A rapid equivalent detection device and method for thermal decomposition characteristics of nitrate compounds

Abstract

The application discloses a kind of nitrate compound thermal decomposition characteristics fast equivalent detection device and method.Equipment includes closed heating table, carrier gas supply module, infrared flow cell and infrared detection and data processing module.Method is: sample is placed in closed heating table, under preset thermal condition, using carrier gas to be released gas of it into infrared flow cell, gas phase infrared time series spectrum and thermal decomposition time are synchronously collected;Spectrum is pretreated, selects multiple gas phase characteristic peaks, extracts the response time and response rate of each peak, and is spliced into initial feature vector;With measured stability time as supervision label, train multilayer perception machine neural network model;Prediction time only needs to collect the gas phase infrared data of sample to be measured within preset duration, and model can be output equivalent thermal decomposition time by inputting.The application can use early gas release dynamic information to quickly predict the thermal stability of nitrate compound, reduce the dependence on artificial end interpretation, and shorten the detection waiting time.

Description

Technical Field

[0001] This invention belongs to the field of spectroscopic quantitative analysis and intelligent modeling technology, specifically relating to a rapid equivalent detection device and method for the thermal decomposition characteristics of nitrate ester compounds. Background Technology

[0002] Nitrate esters are an important class of nitro compounds widely used in energetic systems such as propellants, explosives, propellants, and pyrotechnics. Typical examples include nitrated ethylene glycol, isoamyl nitrate, and pentaerythritol tetranitrate (PETN). Because nitrate ester molecules contain the nitrate ester (–ONO2) functional group, they are prone to slow decomposition under heating conditions, releasing volatile decomposition products such as nitrogen oxides. Therefore, their thermal stability directly affects the storage safety, transportation safety, service reliability, and service life of the products. Accurate evaluation of the thermal stability of nitrate ester compounds has always been a crucial aspect of quality control and safety assessment of energetic materials.

[0003] Currently, the stability test is a commonly used method for evaluating the thermal stability of nitrate ester compounds. This method typically involves isothermal heating of the sample under specified temperature conditions, causing a color change in indicator paper through the volatile decomposition products. The time it takes for the indicator paper to reach the specified color change is used as the evaluation index of thermal stability. Due to its mature testing method, simple operation, and good repeatability, the stability test has been widely used for detecting the thermal stability of energetic materials containing nitrate esters.

[0004] However, the stability test mainly relies on the color change of the test strip to determine the endpoint, and can only obtain the final thermal stability evaluation result. The detection process has a low degree of automation and a long detection cycle, which makes it difficult to meet the needs of rapid evaluation and process analysis of thermal stability.

[0005] In recent years, infrared spectroscopy has been widely used in chemical analysis and process monitoring due to its advantages such as fast detection speed, non-destructive nature, and ability to reflect the vibrational characteristics of gas molecules. For nitrate esters, the volatile decomposition products released during thermal decomposition exhibit corresponding characteristic absorptions in the infrared spectrum. Therefore, using infrared spectroscopy to detect gases released during thermal decomposition has certain application potential. However, existing thermal stability detection devices are generally not designed for online infrared detection in the gas phase, making it difficult to achieve continuous and stable extraction and online detection of volatile decomposition products, and also making it difficult to fully utilize the continuous spectral information generated during thermal decomposition for thermal stability evaluation.

[0006] Therefore, there is an urgent need to provide a method and apparatus suitable for the thermal stability detection of energetic compounds such as nitrate esters, so as to realize the online detection of volatile decomposition products during thermal decomposition and improve the automation level and detection efficiency of thermal stability evaluation. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a device and method for predicting the gas phase infrared equivalent thermal decomposition time of nitrate ester compounds. This invention addresses the problems of long detection cycles, reliance on manual endpoint interpretation, difficulty in utilizing early gas phase dynamic information, and lack of suitable online gas phase infrared detection devices in the current evaluation of the thermal stability of nitrate ester compounds.

[0008] This invention uses a sealed heating stage to controllably heat the sample and uses a carrier gas to transport the volatile gases released from the sample to an infrared flow cell for online detection. It extracts the response time and response rate characteristics of multiple gas phase characteristic peaks and then outputs the equivalent thermal decomposition time of the sample based on a multilayer perceptron neural network, thereby achieving rapid and intelligent assessment of the thermal stability of nitrate ester compounds.

