive feature data; S5. Perform numerical mapping and classification discrimination calculation on the comprehensive feature data to obtain the predicted flame retardant index value and flame retardant level classification result of the composite material sample.

[0007] As a preferred embodiment of the method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data described in this invention, the following steps are taken: Before the data acquisition step, the sample is processed to a preset uniform particle size range, and surface impurities are removed. Specifically, the data acquisition according to the preset gas environment refers to continuously introducing a preset flow rate of protective gas into the heating chamber of the thermogravimetric analyzer during the data acquisition process to isolate external air and maintain a stable inert atmosphere.

[0008] As a preferred embodiment of the method for predicting the flame retardant performance and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectral data described in this invention, the following steps are taken: Based on the time difference between the peak occurrences of the thermogravimetric data and the infrared spectral data, the pyrolysis product gas transport lag time is determined, and the infrared spectral data is corrected by shifting the time axis forward using the pyrolysis product gas transport lag time to obtain time-aligned infrared spectral data. Specifically, determining the pyrolysis product gas transport lag time based on the time difference between the peak occurrences of the thermogravimetric data and the infrared spectral data includes: extracting a first characteristic curve reflecting the change in the material decomposition rate from the thermogravimetric data, and identifying the first characteristic time point with the largest decomposition rate in the first characteristic curve; extracting a first characteristic time point reflecting the total intensity of gas release from the infrared spectral data... The system uses two characteristic curves to identify the second characteristic time point where the gas release intensity is greatest. It calculates the time interval between the second characteristic time point and the first characteristic time point, and determines this time interval as the pyrolysis product gas transport lag time. The system uses this pyrolysis product gas transport lag time to perform time-axis shift correction on the infrared spectral data. Specifically, this includes: obtaining the initial recording time point corresponding to each frame of the infrared spectral data; subtracting the pyrolysis product gas transport lag time from the initial recording time point to obtain the corrected actual reaction time point; and remapping the infrared spectral data to the actual reaction time point, ensuring that each frame of the infrared spectral data is synchronized with the thermogravimetric data at the same physical reaction time point, thereby generating time-aligned infrared spectral data.

[0009] As a preferred embodiment of the method for predicting the flame retardant performance and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectral data described in this invention, the following steps are taken: Feature extraction is performed on the thermogravimetric data and the time-aligned infrared spectral data respectively to construct solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors, which are then concatenated in the time dimension to generate a fused feature matrix. The specific steps are as follows: Constructing the solid-phase thermogravimetric feature vector specifically includes: discretizing the thermogravimetric data at preset time intervals to obtain time-series sample points; extracting the mass value and the derivative value that changes with time corresponding to each time-series sample point; extracting… The key statistical features of the initial decomposition temperature, maximum decomposition rate temperature, and residual mass at the final temperature in the thermogravimetric data are combined with the mass values ​​and their time-varying derivative values ​​to generate a solid-phase thermogravimetric feature vector. The construction of the gas-phase infrared feature vector specifically includes: determining the characteristic wavelength range in the infrared spectral data according to the preset target decomposition gas list and its corresponding infrared characteristic absorption bands, calculating the spectral response intensity index of each time sampling point in the characteristic wavelength range, arranging the spectral response intensity index corresponding to the characteristic wavelength range in chronological order, and generating a multi-channel gas-phase infrared feature vector.

[0010] As a preferred embodiment of the method for predicting the flame retardant performance and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data described in this invention, the fused feature matrix is ​​processed by channel segmentation to extract solid mass change features and gas release features respectively. The extracted features are then fused and weighted according to feature importance weights to generate comprehensive feature data. The specific steps are as follows: The channel segmentation of the fused feature matrix is ​​achieved using a pre-set dual-channel long short-term memory network structure, including parallel solid-phase processing channels and gas-phase processing channels; the solid-phase processing channel is configured to perform time-series modeling of the solid-phase thermogravimetric feature vector and extract the time-dependent features of solid-phase decomposition; the gas-phase processing channel is configured... The system is used to perform time-series modeling of gas phase infrared feature vectors and extract the dynamic evolution features of gas phase release. The dual-channel long short-term memory network structure also includes a feature fusion layer, which splices and fuses the features output from the solid phase processing channel and the gas phase processing channel. Weighting based on correlation is performed by a feature weighting calculation unit, which performs the following operations: receiving fused feature data from the feature fusion layer; calculating the correlation weight of each dimension of the fused feature data to the prediction task according to a preset weight calculation strategy; and using the correlation weight to perform weighting operations on the fused feature data to enhance the contribution of highly correlated features to the final prediction result and suppress the interference of low-correlation features.

[0011] As a preferred embodiment of the method for predicting the flame retardant performance and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data according to the present invention, the following steps are taken: Numerical mapping and classification calculations are performed on the comprehensive feature data to obtain the predicted flame retardant index and flame retardant level classification results of the composite material sample. Specifically, obtaining the predicted flame retardant index includes: constructing a linear mapping channel for numerical regression; inputting the comprehensive feature data into the linear mapping channel; mapping the high-dimensional comprehensive feature data into a one-dimensional continuous value through a fully connected operation; performing inverse normalization on the one-dimensional continuous value to convert it into a specific value conforming to the flame retardant index measurement; and determining the specific value as the predicted flame retardant index value of the composite material sample.

