Journal of Aeronautics, Astronautics and Aviation, Vol. 56, No. 4 pp.

791 – 799 (2024) 791

DOI: https://doi.org/10.6125/JoAAA.202409_56(4).01

Prediction of Regression Rate of HTPB Solid

Fuel-A Machine Learning Approach

Saravanan Meenakshisundaram 1, Krishnakumari Aharamuthu 1, Alisha Biju 1,

Kaveri Vucha 1, Yash Pal 1 *, Sri Nithya Mahottamananda 2
1
Department of Aeronautical Engineering, Hindustan Institute of Technology and Science, Chennai-603103
2
Department of Aerospace Engineering, BS Abdur Rahman Crescent Institute of Science and Technology,
Chennai-600048

ABSTRACT

The prediction of regression rates in solid fuels plays a critical role in

optimizing the performance of hybrid rocket propulsion systems. This paper
presents a machine learning (ML) approach to predict the regression rate of
HTPB solid fuel with various additives, aiming to reduce both time and costs
associated with experimental investigations. For this study, it utilizes data
from previous research where HTPB was combined with Nano Aluminium,
Polytetrafluoroethylene (PTFE), and Viton to create four distinct samples: (i)
HTPB, (ii) HTPB-nAl, (iii) HTPB-nAl-PTFE, and (iv) HTPB-nAl-Viton.
Additionally, a MATLAB analytical model was developed to generate data
alongside experimental work. Three ML models - Random Forest (RF),
Decision Tree (DT), and Multi-Linear Regression (MLR) - were developed
and trained using a dataset derived from previous experimental work. The
dataset underwent exploratory data analysis to select relevant features and
refine the dataset. The trained ML models were evaluated based on metrics
such as accuracy, recall, F1 score, and R2 score. The RF model demonstrated
superior performance with an R2 score exceeding 0.99, outperforming the
other ML models. The RF model's predictions are closely aligned with the
analytical model values, indicating its robustness and accuracy. This study
signifies the potential of machine learning in accurately predicting regression
rates of solid fuel, contributing to the optimum motor design and performance.

Keywords: Solid propellant, HTPB, regression rate, Machine Learning,

Random Forest, Decision tree, Multi linear regression

I. INTRODUCTION design [5-7]. Recent advancements in machine learning

(ML) techniques offer a promising avenue for enhancing
Hybrid rocket propulsion systems have garnered our ability to predict the regression/burn rate of solid
significant interest due to their promising combination of fuel/propellant in hybrid/solid rocket systems [8-14]. By
safety, efficiency, and versatility [1,2]. Among the key leveraging large datasets encompassing various fuel
components influencing the performance of hybrid rockets, formulations, environmental conditions, and combustion
the regression rate of the solid fuel plays a crucial role. The chamber geometries, ML algorithms can discern complex
regression rate, representing the rate at which the solid fuel relationships and patterns that traditional empirical models
surface recedes under combustion, directly impacts thrust might overlook. Furthermore, ML models can adapt and
and overall engine efficiency [3,4]. Understanding and improve over time as they are exposed to more data,
accurately predicting regression rate parameter is essential providing a dynamic and adaptable framework for
for optimizing hybrid rocket performance and motor regression rate prediction.

Manuscript received, April 23, 2024, final revision, July 4, 2024

* To whom correspondence should be addressed, E-mail: [email protected]
792 Saravanan Meenakshisundaram Krishnakumari Aharamuthu Alisha Biju
Kaveri Vucha Yash Pal Sri Nithya Mahottamananda