[0009] To achieve the above objectives, the present invention proposes the following technical solution:

[0010] This invention first provides a rapid equivalent detection method for the thermal decomposition characteristics of nitrate ester compounds, which includes the following steps:

[0011] 1) Under preset thermal conditions, heating treatment, stabilization test and gas phase infrared time-series acquisition were carried out simultaneously on the nitrate ester compound calibration samples. The sample was placed in a sealed heating stage, and the gas released by the sample was introduced into the infrared flow cell under the action of the carrier gas for gas phase infrared time-series spectral detection.

[0012] 2) Preprocess the acquired gas-phase infrared time-series spectra;

[0013] 3) Select multiple gas phase characteristic peaks from the preprocessed gas phase infrared time series spectrum and extract the response time and response rate characteristics of each gas phase characteristic peak;

[0014] 4) Concatenate the response time and response rate features corresponding to each gas phase characteristic peak to form an initial feature vector;

[0015] 5) The synchronously measured settling time is used as a supervision label to train the multilayer perceptron neural network model;

[0016] 6) For the same nitrate ester compound, the sample is heated and subjected to gas phase infrared time-series acquisition under the same or equivalent preset thermal conditions as in step 1). The acquisition continues until the preset duration is reached and then terminated. The ratio of the preset duration to the stabilization time of the calibration sample is 0.1 to 0.6. The initial feature vector of the sample is obtained according to steps 2) to 4), and it is input into the trained multilayer perceptron neural network model to output the equivalent thermal decomposition time of the sample.

[0017] According to a preferred embodiment of the present invention, in step 1), gas phase infrared acquisition begins when the sample reaches the preset thermal conditions and the stabilization timing begins; during the training phase, gas phase infrared acquisition continues until the sample reaches the stabilization endpoint, which is the moment when the volatile nitrogen-containing decomposition products generated by the thermal decomposition of nitrate esters cause the indicator paper to reach the specified color change.

[0018] According to a preferred embodiment of the present invention, the preset thermal conditions are controlled thermal conditions that are consistent with or equivalent to the standard heating conditions for stabilization testing after calibration. Further, the preset thermal conditions include a target heating temperature, a heating rate, and a constant temperature control accuracy. When the temperature at the sample reaches the target heating temperature and stabilizes within the preset temperature control error range, stabilization timing and gas phase infrared timing data acquisition begin.

[0019] According to a preferred embodiment of the present invention, the apparatus used in step 1) to simultaneously perform heating treatment, stabilization testing, and gas-phase infrared time-series acquisition includes a sealed heating stage, a carrier gas supply module, a gas delivery pipeline, and an infrared flow cell. The sealed heating stage provides a stable thermal environment for the sample and forms a relatively enclosed space for the release of volatile gases; the carrier gas is used to extract the gaseous components released from the sample from the sealed heating stage and deliver them to the infrared flow cell; the infrared flow cell is used to achieve online infrared detection of the gaseous components.

[0020] According to a preferred embodiment of the present invention, the preprocessing in step 2) includes one or more of smoothing filtering, baseline correction, wavenumber correction, noise suppression, and normalization.

[0021] According to a preferred embodiment of the present invention, the gas-phase characteristic peaks in step 3) are multiple characteristic peak regions in the gas-phase infrared time-series spectrum. The gas-phase characteristic peaks refer to infrared absorption peaks or peak regions generated in the infrared flow cell by volatile gas-phase components released by the heated nitrate ester compound sample after subtraction of the carrier gas blank spectrum and the device blank spectrum, which appear or intensify with heating time and whose peak areas can be stably extracted.

[0022] According to a preferred embodiment of the present invention, the gas-phase characteristic peaks are selected according to the following rules: First, a blank spectrum of the carrier gas and a blank spectrum of the device without nitrate ester compounds are collected, and the gas-phase infrared time-series spectrum is subjected to blank subtraction; then, in the gas-phase infrared time-series spectrum after blank subtraction, absorption peaks or absorption peak regions that appear with the sample heating time or whose peak areas continuously increase are selected as candidate gas-phase characteristic peaks. The candidate gas-phase characteristic peaks should meet one or more of the following conditions: their peak area is higher than a preset multiple of the blank noise level; their peak center position remains stable in repeated experiments; their peak area can be extracted at multiple consecutive acquisition times; their peak region does not experience detection saturation; and their peak area changes over time to reflect the release process of the sample's gas-phase decomposition products.