[0012] The specific steps for obtaining the flame retardant rating classification result include: constructing a classification mapping channel for category discrimination, inputting the comprehensive feature data into the classification mapping channel, and calculating the probability score of the comprehensive feature data belonging to the preset flame retardant rating category; normalizing the probability score, selecting the category with the highest probability score as the final flame retardant rating, and determining the flame retardant rating as the flame retardant rating classification result of the composite material sample.

[0013] The beneficial effects of this invention are as follows: By collecting thermogravimetric and infrared spectral data of composite material samples and calculating the gas transport lag time of pyrolysis products to perform time-axis forward correction on the infrared spectral data, the invention achieves temporal alignment between solid-phase and gas-phase data. Solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors are extracted and constructed separately to generate a fused feature matrix. A dual-channel long short-term memory network is used to extract solid-phase mass change and gas-phase release characteristics, and weighted fusion is performed based on correlation weights. Finally, through numerical mapping and classification discrimination calculations, the predicted flame retardant index and flame retardant rating are simultaneously output. This invention achieves in-depth multi-dimensional feature mining of the entire pyrolysis process of composite materials, enabling accurate and rapid prediction of the limiting oxygen index and flame retardant rating of materials, providing an effective digital evaluation method for the research and screening of new flame-retardant materials. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0015] Figure 1 is the overall flowchart of the sorting technology; Figure 2 is a schematic diagram of S2 time alignment and lag compensation; Figure 3 is a flowchart of S3 multimodal feature vector construction; Figure 4 is a flowchart of S4 dual-channel modeling and feature weighting; Figure 5 is a flowchart of S5 dual-branch prediction result output. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. An embodiment appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments. Example 1

[0019] Referring to Figures 1-5, the first embodiment of the present invention provides a method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data, including the following steps: S1, according to a preset heating rate and gas environment, thermogravimetric data and infrared spectroscopy data of composite material samples are collected using a thermogravimetric and infrared spectroscopy combined device. The specific operation steps are as follows: before the data collection step, the sample is further processed to a preset uniform particle size range and surface impurities are removed; collecting data according to a preset gas environment specifically means that during the data collection process, a preset flow rate of protective gas is continuously introduced into the heating chamber of the thermogravimetric analyzer to isolate the external air and maintain a stable inert atmosphere environment.

[0020] It should be noted that before data collection, the samples were processed to a preset uniform particle size range and surface impurities were removed. The specific procedure is as follows: Select the composite material block or sample to be tested, and use a mechanical crusher to initially break it into small pieces. Use a cryogenic grinder to finely grind the small sample pieces under liquid nitrogen cooling conditions to prevent the heat generated during grinding from causing premature thermal degradation of the material.

[0021] The ground powder was placed in a standard sieve set for oscillation screening. In this embodiment, the preset uniform particle size range was set to 100 to 150 mesh (approximately 100 μm to 150 μm). Powder particles within this mesh range were selected as test samples. The purpose of controlling particle size uniformity is to eliminate the influence of particle size differences on thermal conductivity, ensuring that all sample particles decompose synchronously when heated, thereby obtaining sharp and clear thermogravimetric peaks.

[0022] The selected sample powders underwent surface impurity removal treatment. The sample powders were ultrasonically cleaned in anhydrous ethanol for 3-5 minutes to remove any static dust, grease, or residual release agent from the preparation process that may have been adsorbed on the powder surface. After cleaning, the samples were placed in a vacuum drying oven and dried at 60°C for 4 hours to remove solvent residue, resulting in clean and homogeneous test samples.

[0023] It should also be noted that a preset flow rate of protective gas is continuously introduced into the heating chamber of the thermogravimetric analyzer to isolate external air and maintain a stable inert atmosphere. The specific execution process is as follows: Before starting the heating program, the gas path control system is activated, and high-purity nitrogen (N2, purity ≥99.999%) is selected as the protective gas. The system controls the mass flow meter (MFC) and sets the preset flow rate to 50 mL / min.

[0024] Throughout the entire experiment (i.e., the entire period from room temperature to 800℃), nitrogen gas was continuously purged into the heating chamber from the bottom of the thermogravimetric analyzer. The gas flow served two main purposes: first, to create a positive pressure environment within the heating chamber, preventing the infiltration of external ambient air and ensuring that the composite material degrades only under anaerobic pyrolysis conditions, thus avoiding interference from oxidation reactions in determining the material's intrinsic thermal stability; second, as a carrier gas, a stable gas flow ensured that the volatile products (flue gas) generated by the thermal decomposition of the sample were promptly and smoothly transported through connecting pipelines to the gas detection cell of the infrared spectrometer, preventing the products from condensing or remaining in the furnace, thereby ensuring that the infrared spectral data could reflect the material's current release state in real time.