Several recent studies have demonstrated the efficacy model was presented for predicting the dynamic
of ML approaches in predicting regression/burn rates in behaviours and steady-state performance of solid
solid fuel/propellant for hybrid/soild rockets [10,12,13,15- propellant combustion which improved the understanding
21]. For instance, research by Tang et al. [10] utilized and prediction of complex combustion behaviours in solid
support vector regression (SVR) to accurately predict propellant systems [12]. Rivera et al.[14] applied ML
burning rates based on parameters such as fuel algorithms to predict the critical heat flux for ignition of
composition, oxidizer type, and chamber pressure. solid fuels. The study compiled a dataset from various
Similarly, Klinger et al. [13] employed Machine learning experiments and evaluated the performance of ML models
models have been developed to predict the burning rates in predicting CHFI. The models were able to adapt to
of solid rocket propellants based on composition and different scenarios and served as a useful tool for
microstructural features, offering potential for tailoring predicting CHFI in solid fuel ignition. Based on the
burn rate properties and providing initial estimates for new information gathered from the above literature, it is
propellant formulations. understood that ML techniques have been employed for
A cluster analysis method was performed for the prediction of various parameters associated with solid
aluminium agglomeration in solid propellant which propellants, including burning rate, combustion efficiency,
examined the aluminium agglomeration phenomena, for solid propellant breakage, factors affecting burning rate,
optimizing combustion efficiency [18]. Application of and the additive agglomeration.
machine learning in the design process, aiming to enhance This study aims to contribute to the growing body of
the efficiency and effectiveness of solid propellant grain literature by presenting a comprehensive analysis of
configurations were introduced [20]. The current state of regression rate prediction in HTPB solid fuels using a
machine learning applications in the field of combustion machine learning approach. Specifically, this paper
which provides insights into the evolving principles and proposes constructing a machine-learning model to predict
showcasing the progress made in leveraging machine the regression rate of HTPB solid fuel with different
learning techniques were understood [19]. Surina et al. [15] additives. The primary objective is to reduce both the time
performed the measurement of hybrid rocket solid fuel and costs associated with experimental testing, thereby
regression rate for a slab burner using deep learning enhancing the efficiency of the research and development
techniques, which offer potential improvements in process in the field of solid propellants.
monitoring and optimizing fuel regression rates. ML
models such as neural networks, support vector regression, II. METHODOLOGY
genetic programming, and Gaussian process regression
were used to predict the solid propellant breakage In our previous study, we collected experimental data
efficiency of cavitation water jets which had the potential on the regression rate of HTPB-based solid fuels using an
to enhance the efficiency and precision of solid propellant opposed flow burner.
breakage predictions in industrial applications [22]. A
dynamic model and a deep neural network-based surrogate

Figure 1 Flow chart showing the employed methodology (1) Data is collection (2) Dataset analysis for features
selections (3) ML modelling and training (4) regression rate prediction.
 Prediction of Regression Rate of HTPB Solid Fuel-A Machine Learning Approach 793

This data forms part of the dataset for the current get more data sets on regression rate, additional 𝑃𝑖 values
work [23]. Further, the data augmentation has been done were added to calculate it. Since the 𝑃𝑓 values were
through an analytical model. Utilizing this dataset, we absent for the newly generated 𝑃𝑖 values, a
employed a machine learning approach to predict the dimensionless number was introduced based on the
regression rate. available data, which is (𝑃𝑖 -𝑃𝑓 )/ 𝑃𝑖 (Figure 2).
This integration of experimental dataset and
analytical data set allowed for the creation of a robust
predictive model, facilitating accurate regression rate
estimations for HTPB-based solid fuels. A detailed process
for prediction of regression rate using ML is presented in
Figure 1. Following data collection, thorough analysis is
conducted to select pertinent features. In this study, linear
regression, decision trees, and multi linear regression
algorithms are used to develop the ML models. These ML
models are trained, tested and validated using the pre-
processed dataset. Finally, the validated model is deployed
to predict regression rates, with predictions compared
against actual data to assess accuracy and reliability.