[0023] Preferably, the gas-phase characteristic peaks include infrared absorption peaks or absorption peak regions corresponding to nitrogen-containing oxides, carbon dioxide, carbon monoxide, or other volatile decomposition products released during the thermal decomposition of nitrate esters. The gas-phase characteristic peaks can be confirmed by standard gas spectra, infrared spectral libraries, blank control experiments, stability analysis of repeated experiments, or correlation analysis between peak area and measured stability time. More preferably, the number M of the gas-phase characteristic peaks is an integer greater than or equal to 2, preferably 3 to 10, and more preferably 3 to 6.

[0024] Furthermore, the response time mentioned in step 3) is the earliest time when the peak area of ​​the characteristic peak first reaches the preset response threshold, and the response rate is used to represent the average growth rate of the characteristic peak after reaching the response threshold.

[0025] Let the peak area of ​​the m-th gas-phase characteristic peak at time t be... ,in M represents the total number of characteristic peaks.

[0026] According to a preferred embodiment of the present invention, the response time of the m-th gas-phase characteristic peak is defined as:

[0027]

[0028] in, For the m-th gas-phase characteristic peak to reach the preset response threshold The earliest time.

[0029] According to a preferred embodiment of the present invention, the response rate of the m-th gas-phase characteristic peak is defined as:

[0030]

[0031] in, The time from when the m-th gas-phase characteristic peak reaches the response threshold to the preset time period (the end of the preset early time window). The average growth rate between them.

[0032] According to a preferred embodiment of the present invention, the initial feature vector in step 4) is represented as:

[0033]

[0034] in, The initial feature vector is formed by splicing together the response times and response rates of multiple gas phase characteristic peaks.

[0035] According to a preferred embodiment of the present invention, the multilayer perceptron neural network model in step 5) includes an input layer, one or more hidden layers, and an output layer; the input layer receives an initial feature vector. The output layer outputs the predicted equivalent thermal decomposition time of the sample. .

[0036] According to a preferred embodiment of the present invention, the measured stabilization time is used during the training phase. As a supervision label, to minimize the predicted equivalent thermal decomposition time Compared with the measured stability time The error between them is used as the training target.

[0037] According to a preferred embodiment of the present invention, in step 6), for the sample to be tested, gas phase infrared acquisition continues for a preset duration. Then terminate; the preset duration (also referred to as the early time window in this invention) is then used. Based on the measured stabilization time of the training or calibration samples Predetermined, and smaller than Preferably, in the training or calibration samples, The value is 0.10–0.60; more preferably 0.20–0.50; and even more preferably 0.25–0.40. For the sample to be tested, The stability time can be preset based on the average, median, quantile of the stability time in the training set of similar samples or the measured stability time of standard samples.

[0038] In a preferred embodiment, the preset duration It also meets the following conditions: Previously, the peak areas of at least two gas phase characteristic peaks reached the corresponding preset response threshold, or no less than 50% of the total number of selected gas phase characteristic peaks reached the corresponding preset response threshold.

[0039] The present invention also provides an equivalent thermal decomposition time prediction device for implementing the aforementioned method, the device comprising a sealed heating stage, a carrier gas supply module, an infrared flow cell, a thermal decomposition time measurement module, and an infrared detection and data processing module;

[0040] The sealed heating stage is used to control the heating of nitrate ester compound samples and to form a relatively sealed sample release space.

[0041] The carrier gas supply module is used to provide carrier gas and export the gas released from the sample;

[0042] The infrared flow cell is used to provide a fixed optical path and detection environment for gas phase components;

[0043] The infrared detection and data processing unit includes infrared spectrometer hardware and a multilayer perceptron neural network model. The infrared spectrometer hardware generates infrared light and acquires raw gas-phase infrared time-series spectral data. The multilayer perceptron neural network model is used to preprocess the raw gas-phase infrared time-series spectral data, extract the response time and response rate features of gas-phase characteristic peaks, form an initial feature vector, and train the multilayer perceptron neural network model using the stabilization time as a supervision label. In the prediction application stage, the infrared detection and data processing unit extracts the initial feature vector of the sample to be tested and outputs the equivalent thermal decomposition time of the sample to be tested based on the trained multilayer perceptron neural network model.

[0044] Through the above technical solution, the present invention has the following significant advantages:

[0045] (1) Improved the stability of online acquisition of gases released by nitrate ester compounds upon heating.

[0046] This invention employs a detection structure combining a sealed heating stage, a carrier gas supply module, a gas delivery pipeline, and an infrared flow cell. This allows the volatile decomposition gases released from nitrate ester samples under controlled thermal conditions to be continuously extracted by the carrier gas and fed into the infrared flow cell for online detection. Based on this structure, the sample heating space, gas extraction path, and infrared detection position are relatively fixed, reducing fluctuations caused by gas diffusion, retention, and manual sampling in open environments. This provides a data foundation for the subsequent extraction of stable gas-phase infrared time-series characteristics.