[0025] S2. Based on the time difference between the peak occurrences of thermogravimetric data and infrared spectral data, determine the lag time of gas transport of pyrolysis products, and use this lag time to perform time-axis shift correction on the infrared spectral data to obtain time-aligned infrared spectral data. The specific steps are as follows: Determine the lag time of gas transport of pyrolysis products based on the time difference between the peak occurrences of thermogravimetric data and infrared spectral data. This includes: extracting a first characteristic curve reflecting the change in material decomposition rate from the thermogravimetric data, and identifying the first characteristic time point with the largest decomposition rate in the first characteristic curve; extracting a second characteristic curve reflecting the total intensity of gas release from the infrared spectral data, and identifying the gas in the second characteristic curve... The second characteristic time point with the greatest release intensity is identified; the time interval between the second characteristic time point and the first characteristic time point is calculated, and this time interval is determined as the pyrolysis product gas transport lag time; the infrared spectral data is then time-axis shifted and corrected using the pyrolysis product gas transport lag time. Specifically, this includes: obtaining the initial recording time point corresponding to each frame of the infrared spectral data; subtracting the pyrolysis product gas transport lag time from the initial recording time point to obtain the corrected actual reaction time point; and remapping the infrared spectral data to the actual reaction time point, so that each frame of the infrared spectral data is synchronized with the thermogravimetric data at the same physical reaction time point, thereby generating time-aligned infrared spectral data.

[0026] It should be noted that a first characteristic curve reflecting the change in material decomposition rate is extracted from the thermogravimetric data, and the first characteristic time point with the largest decomposition rate in the first characteristic curve is identified. In this embodiment, the original thermogravimetric data (TG) records the change in mass percentage over time. The system generates a derivative thermogravimetric curve (DTG curve) as the first characteristic curve by performing a first-order derivative of the original TG data with respect to time. Each point on the DTG curve represents the material decomposition rate at that moment. The system executes a peak-finding algorithm to search for the peak and trough positions with the smallest values ​​(representing the maximum weight loss rate, usually negative) on the DTG curve, and marks the time coordinates corresponding to the extreme points as the first characteristic time points. The point in time, in a physical sense, corresponds to the moment when the solid material undergoes the most violent decomposition reaction.

[0027] The system extracts a second characteristic curve reflecting the total intensity of gas release from infrared spectral data and identifies the second characteristic time point where the gas release intensity is highest. Since infrared spectral data is three-dimensional (time-wavenumber-absorbance), the system employs the Gram-Schmidt orthogonal reconstruction method (or full-band integration method) to process the spectral data to obtain a one-dimensional intensity curve. Specifically, the system calculates the vector difference or total integral area of ​​each frame of infrared spectrum relative to the background spectrum, generating a Gram-Schmidt reconstruction map (GS curve) that varies over time, which serves as the second characteristic curve. The peak value on the GS curve represents the highest detected total concentration of volatile gases at that moment. The system also uses a peak-finding algorithm to identify the position of the highest peak on the GS curve and marks the time coordinate corresponding to this extreme point as the second characteristic time point. The time point physically corresponds to the moment when a large amount of gaseous products flow into the infrared detection cell.

[0028] Calculate the time interval between the second characteristic time point and the first characteristic time point, and determine this time interval as the lag time for the gas transport of pyrolysis products. The calculation formula is as follows:

[0029] in, : Gas transport lag time of pyrolysis products, in minutes (min).

[0030] Reference range: usually between 0.1 min and 1.0 min, depending on the length of the pipeline and the carrier gas flow rate (e.g., 50-100 mL / min).

[0031] (Second characteristic time point): The time point at which the total intensity of gas release (Gram-Schmidt curve) in the infrared spectral data reaches its maximum value.

[0032] First characteristic time point: The time point at which the thermogravimetric curve (DTG) of the micro-merchant reaches its extreme value (i.e., the maximum decomposition rate) in the thermogravimetric data.

[0033] It should also be noted that the infrared spectral data is time-axis shifted forward using the pyrolysis product gas transport lag time to obtain time-aligned infrared spectral data. The specific steps are as follows: Obtain the initial recording time point corresponding to each frame of the infrared spectral data. Let the first frame be... The initial recording time point for the frame infrared spectrum is .

[0034] Subtract the pyrolysis product gas transport lag time from the initial recording time point. The corrected actual reaction time point was obtained. The correction formula is: .

[0035] Remapping infrared spectral data to actual reaction time points Through this mathematical transformation, what was originally recorded in The gas phase data at a given moment is logically shifted to the moment of its physical generation, thereby ensuring that each frame of the infrared spectral data is precisely synchronized with the thermogravimetric data at the same physical reaction time, eliminating system errors caused by pipeline transmission.