III. ANALYTICAL MODEL

Analytical models employing mathematical

equations and statistical techniques are utilized to predict
the regression rate of solid fuel in hybrid rocket systems, Figure 2 Graphical representation of dimensionless
with a focus on its dependency on factors such as oxidizer number
mass flux, oxidizer injection pressure, fuel density, and
composition. Through iterative refinement and validation Using interpolation, 𝑃𝑓 values were estimated for all
using experimental data, these models provide a structured 𝑃𝑖 values within the pressure range. The parameters
framework for understanding the dynamics of solid fuel obtained through interpolation were then utilized in a
regression and supporting design decision-making series of theoretical formulae (Equations 1-4) in analytical
processes [10,15,24,25]. In the present work, utilizing an model to calculate the regression rate.
experimental dataset obtained from previous studies [23],
(𝑃𝑖 −𝑃𝑓 )×𝑣
with four distinct samples, they are HTPB (sample 1), 𝑚0 𝑥 = (1)
𝑅×𝑇
HTPB-nAl (sample 2), HTPB-nAl-PTFE (sample 3),
HTPB-nAl-Viton (sample 4) which was tested under 𝑚0 𝑥
gaseous oxygen oxidizer at pressure range between 2.38 𝑚̇0 𝑥 = (2)
𝑡𝑏
MPa -7.16 MPa. The final oxidiser settling-chamber
pressure (𝑃𝑓 ) and regression rate (r) were constructed G0 x =
ṁ0 x
(3)
An
comprising approximately 2000 pressure points. Initially,
there were only 16 sets of experimental data with
r = 𝑎[𝐺0 𝑥 𝑛 ] (4)
parameters such as initial oxidiser settling- chamber
pressure (𝑃𝑖 ), final oxidiser settling-chamber pressure
Finally, we trained the analytical model in MATLAB
(𝑃𝑓 ), volume of settling chamber (V), gas constant (R),
to utilize this information and further additional data.
temperature (T), burning time (tb), opposed-flow burner Table 1 presents a sample dataset from the analytical
nozzle area (An), pre-factor (a), and regression rate factor model, showing the outcomes of the model's predictions
(n) which were derived from regression rate power law. To and calculations.

Table 1 Sample dataset prepared using the analytical model

Initial
Final Burn Nozzle Regressi
pressure, T Pre- Regression
Sample pressure, V (m3) R time, Area, on rate
𝑷𝒊 (K) factor a factor n
𝑷𝒇 (MPa) tb(s) An (m2) (mm/s)
(MPa)
HTPB 2.38 1.35 0.0136 259.8 298 5 0.000113 0.019 0.86 0.135
HTPB-nAl 2.4 1.36 0.0136 259.8 298 5 0.000113 0.016 0.79 0.172
HTPB-nAl-PTFE 2.42 1.37 0.0136 259.8 298 5 0.000113 0.05 0.68 1.011
HTPB-nAl-Viton 2.44 1.38 0.0136 259.8 298 5 0.000113 0.002 1.3 0.001
794 Saravanan Meenakshisundaram Krishnakumari Aharamuthu Alisha Biju
Kaveri Vucha Yash Pal Sri Nithya Mahottamananda

IV. DATASET PREPARATION 11 columns. Subsequently, Exploratory Data Analysis

(EDA) has been conducted on this regression rate data to
The dataset has been compiled by merging data acquire comprehensive insights. The heatmap presented in
generated from the analytical model using MATLAB with Figure 3 illustrates the correlation among all the features
experimental data. The dataset consists of 1916 rows and within the dataset.

Figure 3 Heat map showing correlation between solid fuel features

Figure 4 Heat map showing correlation between solid fuel features (after feature selection)
 Prediction of Regression Rate of HTPB Solid Fuel-A Machine Learning Approach 795