[0047] (2) The synchronous correlation between the stabilization test process and the gas phase infrared time-series detection was realized.

[0048] This invention simultaneously acquires gas-phase infrared time-series spectra when the sample reaches the preset thermal conditions and begins stabilization timing, thus providing a unified time reference for the gas-phase infrared data and thermal decomposition time. Based on this synchronous acquisition method, a correspondence can be established between the appearance time and growth process of gas-phase components in the infrared spectrum and the endpoint of the stabilization test, avoiding time deviations caused by separate spectral data acquisition or post-test comparison, and improving the consistency and comparability of model training samples.

[0049] (3) It can use multi-peak dynamic response information to characterize the early thermal decomposition behavior of nitrate esters.

[0050] This invention does not rely solely on peak intensity or a single characteristic peak at a single moment as the basis for judgment. Instead, it selects multiple gas-phase characteristic peaks from the gas-phase infrared time-series spectrum and extracts response time and response rate features for each. The response time characterizes when the corresponding gas-phase component first reaches a detectable level, while the response rate characterizes the release and growth rate of the gas-phase component in the early stages. These two types of features can reflect the release dynamics of volatile gases during the thermal decomposition of nitrate esters from both the time of appearance and the growth rate, thus better reflecting differences in sample thermal stability compared to single-peak static features.

[0051] (4) It can predict the equivalent thermal decomposition time before the actual stable endpoint is reached.

[0052] This invention utilizes the measured stability time obtained from a complete stability test as a supervisory label during the training phase, and establishes a mapping relationship between the early gas-phase infrared multi-peak response characteristics and the final thermal decomposition time. During the prediction phase, only gas-phase infrared data of the test sample within a preset time period (the ratio of the preset time to the stability time of the calibration sample is 0.1–0.6) needs to be collected, and the equivalent thermal decomposition time can be output by inputting it into the trained model. Therefore, this invention can obtain thermal stability evaluation results in advance without waiting for the test strip to reach the traditional stability endpoint, thereby shortening the detection cycle. This effect is particularly suitable for scenarios requiring batch screening or rapid comparison of the thermal stability of different samples.

[0053] (5) It reduces the impact of manual endpoint interpretation on the test results.

[0054] Traditional stabilization tests typically rely on color changes in test strips to determine the endpoint, which suffers from issues such as differences in human observation, unclear color change boundaries, and recording time lag. This invention automatically extracts multi-peak response features through a gas-phase infrared time-series spectral acquisition and data processing module, and outputs the equivalent thermal decomposition time based on a trained multilayer perceptron neural network model. This reduces the impact of manual endpoint interpretation on the results and improves the objectivity and repeatability of the detection process.

[0055] (6) Enhanced the nonlinear modeling capability of complex gas phase decomposition processes.

[0056] During the thermal decomposition of nitrate esters, the formation time, formation rate, and infrared response intensity of different volatile gaseous components may vary. A single linear criterion is insufficient to fully describe the relationship between the early gaseous response and the final thermal decomposition time. This invention concatenates the response times and response rates of multiple gaseous characteristic peaks to form a feature vector, which is then input into a multilayer perceptron neural network model for training. This allows for the learning of the nonlinear mapping relationship between multi-peak dynamic characteristics and equivalent thermal decomposition time, thereby improving the utilization of early gaseous information.

[0057] (7) The apparatus and method have good modular implementation conditions.

[0058] The device of this invention consists of a sealed heating stage, a carrier gas supply module, a gas delivery pipeline, an infrared flow cell, a thermal decomposition time measurement module, and a data processing module. Each module has clearly defined functional boundaries and is capable of performing controlled sample heating, gas extraction, online spectral acquisition, thermal decomposition time recording, and data modeling and prediction functions, respectively. This modular structure facilitates the addition of a gas-phase infrared detection module and a data processing module to existing stability testing procedures. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of a detection device consisting of a sealed heating stage, a carrier gas delivery system, and an infrared flow cell, according to one embodiment of the present invention.