[0036] S3. Extract features from the thermogravimetric data and the time-aligned infrared spectral data respectively, construct solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors, and concatenate them in the time dimension to generate a fused feature matrix. The specific steps are as follows: Constructing the solid-phase thermogravimetric feature vector specifically includes: discretizing the thermogravimetric data at preset time intervals to obtain time-series sample points; extracting the mass value and the derivative value changing with time corresponding to each time-series sample point; extracting key statistical features of the initial decomposition temperature, maximum decomposition rate temperature, and residual mass at the final temperature from the thermogravimetric data; combining the mass value and the derivative value changing with time with the key statistical features to generate the solid-phase thermogravimetric feature vector; Constructing the gas-phase infrared feature vector specifically includes: determining the characteristic wavelength range in the infrared spectral data according to the preset target decomposition gas list and its corresponding infrared characteristic absorption bands; calculating the spectral response intensity index of each time sampling point within the characteristic wavelength range; arranging the spectral response intensity index corresponding to the characteristic wavelength range in time sequence to generate a multi-channel gas-phase infrared feature vector.

[0037] It should be noted that after time alignment is completed, the system performs structured feature extraction on the thermogravimetric data of the solid phase and the infrared spectral data of the gas phase, respectively.

[0038] 1. Constructing a solid-phase thermogravimetric feature vector: Discretize the thermogravimetric data at preset time intervals to obtain time-series sample points. The system resamples continuous thermogravimetric (TG) and derivative thermogravimetric (DTG) curves at fixed sampling intervals (e.g., 0.05 minutes) to obtain equally spaced time points. (in (where the total number of sampling points is ), thus converting a continuous analog signal into a discrete digital sequence.

[0039] Extract the quality value and its derivative over time for each time series sample point. For each sampling time... The system simultaneously reads two key physical quantities: percentage of remaining mass. (Directly from the TG curve) and instantaneous decomposition rate (Directly from DTG curves). These two quantities together describe the material's properties. The thermal decomposition state and its changing trend at any given time.

[0040] Simultaneously, key statistical features were extracted from the thermogravimetric data, including the initial decomposition temperature. Maximum decomposition rate temperature and the residual mass at final temperature The system automatically calculates and stores these three global scalar features from the overall thermogravimetric curve: the temperature at which weight loss reaches 5% is used as the starting point for thermal stability. The temperature corresponding to the peak point of the DTG curve is taken as the temperature of most severe decomposition. The percentage of remaining mass at the test endpoint temperature (e.g., 800℃) is used as the indicator of char-forming capacity. .

[0041] By combining the mass value and its time-varying derivative with key statistical features, a solid-phase thermogravimetric feature vector is generated. For each time step... The system constructs a feature vector that integrates dynamic temporal information and static global information. Its mathematical expression is:

[0042] time The remaining percentage of mass (mass value), reference range: 0% to 100%.

[0043] time The first derivative of mass with respect to time (decomposition rate / DTG value), reference range: usually -30% / min to 0% / min (negative values ​​indicate weightlessness).

[0044] Initial decomposition temperature (temperature at which 5% weight loss occurs), reference range: approximately 200℃ to 500℃.

[0045] Maximum decomposition rate temperature, reference range: approximately 300°C to 600°C.

[0046] The percentage of residual mass at the final temperature (e.g., 800°C), with a reference range of 0% to 60%.

[0047] In the vector, the first two elements and It changes over time, constituting the dynamic part of the characteristic; the last three elements For all samples of the same type These are all the same constants, constituting the static part of the feature. Through this combination of dynamic and static methods, the model can simultaneously perceive the overall thermal stability and char-forming properties of the material when analyzing local behavior at any given moment.

[0048] 2. Constructing Gas-Phase Infrared Feature Vectors: Based on a pre-defined list of target decomposition gases and their corresponding infrared characteristic absorption bands, the system determines the characteristic wavelength ranges in the infrared spectral data. The system includes a built-in knowledge base storing common flame-retardant related gaseous products. Standard infrared absorption peak position For each preset target gas (N types in total), the system locks the corresponding characteristic absorption band. .

[0049] Calculate the spectral response intensity index of each sampling point within the characteristic wavelength range. For each sampling time... and each target gas The system does not read the absorbance at a single wavenumber, but rather calculates the area under the absorbance curve across the entire characteristic waveband, using this area as the relative release intensity of the gas at that moment. The calculation formula is:

[0050] At any moment wave number is The absorbance value at the specified location, with a reference range of 0.0 to 2.0 (absorbance units).

[0051] No. The characteristic wavelength range of a gas, for example. 2300-2400 CO is 2000-2200 .

[0052] No. The spectral response intensity (integral area) of a gas represents the gas's performance in... The relative concentration at any given time.

[0053] in, Represents the moment , wave number The infrared absorbance at a certain point. This band integration method is more resistant to noise interference than single-point readings, and its integrated value is positively correlated with the gas concentration over a wide range, thus providing a more robust characterization of gas release.

[0054] The spectral response intensity indices corresponding to the characteristic wavelength ranges are arranged in time sequence to generate a multi-channel gas phase infrared feature vector. The N target gases are then analyzed at time... The release intensity indices are arranged in a predetermined order, thus obtaining the multidimensional gas phase characteristic vector at that moment. :

[0055] This vector constitutes a multi-channel time series, with each channel corresponding to the release process of a specific gas, fully recording the evolution information of gaseous products during the pyrolysis of the material.