Correlation values depicted on the heatmap range robust positive correlation, indicating a proportional
from -1 to +1, reflecting the strength and direction of the increment in one feature associated with changes in other
relationships between the features. During correlation features. Conversely, some features demonstrate a
analysis, it was observed that several features, such as negative correlation, implying an inversely proportional
volume of settling chamber (V), gas constant (R), relationship, where an increase in one feature corresponds
temperature (T), burn time (tb), and nozzle area (An), to a decrease in another feature.
maintained constant values and did not significantly In Table 2, a sample dataset generated after
contribute to predicting regression rate. Consequently, exploratory data analysis is presented. Subsequently, the
these constant values were systematically eliminated final dataset underwent refinement in preparation for
through feature selection technique. The resulting heatmap, training the machine learning model, ensuring the
displayed in Figure 4, represents the dataset after feature optimization of input variables for accurate predictive
selection, which now comprises 1916 rows and 5 columns. modelling.
From Figure 4, it is evident that the features exhibit a

Table 2 Sample Dataset prepared using Exploratory Data Analysis

Initial settling chamber Regression Regression rate,

Propellant Prefactor, a
pressure, (MPa) factor, n r (mm/s)
HTPB 2.38 0.019 0.86 0.135
HTPB-nAl 2.4 0.016 0.79 0.172
HTPB-nAl-PTFE 2.42 0.050 0.68 1.011
HTPB-nAl-Viton 2.44 0.002 1.30 0.001

V. ML MODEL AND TRAINING 1 𝑁

(6)
𝑀𝑆𝐸 = √ ∑𝑖=1(𝑦𝑖′ − 𝑦𝑖 )2
𝑁
Building upon prior research in the field, we
developed ML models aimed at predicting the regression 1 𝑁
𝑀𝐴𝐸 = ∑𝑖=1|𝑦𝑖′ − 𝑦𝑖 | (7)
rate of solid fuel. Various techniques including random 𝑁

forest, decision tree, and multi-linear regression were

Where ( 𝑦𝑖 ) represents the original regression rate
utilized. These models were trained on a compiled dataset
values in the dataset calculated by QC/K-J equation, (𝑦𝑖′ )
to understand the intricate relationships between solid fuel
compositions and their impact on regression rates. This is the predicted value by an ML model, (𝑦̅ ) is the average
study provides a comprehensive analysis of the accuracy, dataset value, and (N) is the sample size [24].
adaptability, and predictive capabilities of each model,
highlighting their efficacy in solid fuel formulation VI. EVALUATION OF ML MODELS
development. Specifically, ML models were trained to
differentiate the correlation between initial settling ML models used to predict regression rates have been
chamber pressure and regression rate of sampled fuels. successfully developed through the implementation of
The dataset comprising 2000 pressure points was divided Decision Tree (DT), Random Forest (RF), and multi-linear
into training and test sets at a 70:30 ratio. To determine the regression (MLR) algorithms. To evaluate the trained
optimal parameters, a hyperparameter optimization models' robustness and performance, accuracy, recall and
algorithm was employed during model training. All F1 score were calculated for both the training and test sets.
training procedures were conducted using Scikit-learn The prediction accuracy of each model for the specific
within the Anaconda Python virtual environment. property was assessed on an independent test set. Table 3
In assessing model accuracy, several metrics were demonstrates that all the models display linear correlations
considered. The sum squared regression quantifies the sum greater than 0.84.
of squared residuals, while the total sum of squares
measures the squared distances of data from the mean. Table 3 Metrices of the ML Models (DT, RF) for
Additionally, the linear correlation coefficient (R2), root Regression rate
mean squared error (RMSE), and mean absolute error
(MAE) were evaluated to provide a comprehensive Random
Metrics Decision Tree
understanding of model performance. Forest
Accuracy 0.84635 0.84635
𝑁 2
∑ (𝑦𝑖′ −𝑦𝑖 ) Recall 0.84635 0.84635
𝑅2 = 1 − 𝑖=1 (5)
𝑁
∑𝑖=1(𝑦−𝑦̅)2 F1 Score 0.80699 0.80699
796 Saravanan Meenakshisundaram Krishnakumari Aharamuthu Alisha Biju
Kaveri Vucha Yash Pal Sri Nithya Mahottamananda

To evaluate the trained RF model's robustness and above 0.99 as shown in Table 4. The RF model showcases
performance 𝑅2 is calculated for all ML models. All the exceptional performance, outperforming all other machine
ML models for regression rates have 𝑅2 values above learning models. For instance, Daniel's RF model achieved
0.96 as shown in Table 4. It is determined that the RF an impressive R2 value of 0.94 according to the literature
model is best among all the three models with 𝑅2 value [13] .