[0060] Explanation of reference numerals in the attached drawings: 1-Carrier gas supply module; 2-Flow regulator; 3-Sealed heating platform; 4-Sealed cover or temperature control component; 5-Infrared flow cell; 6-Infrared detection interface; 7-Infrared detection and data processing unit. Detailed Implementation

[0061] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0062] Example: Prediction of equivalent thermal decomposition time of nitrated ethylene glycol based on a closed heating stage and a gas-phase infrared flow cell

[0063] like Figure 1 As shown, this embodiment provides a device for predicting the equivalent thermal decomposition time of nitrated ethylene glycol based on the multi-peak response characteristics of gas-phase infrared radiation. The device in this embodiment includes a sealed heating stage 3, a carrier gas supply module 1, a gas pipeline, an infrared flow cell 5, a thermal decomposition time measurement module (used only in the training data acquisition stage), and an infrared detection and data processing unit 7.

[0064] The sealed heating stage 3 is used to place the nitrated ethylene glycol sample and provide preset thermal conditions. A sealing cover or temperature control component 4 is installed on its top. The carrier gas supply module 1 continuously supplies carrier gas to the sealed heating stage 3. A gas delivery pipeline is used to transport the carrier gas, and a flow regulator 2 is installed on the pipeline to adjust the carrier gas flow rate as needed. Gas released from the sample under preset thermal conditions is introduced into the infrared flow cell 5 through the gas delivery pipeline under the action of the carrier gas. The infrared flow cell 5 provides a fixed optical path and detection environment for the introduced gas phase components and is equipped with an infrared detection interface 6. The thermal decomposition time measurement module is used to synchronously record the thermal decomposition time corresponding to the color change of the test strip when acquiring training data. The thermal decomposition time measurement module is not required during the application phase. The infrared detection and data processing unit 7 includes infrared spectrometer hardware and a multilayer perceptron neural network model. The infrared spectrometer hardware generates infrared light and detects the light intensity after passing through the infrared flow cell, thereby generating raw gas phase infrared time-series spectral data. The multilayer perceptron neural network model is used to preprocess the gas phase infrared time series spectrum, extract the response time and response rate features of the gas phase characteristic peaks, and train the multilayer perceptron neural network model based on the extracted features. In the application stage, the infrared detection and data processing unit 7 extracts the features of the sample to be tested and outputs the equivalent thermal decomposition time of the sample to be tested based on the trained multilayer perceptron neural network model.

[0065] The core of this invention lies in: utilizing a sealed heating stage and a carrier gas extraction structure to simultaneously acquire the gas phase infrared time-series spectrum of the gas released from the sample during the stabilization test; by extracting the response time and response rate of multiple gas phase characteristic peaks, a dynamic feature vector capable of characterizing the onset time and growth rate of gas release is constructed; then, using the synchronously measured thermal decomposition time as a supervisory label, a multilayer perceptron neural network model is trained to establish a mapping relationship between early gas phase dynamic characteristics and the final thermal decomposition time. In the prediction stage, only gas phase infrared data of the sample under test within a preset time period (in this invention, the preset time period is also called the preset early time window, and the ratio of the preset time period to the stabilization time of the calibration sample is 0.1 to 0.6) needs to be acquired. The trained model can then quickly output the equivalent thermal decomposition time, thereby significantly shortening the detection cycle while ensuring prediction accuracy. The following detailed description of the method in this embodiment, combined with specific operational steps, further illustrates this approach.

[0066] In this embodiment, during the training phase, the nitrated ethylene glycol sample is placed in a sealed heating stage, and the moment when the sample reaches the preset thermal conditions and the stabilization time begins is defined as t=0. From this moment, gas phase infrared time-series acquisition and stabilization tests are carried out simultaneously until the test strip changes color; the corresponding endpoint time is recorded as t=0. ,in This represents the actual thermal decomposition time of the sample.

[0067] In this embodiment, the preset thermal conditions include a target heating temperature, a heating rate, and a temperature stability criterion. The target heating temperature can be determined according to the standard conditions of the stability test or the company's internal control testing conditions, for example, 60–90°C, preferably 65–85°C; the heating rate is 0.5–10°C / min, preferably 1–5°C / min; when the temperature near the sample position in the sealed heating stage reaches the target temperature and stabilizes within the range of ±0.5–±2°C, this moment is defined as t=0, and the stability timing and gas phase infrared timing acquisition begin simultaneously. If the standard stability test temperature is used... The target heating temperature is preferably... Temperature control deviation should not exceed ±0.5~±2℃.

[0068] In this embodiment, the acquired gas phase infrared time-series spectrum is preprocessed, preferably including smoothing filtering, baseline correction, and normalization. Then, multiple gas phase characteristic peaks are selected from the preprocessed gas phase infrared time-series spectrum. Let the peak area of ​​the m-th gas phase characteristic peak at time t be... ,in .