[0056] 3. The system generates a fusion feature matrix at the same time. solid-phase eigenvectors With gas phase eigenvectors The feature vectors are concatenated along the feature dimension to form the final fused feature vector at that moment. The fused feature vectors from all M time points are then sorted chronologically. Stacking, we get a dimension of The fusion feature matrix.

[0057] S4. Perform channel-specific processing on the fused feature matrix, extracting solid mass change features and gas release features separately. Merge the extracted features and weight them according to their importance to generate comprehensive feature data. The specific steps are as follows: Channel-specific processing of the fused feature matrix is ​​achieved using a pre-built dual-channel long short-term memory network structure, including parallel solid-phase processing channels and gas-phase processing channels. The solid-phase processing channel is configured to perform time-series modeling of the solid-phase thermogravimetric feature vector, extracting the time-dependent features of solid-phase decomposition. The gas-phase processing channel is configured to perform time-series modeling of the gas-phase infrared feature vector, extracting gas-phase... The dynamic evolution characteristics of the released data are shown. The dual-channel long short-term memory network structure also includes a feature fusion layer, which splices and fuses the features output from the solid-phase processing channel and the gas-phase processing channel. Weighting based on correlation is performed by a feature weighting calculation unit, which performs the following operations: receiving fused feature data from the feature fusion layer; calculating the correlation weight of each dimension of the fused feature data to the prediction task according to a preset weight calculation strategy; and using the correlation weight to perform weighting operations on the fused feature data to enhance the contribution of highly correlated features to the final prediction result and suppress the interference of low-correlation features.

[0058] It should be noted that the fusion feature matrix is ​​processed by channel-specific processing to extract solid mass change features and gas release features separately. The specific implementation process is as follows: 1. Dual-channel network structure processing: The fusion feature matrix is ​​processed by channel-specific processing using a pre-set dual-channel long short-term memory network structure, which includes parallel solid phase processing channels and gas phase processing channels.

[0059] Solid-phase processing channel: Used for temporal modeling of solid-phase thermogravimetric feature vectors to extract time-dependent features of solid-phase decomposition. Specifically, this channel receives a temporal sequence composed of solid-phase feature vectors as input. The network learns and memorizes the long-term dynamic trends of material mass changes during heating through its internal recurrent neural units (such as LSTM units). For example, it can capture typical staged evolution patterns such as initial slow decomposition, followed by rapid and violent decomposition, and finally entering residual slow decomposition, or identify complex kinetic behaviors such as multi-stage decomposition plateaus. This channel outputs a set of high-order abstract features that can highly summarize the overall thermal decomposition process of the material.

[0060] The gas phase processing channel is used for time-series modeling of gas phase infrared feature vectors to extract the dynamic evolution characteristics of gas phase release. This channel receives a time-series sequence composed of gas phase feature vectors as input. The network focuses on analyzing the temporal order and intensity changes of different gas release peaks. For example, it can learn and identify the deep chemical reaction logic, such as whether the early release of flame-retardant gases effectively inhibits the formation of subsequent flammable hydrocarbons, or capture the synergistic and competitive relationships between different gas phase products. Finally, this channel outputs a set of high-order abstract features that reflect the flame-retardant mechanism and gas phase reaction pathway of the material.

[0061] 2. Feature Fusion Preparation: After feature extraction is completed in both channels, the dual-channel long short-term memory network structure also includes a feature fusion layer, which concatenates and fuses the features output from the solid-phase processing channel and the gas-phase processing channel. The system connects the high-order features output from the two channels along the feature dimension, preparing the data for subsequent unified weighted processing.

[0062] It should also be noted that the extracted features are fused and weighted according to relevance weights to generate comprehensive feature data. The specific execution steps are as follows: 1. Feature splicing and preliminary fusion: The dual-channel long short-term memory network structure also includes a feature fusion layer, which splices and fuses the features output from the solid-phase processing channel and the gas-phase processing channel. The system performs a vector splicing operation, connecting the high-order time-series feature vectors output from the solid-phase processing channel and the gas-phase processing channel end-to-end along the feature dimension, combining them into a long vector containing all physicochemical information, i.e., fused feature data. At this point, although the data encompasses information from all dimensions, the primary and secondary features are not yet distinguished, and their status is equal.

[0063] 2. Intelligent weight calculation, which assigns weights based on relevance, is performed by a feature weighting calculation unit. The unit first receives fused feature data from the feature fusion layer. Then, according to a preset weight calculation strategy, it calculates the relevance weights of each dimension of the fused feature data to the prediction task. In this embodiment, the preset weight calculation strategy employs a self-attention mechanism. The system automatically evaluates the importance of each feature dimension in the fused feature data for the final predicted flame retardant performance through its internal network layer. The system generates a numerical weight between 0 and 1 for each feature dimension. For example, a trained model might automatically identify that the feature dimension representing the final temperature char rate is crucial for flame retardant judgment and thus assigns a higher weight value; while the feature dimension representing trace moisture release at low temperatures has minimal impact on flame retardant performance and is assigned a lower weight value.

