Dissertation
Physics-based machine-learned models for multi-scale materials response to shock loads
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2019
DOI: 10.17077/etd.005208
Abstract
Energetic materials are distinguished from other combustibles by their relatively high reaction rates. These reaction rates differ between energetic materials and, therefore, indicate the relative sensitivity for each material. Sensitivity, in energetic materials depend on the molecular structure of the material and on the microstructure of a particular sample of this material. It is understandable that different heterogenous energetic materials possess different sensitivity threshold and/or performance due to their different molecular structures. However, it has repeatedly found in the literature that sensitivity and/or performance could also change for the same material. Since, nevertheless, those materials are heterogenous (i.e., filled with voids, cracks, etc.) in their microstructures, variations in microstructures affect performance and sensitivity for each produced sample of a given material. Moreover, even if the manufacturing process attempts to minimize those variations; in the microstructure, the targeted performance and/or safety requirements could be compromised by inevitable, post-manufacturing changes to the originally produced microstructures, for the same material. These changes are due to the high gradients in the operational environmental conditions and/or aging due to extensive periods of storage time.
The linkage between different microstructures and different performance and/or sensitivity for the same energetic material has been established in the literature. It was found that the inherent voids available in the microstructure of such heterogenous materials play a key role in initiation and, consequently, variation in performance and sensitivity. This variation is due to the void-collapse phenomenon occurs as a shock wave passes through a given sample of those materials. As voids collapse, they form localized areas of high temperatures knows as hotspots. Thermal decomposition of the material is initiated at these hotspots. The dynamics of theses hotspots at the meso-scale (shape, intensity, and evolution) are studied in the work presented in this thesis. The goal is to bridge this meso-scale phenomenon (i.e., void-collapse) to sensitivity and performance at the macro-scale. High fidelity direct numerical simulation (DNS) approach is used to conduct reactive void-collapse numerical experiments, by varying the void size and shock loading conditions. Thereafter, the reaction rate value for each case is collected. These experiments are done for two energetic materials; namely, HMX and TATB. The data collected are not only used to study the void-collapse phenomenon in a multi-scaling approach for each sole material, but to also determine what are the key parameters that affect the extraordinarily high chemical reaction rates in energetic materials.
This is done by utilizing machine learning techniques. Even though those techniques devolve into a form of a correlation method due to the data limitation; only two materials, the available data are successfully used to train a predictive model. This model is used to estimate the reaction rates for a third, hypothetical, energetic material. The chemical properties of this third material are bounded by those for HMX and TATB. Along with the capability of predicting the reaction rates of HMX and TATB under various conditions, the model shows a promising performance in estimating the reaction rates for unseen data; a third material. This is done by finding the key parameters that dictate the behavior of energetic materials. Therefore, those parameters could be used to generalize the prediction of reaction rates for any energetic material.
In addition to recommendations for any future work, the results of this preliminary investigation (i.e., the use of machine learning in this field) along with the insights gained about hotspots dynamics in energetic materials are presented in this thesis.
Details
- Title: Subtitle
- Physics-based machine-learned models for multi-scale materials response to shock loads
- Creators
- Anas Nassar
- Contributors
- H.S. Udaykumar (Advisor)Albert Ratner (Committee Member)James Buchholz (Committee Member)Hongtao Ding (Committee Member)Stephen Baek (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Mechanical Engineering
- Date degree season
- Autumn 2019
- DOI
- 10.17077/etd.005208
- Publisher
- University of Iowa
- Number of pages
- xvi, 171 pages
- Copyright
- Copyright 2019 Anas Nassar
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 167-171)
- Public Abstract (ETD)
Energetic materials are materials that burn in distinguishably high rates. Propellants, explosives, and pyrotechnics are classifications of such materials. A given material does not need to have high energy density values stored in its molecules to be considered an energetic material. In fact, a chocolate bar has more energy per unit mass than many explosives do. However, the energy release rate in chocolate could be hours or even days. On the other hand, the time it takes a specimen of explosive to release the energy stored in its molecules is in the order of microseconds, or even nanoseconds. This is the character which makes those materials called energetic materials – the rate of energy release.
This high reaction rate is studied in the work presented in this thesis. The study is done for two different energetic materials. Numerical experiments are conducted to achieve this goal. The data collected from all different experiments for those two materials are used to build a machine learning model that can predict the behavior of a third material, along with the original two. The machine learning model is capable of doing that without running any numerical experiments for this third material; the model has not seen any data for this material before.
The preliminary results presented in this thesis are promising and suggesting for more work to be done in this direction –applying machine learning techniques to better understand physics in energetic materials. This will enhance both performance and safety when dealing with such materials under various operational conditions.
- Academic Unit
- Mechanical Engineering
- Record Identifier
- 9983779697402771
Metrics
87 File views/ downloads
613 Record Views