[0069] In this embodiment, M=5 gas-phase characteristic peak regions are selected as the objects of gas-phase infrared dynamic feature extraction. The gas-phase characteristic peak regions may include the exemplary peak regions shown in Table 1 below. Specific peak regions can be corrected according to the infrared flow cell resolution, actual spectrum peak position drift, standard gas spectra, or infrared spectrum libraries.

[0070] Table 1

[0071]

[0072] The aforementioned peak regions are not limited to a single peak position, but are determined by the integration interval based on the resolution of the infrared flow cell, the actual peak position drift of the gas components, and the baseline correction results. For the m-th gas-phase characteristic peak region, its peak area... It can be obtained by integrating the absorbance curve within the peak region.

[0073] The selection criteria for gas-phase characteristic peaks include the following conditions: First, the peak region has no obvious response in the carrier gas blank spectrum and the device blank spectrum, or the background response after blank subtraction is lower than the preset noise threshold; Second, the peak region appears or intensifies over time after the nitrated ethylene glycol sample is heated, reflecting the release process of volatile decomposition gases from the sample; Third, the peak center position of the peak region is relatively stable in repeated experiments, preferably with a peak position shift not exceeding ±5 cm. -1 Fourth, the peak area of ​​the peak region can be stably extracted over multiple consecutive acquisition times, and there is no detection saturation; fifth, the response time or response rate of the peak region is correlated with the measured settling time, or has high feature importance in the trained model.

[0074] For each gas-phase characteristic peak, its response time is first defined as:

[0075]

[0076] in, The preset response threshold is the m-th gas phase characteristic peak.

[0077] Then, the peak is defined as the point from reaching the response threshold to the end of a preset early time window. The response rates between them are:

[0078]

[0079] Therefore, for the M gas-phase characteristic peaks, an initial eigenvector can be formed:

[0080]

[0081] In this embodiment, the initial feature vector is... The input is a multilayer perceptron neural network model. This model includes an input layer, one or more hidden layers, and an output layer. The output layer outputs the predicted equivalent thermal decomposition time of the sample. .

[0082] During the training phase, complete gas-phase infrared time-series spectra were simultaneously acquired for multiple groups of nitrated ethylene glycol samples with different thermal stability, and their measured stability times were measured. Based on actual measured stability time As supervisory labels, the multilayer perceptron neural network model is trained. During the training phase, the measured thermal decomposition time is used as the training metric. As a supervision label, to minimize the predicted equivalent thermal decomposition time Compared with the measured stability time The error between the samples is used as the training objective. Specifically, the training process is as follows: the initial feature vector of each sample group is used as the model input, the corresponding measured settling time is used as the output objective, the mean squared error is used as the loss function, and an adaptive moment estimation optimizer is used to update the parameters. The termination condition for model training is: the validation set loss no longer decreases for 10 consecutive training epochs, or the preset maximum number of training epochs (e.g., 500 epochs) is reached. The above training strategies (including optimizer selection, loss function setting, early stopping strategy, etc.) are all conventional techniques in this field, and those skilled in the art can make appropriate adjustments according to the actual dataset size. After training is completed, the model can be used to predict the equivalent thermal decomposition time of the test samples.

[0083] The trained multilayer perceptron neural network model can be used to make equivalent predictions of the thermal decomposition time of the nitrated ethylene glycol sample. The prediction stage can be carried out using the following process:

[0084] First, in the prediction phase, the gas phase infrared time-series spectrum of the sample to be tested is also collected starting from t=0, but only up to the end of the preset early time window. The preset duration Based on the measured stabilization time of similar training or calibration samples Predetermined. Preferably, To measure the stabilization time of training or calibration samples. It is 20% to 50%, more preferably 25% to 40%, and even more preferably 30% to 35%.

[0085] In one specific implementation, take ,in This represents the median measured stabilization time of similar calibration or training samples. Therefore, the prediction phase only requires collecting gas-phase infrared data equivalent to approximately 30% of the reference stabilization test duration, theoretically reducing detection waiting time by about 70% compared to waiting for the actual stabilization endpoint. When the response time is too short, some gas-phase characteristic peaks have not yet reached the response threshold, resulting in insufficient extraction of response time and response rate features, which can easily reduce prediction accuracy; when While more gas phase infrared information can be obtained over a longer period, the time-saving effect compared to traditional stabilization tests is reduced.

[0086] During the prediction phase, it is only necessary to collect the sample to be tested for a preset time. By extracting the response time and response rate features of multiple gas phase characteristic peaks from the gas phase infrared data, and inputting these features into the trained model, the equivalent thermal decomposition time of the sample under test can be output. .