[0064] 3. Weighted Enhancement and Noise Suppression: The system utilizes correlation weights to perform weighted operations on the fused feature data, enhancing the contribution of highly correlated features to the final prediction result and suppressing the interference of low-correlation features. The system performs element-wise multiplication, multiplying the original value of each feature dimension by its corresponding weight value.

[0065] For highly correlated features: due to the multiplication by a large weight value, the numerical signal is preserved or even amplified in subsequent calculations, thus taking a dominant position in the prediction results (i.e., enhancing the contribution).

[0066] For low-correlation features: due to the multiplication by a very small weight value, the numerical signal is significantly reduced, thereby reducing its impact on the prediction results (i.e., suppressing interference).

[0067] The processed data is the comprehensive feature data, which retains and highlights only the flame-retardant properties of the material, providing input for the next step of prediction.

[0068] S5. Perform numerical mapping and classification calculation on the comprehensive feature data to obtain the predicted flame retardant index and flame retardant level classification results of the composite material samples. The specific operation steps are as follows: To obtain the predicted flame retardant index, the following steps are taken: Construct a linear mapping channel for numerical regression, input the comprehensive feature data into the linear mapping channel, and map the high-dimensional comprehensive feature data into a one-dimensional continuous value through a fully connected operation; perform inverse normalization on the one-dimensional continuous value to convert it into a specific value that conforms to the flame retardant index measurement, and determine the specific value as the predicted flame retardant index value of the composite material samples.

[0069] The specific steps to obtain the flame retardant rating classification result include: constructing a classification mapping channel for category discrimination, inputting comprehensive feature data into the classification mapping channel, and calculating the probability score of the comprehensive feature data belonging to the preset flame retardant rating category; normalizing the probability score, selecting the category with the highest probability score as the final flame retardant rating, and determining the flame retardant rating as the flame retardant rating classification result of the composite material sample.

[0070] It should be noted that the process includes two parallel processing branches: Branch 1: Flame retardant index numerical prediction (regression task). A linear mapping channel is constructed for numerical regression. The comprehensive feature data is input into the linear mapping channel, and a fully connected operation maps the high-dimensional comprehensive feature data into a one-dimensional continuous value. The system uses a fully connected layer to perform linear weighted summation and dimensionality reduction operations on the comprehensive feature data, compressing it into a single numerical output. The value is typically processed by an activation function (such as Sigmoid or Tanh) and falls within a normalized range (e.g., between 0 and 1). The one-dimensional continuous value is then de-normalized to convert it into a specific value that conforms to the flame retardant index metric, and this specific value is determined as the predicted flame retardant index value for the composite material sample. Since the model output is a relative value, the system needs to restore it in conjunction with a preset physical range. In this embodiment, the flame retardant index specifically refers to the limiting oxygen index (LOI). The system maps the normalized value output by the model back to the actual physical range of the LOI (e.g., 20.0% to 60.0%). For example, if the model output is 0.6, after inverse normalization, the system will finally output a predicted limiting oxygen index of 44.0%.

[0071] Branch 2: Flame Retardant Rating Classification Determination (Classification Task) A classification mapping channel is constructed for category discrimination. Comprehensive feature data is input into the classification mapping channel, and the probability score of the comprehensive feature data belonging to a preset flame retardant rating category is calculated. The system uses another fully connected layer in conjunction with a Softmax classifier to map the comprehensive feature data into a probability vector. The dimension of this vector corresponds to the preset number of flame retardant ratings. This embodiment, based on the UL-94 vertical burning standard, presets four rating categories: V-0, V-1, V-2, and No Rating (NR). The system calculates the probability score of the sample belonging to these four ratings, for example, obtaining a probability vector [0.85, 0.10, 0.03, 0.02]. The probability scores are normalized, and the category with the highest probability score is selected as the final flame retardant rating, which is then determined as the flame retardant rating classification result for the composite material sample. The system compares the values ​​in the probability vector and identifies the category index (V-0) corresponding to the maximum value (0.85). Based on this, the system determines the flame retardancy rating of the composite material sample to be V-0 and outputs it to the user as the final classification result. Through the aforementioned dual-channel output, this invention achieves simultaneous prediction of quantitative assessment (LOI value) and qualitative classification (UL-94 rating) of the flame retardancy performance of composite materials. Example 2

[0072] To further clarify the present invention, the technical solution of the present invention will be described in more detail below, taking the research and evaluation of a novel flame-retardant composite material as an example.

[0073] In this embodiment, the core objective of evaluating the flame retardant properties (limiting oxygen index LOI and UL-94 rating) of a novel phosphorus-nitrogen synergistic flame-retardant epoxy resin composite material is to rapidly and accurately assess the material's flame retardant properties, thereby accelerating the product iteration cycle.