Table 4 R2 Score of the ML Models for Regression rate

Metrics Decision Tree Random Forest Multi Linear Regression

𝑹𝟐 Score 0.99 0.99 0.96

Figure 5 illustrates the linear correlations of RF kg/m2s to 80.3 kg/m2s, falls within the range of 0.1 mm/s
model predictions for each sample with oxidizer mass flux. to 0.35 mm/s. Furthermore, in the authors' previous work
It is noted that the average regression rate obtained from [23], Sample 1 exhibited an average regression rate
both the analytical model and ML models for sample 1, ranging from 0.1 mm/s to 0.336 mm/s when subjected to
when tested under an oxidizer mass flux range of 29 an oxidizer mass flux range of 29 kg/m2s to 80.3 kg/m2s.

Figure 5 Linear correlations of RF model predictions of regression rate for a) HTPB; b) HTPB-nAl; c) HTPB-
nAl-PTFE; d) HTPB-nAl-viton

The linear correlations of RF model predictions of results closely match the analytical model values.
each sample with oxidizer settling chamber pressures can Consequently, the RF model is utilized to forecast the
be seen in Figure 6. The alignment of data points with the regression rate of the remaining data points.
linear fitting lines indicates that the test set prediction
 Prediction of Regression Rate of HTPB Solid Fuel-A Machine Learning Approach 797

Figure 6 Linear correlations of RF model predictions of each sample with Initial settling chamber pressure for
a) HTPB; b) HTPB-nAl; c) HTPB-nAl-PTFE; d) HTPB-nAl-viton

VII. CONCLUSIONS REFERENCES

In this study, we utilized ML to predict the regression [1] Galfetti L, Nasuti F, Pastrone D, Russo AM, “An
rates of HTPB solid fuel loaded with additives and Italian Network to Improve Hybrid Rocket
fluoropolymers. By employing three predictive ML Performance: Strategy and Results,” Acta
models—Random Forest, Decision Tree, and Multi-Linear Astronautica, Vol. 96, 2014, pp. 246-260.
Regression algorithms—we evaluated the solid fuel https://doi.org/10.1016/j.actaastro.2013.11.036
regression rate. The dataset for training the ML models [2] Pal Y, Mahottamananda SN, Palateerdham SK,
was generated from previous experimental works by one Subha S, Ingenito A, “Review on the Regression
of the authors, supplemented with data from an analytical Rate-Improvement Techniques and Mechanical
model developed in MATLAB. The results demonstrate Performance of Hybrid Rocket Fuels,”
that, while all three ML models offer improved predictive FirePhysChem, Vol. 1, No. 4, 2021, pp. 272-282.
capabilities, the Random Forest model stands out with the https://doi.org/10.1016/j.fpc.2021.11.016
highest accuracy. The Random Forest model achieved a [3] Karabeyoglu A, Zilliac G, Cantwell BJ, DeZilwa S,
coefficient of determination (R²) of 0.99, along with a Castellucci P, “Scale-Up Tests of High Regression
mean absolute error (MAE) of 0.01, indicating its superior Rate Paraffin-Based Hybrid Rocket Fuels,” Journal
performance in predicting the regression rates. The of Propulsion and Power, Vol. 20, No. 6, 2004, pp.
Decision Tree and Multi-Linear Regression models also 1037-1045.
performed well, but with slightly lower R² values and DOI:10.2514/1.3340
higher MAEs. Key parameters predicted include the [4] Sabri MEE bin, Azami MH, Abdullah NA, Nordin
regression rate of the solid fuel, the impact of various NH, “Investigation of Regression Rate End-Burning
additives, and the influence of fluoropolymers on the Typed Hybrid Rocket Motor,” Journal of
combustion characteristics. This study highlights the Aeronautics, Astronautics and Aviation, Vol. 55, No.
potential of machine learning in reducing experimental 3S, 2023, pp. 485-494.
time and costs while providing accurate predictions of https://doi.org/10.6125/JoAAA.202309_55(3S).08
solid fuel behavior.
798 Saravanan Meenakshisundaram Krishnakumari Aharamuthu Alisha Biju
Kaveri Vucha Yash Pal Sri Nithya Mahottamananda