[0087] Using the above method, this embodiment can predict the final equivalent thermal decomposition time by utilizing the early gas phase release dynamic information of the sample before it reaches the true stability endpoint, thereby achieving rapid and intelligent assessment of the thermal stability of nitrated ethylene glycol.

[0088] The accuracy of the prediction method of the present invention is verified through specific embodiments below.

[0089] In this embodiment, 12 groups of nitrated ethylene glycol samples with varying thermal stability were selected as verification objects. Each sample underwent stability testing and gas-phase infrared time-series spectroscopy acquisition under the same preset thermal conditions. During the training phase, data was collected for each sample from t=0 to the actual stability endpoint. The complete gas-phase infrared time-series spectrum was obtained, and the measured settling time was recorded simultaneously. During the prediction phase, only gas-phase infrared data within a preset time period was used for prediction. The median measured settling time of similar training samples was used as the reference thermal decomposition time. In this embodiment ,Pick .

[0090] In this embodiment, five gas-phase characteristic peak regions are selected from the preprocessed gas-phase infrared time-series spectrum, and the response time and response rate features of each peak region are extracted to form an initial feature vector f. Eight sets of samples are used as training samples, and four sets of samples are used as validation samples. The initial feature vector of the training samples is used as the model input, and the measured stabilization time is used as the model input. The multilayer perceptron neural network model is trained using these supervisory labels. After training, the gas-phase infrared multi-peak response features extracted from the validation samples within a preset time window (preset early time window) are input into the model to obtain the predicted equivalent thermal decomposition time. The verification results are shown in Table 2 below.

[0091] Table 2

[0092]

[0093] As shown in the table above, the equivalent thermal decomposition time predicted by this invention and the measured stabilization time are in good agreement in the verification samples. The mean absolute error (MAE), root mean square error (RMSE), and mean relative error (MAPE) are calculated using the following formulas:

[0094]

[0095]

[0096]

[0097] Where N is the number of validation samples. Let be the measured stability time of the i-th verification sample. Let be the predicted equivalent thermal decomposition time of the i-th verification sample. Based on the above verification data, the calculated values ​​are: MAE = 1.45 min, RMSE = 1.47 min, and MAPE = 2.82%. Therefore, under the set sample and test conditions, this invention can accurately predict the equivalent thermal decomposition time of the sample by collecting only early gas phase infrared data.

[0098] Furthermore, since the preset early acquisition window for the prediction stage in this embodiment is 15.0 min, compared to traditional stability tests that require waiting for the sample to reach the true stability endpoint, this invention can shorten the detection waiting time. Taking sample S4 as an example, its true thermal decomposition time is 61.5 min. Traditional methods require waiting approximately 61.5 min to obtain the endpoint result; however, using the method of this invention, only about 15.0 min of early gas phase infrared data needs to be acquired to obtain the predicted equivalent thermal decomposition time of 63.2 min, with a relative error of 2.76%, theoretically reducing the detection waiting time by about 75.6%. This result shows that this invention can maintain good predictive ability for the evaluation results of the thermal stability of nitrated ethylene glycol while shortening the detection time.

[0099] Furthermore, the verification results from different samples show that samples with shorter measured stability times generally have earlier gas-phase characteristic peak response times and relatively higher response rates; samples with longer measured stability times have relatively delayed gas-phase characteristic peak response times and relatively lower response rates. These patterns indicate that the gas-phase characteristic peak response times and response rates extracted in this invention can reflect the early thermal decomposition gas release behavior of nitrated ethylene glycol samples and have a corresponding relationship with the final thermal decomposition time.

[0100] In summary, this embodiment demonstrates that, under the set sample and test conditions, the present invention can predict the equivalent thermal decomposition time using the early gas-phase infrared multi-peak response characteristics, and can obtain thermal stability evaluation results without waiting for the actual stable endpoint.

