Deep-Neural-Networks-Based Data-Driven Methods for Characterizing the Mechanical Behavior of Hydroxyl-Terminated...

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Modeling

Background:

• Hydroxyl-terminated polyether (HTPE) propellants are crucial in the weapons industry due to their insensitive nature.
• Accurate mechanical modeling of solid propellants is vital for understanding their behavior during storage, combustion, and explosion.
• Existing mechanical models may not fully capture the complex stress-strain relationships of composite propellants.

Purpose of the Study:

• To apply deep neural networks for the first time to model the mechanical behavior of composite solid propellants.
• To develop a data-driven framework with a novel training-testing splitting strategy for propellant mechanical modeling.
• To evaluate the performance of different neural network architectures, including FFNNs, KANs, and LSTMs.

Main Methods:

• Implementation of Feedforward Neural Networks (FFNNs), Kolmogorov-Arnold Networks (KANs), and Long Short-Term Memory (LSTM) networks.
• Utilization of a Bayesian optimization algorithm for optimizing model frameworks and parameters.
• Application of the Shapley Additive Explanations (SHAP) method for interpreting model predictions and identifying key formulation factors.

Main Results:

• The LSTM model achieved a Root Mean Square Error (RMSE) of 0.053 MPa, significantly outperforming FFNNs (62.7% improvement) and KANs (48.5% improvement).
• LSTM model R-squared values exceeded 0.99 on the testing set, demonstrating high accuracy in predicting the stress-strain curve.
• The LSTM model effectively captured the influence of tensile rate and temperature on tensile strength, accurately predicting yield points and slope changes.

Conclusions:

• Deep neural networks, particularly LSTM, provide a highly accurate and effective method for modeling HTPE propellant mechanical behavior.
• The SHAP analysis revealed that fine-grained ammonium perchlorate (AP) enhances tensile strength, while plasticizers improve elongation at break.
• This data-driven approach offers a powerful tool for optimizing HTPE propellant formulations, enhancing material properties and safety.