[5] Thomas JC, Paravan C, Stahl JM, Tykol AJ, 103968.

Rodriguez FA, Galfetti L, Petersen EL, https://doi.org/10.1016/j.firesaf.2023.103968
“Experimental Evaluation of HTPB/Paraffin Fuel [15] Surina G, Georgalis G, Aphale SS, Patra A,
Blends for Hybrid Rocket Applications,” DesJardin PE, “Measurement of Hybrid Rocket
Combustion and Flame, Vol. 229, 2021, p. 111386. Solid Fuel Regression Rate for a Slab Burner Using
https://doi.org/10.1016/j.combustflame.2021.02.03 Deep Learning,” Acta Astronautica, Vol. 190, 2022,
2 pp. 160-175.
[6] Razak NA, Azami MH, “Investigation on https://doi.org/10.1016/j.actaastro.2021.09.04
Regression Rate Behavior of Polylactic Acid [16] Choi JB, Nguyen PCH, Sen O, Udaykumar HS,
Manufactured by Fused Deposition Modelling for Baek S, “Artificial Intelligence Approaches for
Hybrid Rocket Motor,” Journal of Aeronautics, Energetic Materials by Design: State of the Art,
Astronautics and Aviation, Vol. 56, No. 1S, 2024, pp. Challenges, and Future Directions,” Propellants,
157-165. Explosives, Pyrotechnics, Vol. 48, No. 4, 2023, p.
https://doi.org/10.6125/JoAAA.202403_56(1S).08 e202200276.
[7] Ahmad MT, Jagannathan A, Abidin R, Zhahir A, https://doi.org/10.1002/prep.202200276
Saniman MNF, “Static Hot-Fire Testing of a Green [17] Wan'e W, Zuoming Z, “Calculation for Primary
Hybrid Rocket Engine,” Journal of Aeronautics, Combustion Characteristics of Boron-Based Fuel-
Astronautics and Aviation, Vol. 56, No. 1S, 2024, pp. Rich Propellant Based on BP Neural Network,”
405-417. Journal of Combustion, Vol. 2012, 2012, p. 635190.
https://doi.org/10.6125/JoAAA.202403_56(1S).29 https://doi.org/10.1155/2012/635190
[8] Eisenreich N, Kugler HP, Sinn F, “An Optical [18] Xiao L, Pang W, Qin Z, Li J, Fu X, Fan X, “Cluster
System for Measuring the Burning Rate of Solid Analysis of Al Agglomeration in Solid Propellant
Propellant Strands,” Propellants, Explosives, Combustion,” Combustion and Flame, Vol. 203,
Pyrotechnics, Vol. 12, No. 3, 1987, pp. 78-80. 2019, pp. 386-396.
https://doi.org/10.1002/prep.19870120304 https://doi.org/10.1016/j.combustflame.2018.12.03
[9] Liu T-K, “Correlations of Uncertainties of 2
Composite Propellant Strand Burner Burning Rate [19] Ihme M, Chung WT, Mishra AA, “Combustion
Measurement for Quality Control,” Propellants, Machine Learning: Principles, Progress and
Explosives, Pyrotechnics, Vol. 36, No. 2, 2011, pp. Prospects,” Progress in Energy and Combustion
131-139. Science, Vol. 91, 2022, p. 101010.
https://doi.org/10.1002/prep.200900098 https://doi.org/10.1016/j.pecs.2022.101010
[10] Tang JL, Cai CZ, Zhao S, Wang GL, “Prediction of [20] Oh S-H, Lee HJ, Roh T-S, “New Design Method of
Burning Rate of HTPB Propellant by Using Support Solid Propellant Grain Using Machine Learning,”