Claims

1. A rapid equivalent detection method for the thermal decomposition characteristics of nitrate ester compounds, characterized in that, Includes the following steps: 1) Under preset thermal conditions, heating treatment, stabilization test and gas phase infrared time-series acquisition were carried out simultaneously on the nitrate ester compound calibration samples. The sample was placed in a sealed heating stage, and the gas released by the sample was introduced into the infrared flow cell under the action of the carrier gas for gas phase infrared time-series spectral detection. 2) Preprocess the acquired gas-phase infrared time-series spectra; 3) Select multiple gas phase characteristic peaks from the preprocessed gas phase infrared time series spectrum and extract the response time and response rate characteristics of each gas phase characteristic peak; 4) Concatenate the response time and response rate features corresponding to each gas phase characteristic peak to form an initial feature vector; 5) The synchronously measured settling time is used as a supervision label to train the multilayer perceptron neural network model; 6) For the same nitrate ester compound, the sample is heated and subjected to gas phase infrared time-series acquisition under the same or equivalent preset thermal conditions as in step 1). The acquisition continues until the preset duration is reached and then terminated. The ratio of the preset duration to the stabilization time of the calibration sample is 0.1 to 0.

6. The initial feature vector of the sample is obtained according to steps 2) to 4), and it is input into the trained multilayer perceptron neural network model to output the equivalent thermal decomposition time of the sample.

2. The method according to claim 1, characterized in that, In step 1), gas phase infrared data acquisition begins when the sample reaches the preset thermal conditions and the stabilization timer starts, and ends when the sample reaches the stabilization endpoint. The preset thermal conditions are controlled thermal conditions that are consistent with the standard heating conditions for stabilization testing or are equivalent to the standard heating conditions for stabilization testing after calibration.

3. The method according to claim 1, characterized in that, The apparatus used in step 1) to simultaneously perform heating treatment, stabilization test and gas phase infrared time-series acquisition includes a sealed heating stage, a carrier gas supply module, a gas pipeline and an infrared flow cell. The sealed heating stage is used to place nitrate ester compound samples and provide preset thermal conditions; The carrier gas supply module is used to supply carrier gas to the sealed heating platform; The gas pipeline is used to transport the gas released from the sealed heating table to the infrared flow cell. The infrared flow cell is used to perform infrared spectral detection on the introduced gas phase components.

4. The method according to claim 1, characterized in that, The preprocessing described in step 2) includes one or more of the following: smoothing filtering, baseline correction, wavenumber correction, noise suppression, and normalization.

5. The method according to claim 1, characterized in that, Let the peak area of ​​the m-th gas-phase characteristic peak at time t be... ,in M represents the total number of gas-phase characteristic peaks; the response time of the m-th gas-phase characteristic peak is defined as: ； in, For the m-th gas-phase characteristic peak to reach the preset response threshold The earliest time.

6. The method according to claim 5, characterized in that, The response rate of the m-th gas-phase characteristic peak is defined as: ； in, The time from the m-th gas phase characteristic peak reaching the response threshold to the preset time. The average growth rate between them.

7. The method according to claim 1, characterized in that, The multilayer perceptron neural network model described in step 5) includes an input layer, one or more hidden layers, and an output layer; the input layer receives an initial feature vector, and the output layer outputs the predicted equivalent thermal decomposition time of the sample.

8. The method according to claim 1, characterized in that, In step 5), the measured stabilization time is used as the supervision label, and the training objective is to minimize the error between the predicted equivalent thermal decomposition time and the measured stabilization time. The training continues until the multilayer perceptron neural network model converges.

9. The method according to claim 1, characterized in that, The stabilization time mentioned in step 5) is the time from the start of the stabilization timing to the stabilization endpoint. The stabilization endpoint is the time when the indicator paper reaches the specified color change due to the volatile nitrogen-containing decomposition products generated by the thermal decomposition of nitrate ester compounds.

10. A device for predicting the gas-phase infrared equivalent thermal decomposition time of nitrate ester compounds, used for implementing the method according to any one of claims 1-9, characterized in that, It includes a sealed heating stage, a carrier gas supply module, an infrared circulation pool, and an infrared detection and data processing module; The sealed heating stage is used to control the heating of nitrate ester compound samples and to form a relatively sealed sample release space. The carrier gas supply module is used to provide carrier gas and export the gas released from the sample; The infrared flow cell is used to provide a fixed optical path and detection environment for gas phase components; The infrared detection and data processing unit includes infrared spectrometer hardware and a multilayer perceptron neural network model. The infrared spectrometer hardware generates infrared light and acquires raw gas-phase infrared time-series spectral data. The multilayer perceptron neural network model is used to preprocess the raw gas-phase infrared time-series spectral data, extract the response time and response rate features of gas-phase characteristic peaks, form an initial feature vector, and train the multilayer perceptron neural network model using the stabilization time as a supervision label. In the prediction application stage, the infrared detection and data processing unit extracts the initial feature vector of the sample to be tested and outputs the equivalent thermal decomposition time of the sample to be tested based on the trained multilayer perceptron neural network model.

Images