[0074] S1. Acquisition of raw physicochemical data: A batch of samples to be tested was prepared according to standards. The samples were ground and sieved into uniform powder of 100-150 μm, placed in a thermogravimetric-infrared spectrometer, and heated from room temperature to 800℃ under a nitrogen atmosphere of 50 mL / min at a heating rate of 20℃ / min. During this process, the system automatically acquired complete mass change curves (thermogravimetric data) and gas release spectra (infrared spectral data).

[0075] S2. Achieve precise synchronization of solid-phase and gas-phase data: Raw data shows a time difference between material decomposition and gas release. The system activates the timing alignment module to automatically process the data: By analyzing thermogravimetric data, it identifies the moment of most intense material decomposition (DTG peak) at 15.5 minutes.

[0076] By analyzing infrared data, the moment with the highest total gas release intensity (GS peak) was identified at 15.8 minutes.

[0077] The system calculated the time difference between the two to be 0.3 minutes and determined it as the gas transport lag time. The system shifted the time axis of the entire infrared spectrum forward by 0.3 minutes to ensure that the mass loss at each moment in the subsequent analysis corresponded to the gas produced at that moment, achieving precise synchronization of the physical reaction process.

[0078] S3. Transforming experimental curves into structured features: For solid-phase data, the system not only extracts the remaining mass and decomposition rate at each moment, but also calculates the material's global thermal stability indices, such as the initial decomposition temperature. Maximum decomposition rate temperature The carbon residue at 800℃ was 28%. These data were combined into a multidimensional solid-phase eigenvector.

[0079] For gas phase data: the system focuses on key aspects based on its built-in chemical knowledge base. (Complete combustion products) The release intensity of key gases, such as the nitrogen-based flame retardant product and the POC characteristic peak (representing the phosphorus-based flame retardant product), is measured. At each moment, the relative concentration of these gases is quantified and combined into a gas-phase characteristic vector.

[0080] Finally, the solid and gas phase feature vectors are spliced ​​together in the time dimension to form a complete and structured material pyrolysis archive (i.e., a fused feature matrix).

[0081] S4. Deeply mine data correlations and focus on key information: Input this material pyrolysis file into the system's dual-channel LSTM network for in-depth analysis. The solid phase channel within the network analyzed the relationship between char residue and decomposition rate, while the gas phase channel analyzed... The timing and intensity of the POC peak were analyzed. The system's built-in feature-weighted calculation unit (self-attention mechanism) found that, for predicting the performance of this phosphorus-nitrogen synergistic flame-retardant material, the 800℃ char residue rate, a condensed phase flame-retardant index, and... The gas-phase quenching indicator, synergistic release with POC, is the most important of all features. Therefore, the model assigns extremely high weights to the feature dimensions representing these two phenomena, while correspondingly reducing the weights of secondary information such as early moisture evaporation.

[0082] S5. Output Quantitative Prediction Report: The essential features after weighted processing are sent to the prediction output module, and the system generates a prediction report almost instantly: Numerical Prediction Section: Through linear mapping and inverse normalization calculation, the predicted value of the limiting oxygen index (LOI) of the material is 42.5%.

[0083] Flame retardancy rating classification: The classifier determined the probability of this sample belonging to each flame retardancy rating as follows: V-0 (95%), V-1 (3%), V-2 (1%), and no rating (1%). The system ultimately determined its flame retardancy rating to be UL-94V-0.

[0084] To verify the accuracy of the prediction results, the samples were subjected to traditional LOI testing and UL-94 vertical burning testing. The experimental results showed that the measured LOI value of the material was 43.0%, and the measured flame retardancy rating was V-0. The prediction results and the measured results were highly consistent, proving that the method of this invention can complete high-precision flame retardancy performance evaluation in a short time, greatly improving the efficiency of new material research and development.

[0085] In summary, this invention achieves temporal alignment of solid-phase and gas-phase data by collecting thermogravimetric and infrared spectral data of composite material samples and calculating the gas transport lag time of pyrolysis products to perform time-axis forward correction on the infrared spectral data. It extracts and constructs solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors respectively, generating a fused feature matrix. A dual-channel long short-term memory network is used to extract solid-phase mass change and gas-phase release characteristics, and weighted fusion is performed based on correlation weights. Finally, through numerical mapping and classification calculations, the predicted flame retardant index and flame retardant rating are simultaneously output. This invention achieves in-depth multi-dimensional feature mining of the entire pyrolysis process of composite materials, enabling accurate and rapid prediction of the limiting oxygen index and flame retardant rating of materials, providing an effective digital evaluation method for the research and screening of new flame-retardant materials.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data, characterized in that, Includes the following steps: S1. According to the preset heating rate and gas environment, thermogravimetric and infrared spectral data of composite material samples are collected using a thermogravimetric and infrared spectral device; S2. Based on the time difference between the peak occurrences of the thermogravimetric data and the infrared spectral data, the gas transport lag time of the pyrolysis products is determined, and the infrared spectral data is corrected by shifting the time axis forward using the gas transport lag time of the pyrolysis products to obtain time-aligned infrared spectral data; S3. Features are extracted from the thermogravimetric data and the time-aligned infrared spectral data respectively to construct solid-phase thermogravimetric feature vectors and gas-phase infrared feature vectors, and they are spliced ​​together in the time dimension to generate a fused feature matrix; S4. Perform channel-specific processing on the fused feature matrix to extract solid mass change features and gas release features respectively. Merge the extracted features and weight them according to the feature importance weights to generate comprehensive feature data. S5. Perform numerical mapping and classification calculation on the comprehensive feature data to obtain the predicted flame retardant index and flame retardant level classification results of the composite material sample.

2. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, Before step S1, the method further includes: processing the sample to a preset uniform particle size range and removing surface impurities; the data collection according to the preset gas environment specifically refers to: during the data collection process, continuously introducing a preset flow rate of protective gas into the heating chamber of the thermogravimetric analyzer to isolate external air and maintain a stable inert atmosphere environment.

3. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S2, the gas transport lag time of pyrolysis products is determined based on the time difference between the peak occurrences of thermogravimetric data and infrared spectral data. Specifically, this includes: extracting a first characteristic curve reflecting the change in the material decomposition rate from the thermogravimetric data, and identifying a first characteristic time point with the largest decomposition rate in the first characteristic curve; extracting a second characteristic curve reflecting the total intensity of gas release from the infrared spectral data, and identifying a second characteristic time point with the largest gas release intensity in the second characteristic curve; calculating the time interval between the second characteristic time point and the first characteristic time point, and determining the time interval as the gas transport lag time of pyrolysis products.

4. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S2, the time-axis shift correction of the infrared spectral data using the pyrolysis product gas transport lag time specifically includes: obtaining the initial recording time point corresponding to each frame of the infrared spectral data; subtracting the pyrolysis product gas transport lag time from the initial recording time point to obtain the corrected actual reaction time point; and remapping the infrared spectral data to the actual reaction time point so that each frame of the infrared spectral data is synchronized with the thermogravimetric data at the same physical reaction time point, thereby generating time-aligned infrared spectral data.

5. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S3, the construction of the solid-phase thermogravimetric feature vector specifically includes: discretizing the thermogravimetric data at preset time intervals to obtain time series sample points; extracting the mass value and the derivative value that changes with time corresponding to each time series sample point; extracting key statistical features of the initial decomposition temperature, the maximum decomposition rate temperature, and the residual mass at the final temperature from the thermogravimetric data; and combining the mass value and the derivative value that changes with time with the key statistical features to generate the solid-phase thermogravimetric feature vector.

6. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S3, the construction of the gas phase infrared feature vector specifically includes: determining the characteristic wavelength range in the infrared spectral data according to the preset target decomposition gas list and its corresponding infrared characteristic absorption bands, calculating the spectral response intensity index of each time sampling point in the characteristic wavelength range, arranging the spectral response intensity index corresponding to the characteristic wavelength range in time sequence, and generating a multi-channel gas phase infrared feature vector.

7. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S4, the channel-specific processing of the fused feature matrix is ​​achieved using a pre-defined dual-channel long short-term memory network structure, which includes parallel solid-phase processing channels and gas-phase processing channels. The solid-phase processing channel is configured to perform time-series modeling on the solid-phase thermogravimetric feature vectors and extract the time-dependent features of solid-phase decomposition. The gas-phase processing channel is configured to perform time-series modeling on the gas-phase infrared feature vectors and extract the dynamic evolution features of gas-phase release. The dual-channel long short-term memory network structure also includes a feature fusion layer, which splices and fuses the features output from the solid-phase processing channel and the gas-phase processing channel.

8. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S4, weighting based on relevance is performed by a feature weighting calculation unit, which performs the following operations: receiving fused feature data from the feature fusion layer; calculating the relevance weights of each dimension of the fused feature data to the prediction task according to a preset weighting calculation strategy; and using the relevance weights to perform weighting operations on the fused feature data to enhance the contribution of highly relevant features to the final prediction result and suppress the interference of low-relevance features.

9. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S5, obtaining the predicted flame retardant index specifically includes: constructing a linear mapping channel for numerical regression, inputting the comprehensive feature data into the linear mapping channel, mapping the high-dimensional comprehensive feature data into a one-dimensional continuous value through a fully connected operation; performing inverse normalization on the one-dimensional continuous value to convert it into a specific value that conforms to the flame retardant index measurement, and determining the specific value as the predicted flame retardant index value of the composite material sample.

10. The method for predicting the flame retardant properties and thermal stability of composite materials based on the fusion of thermogravimetric analysis and infrared spectroscopy data as described in claim 1, characterized in that, In step S5, obtaining the flame retardant rating classification result specifically includes: constructing a classification mapping channel for category discrimination, inputting the comprehensive feature data into the classification mapping channel, and calculating the probability score of the comprehensive feature data belonging to the preset flame retardant rating category; normalizing the probability score, selecting the category with the highest probability score as the final flame retardant rating, and determining the flame retardant rating as the flame retardant rating classification result of the composite material sample.

Images