Vector Regression,” presented at the 2011 6 th IEEE Processes, Vol. 9, No. 6, 2021, p. 910.
International Conference on Nano/Micro https://doi.org/10.3390/pr9060910
Engineered and Molecular Systems, 2011. [21] Liu JM, Tang SC, Xu FM, Zhuo WL, “Prediction of
DOI:10.1109/NEMS.2011.6017463 Burning Rate of HTPB Propellant by Artificial
[11] Parinet J, Julien M, Nun P, Robins RJ, Remaud G, Neural Network Model,” Chinese Journal of
Höhener P, “Predicting Equilibrium Vapour Pressure Explosives and Propellants, Vol. 29, No. 3, 2006,
Isotope Effects by Using Artificial Neural Networks pp. 13-16.
or Multi-Linear Regression – A Quantitative [22] Zhou W, Wang X, Liu B, Zhao M, Zhang Y, Ma Y,
Structure Property Relationship Approach,” “Machine Learning Method to Predict Solid
Chemosphere, Vol. 134, 2015, pp. 521-527. Propellant Breakage Efficiency of Cavitation Water
https://doi.org/10.1016/j.chemosphere.2014.10.079 Jet,” Propellants, Explosives, Pyrotechnics, Vol. 47,
[12] Jung MY, Chang JH, Oh M, Lee C-H, “Dynamic No. 12, 2022, p. e202200131.
Model and Deep Neural Network-Based Surrogate https://doi.org/10.1002/prep.202200131
Model to Predict Dynamic Behaviors and Steady- [23] Koul A, Ojha A, Vimal P, Pal Y, Mahottamananda S
State Performance of Solid Propellant Combustion,” NSS, Trache D, “Enhancement of the Energetic
Combustion and Flame, Vol. 250, 2023, p. 112649. Performance of Solid Fuels with Metal-
https://doi.org/10.1016/j.combustflame.2023.11264 Fluoropolymer Additives,” FirePhysChem, 2023, p.
9 S2667134423000494.
[13] Klinger D, Casey A, Manship T, Son S, Strachan A, https://doi.org/10.1016/j.fpc.2023.09.001
“Prediction of Solid Propellant Burning Rate [24] Qian W, Huang J, Guo S, Duan B, Xie W, Liu J,
Characteristics Using Machine Learning Zhang C, “Searching for the Analogues of 1,1-
Techniques,” Propellants, Explosives, Pyrotechnics, Dinitro-2,2-Diamino Ethylene (FOX-7) by High-
Vol. 48, No. 4, 2023, p. e202200267. Throughput Computation and Machine Learning,”
https://doi.org/10.1002/prep.202200267 FirePhysChem, Vol. 3, No. 4, 2023, pp. 339-349.
[14] Rivera J, San Martin D, Gollner M, Torres CE, https://doi.org/10.1016/j.fpc.2023.04.002
Fernandez-Pello C, “A Machine Learning Approach [25] Guo S, Huang J, Qian W, Liu J, Zhu W, Zhang C,
to Predict the Critical Heat Flux for Ignition of Solid “Discovery of High Energy and Stable Prismane
Fuels,” Fire Safety Journal, Vol. 141, 2023, p. Derivatives by the High-Throughput Computation
 Prediction of Regression Rate of HTPB Solid Fuel-A Machine Learning Approach 799

and Machine Learning Combined Strategy,”

FirePhysChem, Vol. 4, No. 1, 2024, pp. 55-62.
https://doi.org/10.1016/j.fpc.2023.07.002