Modeling Damage Evolution and Damage Healing using Data Driven Method

Advanced aircraft use composite materials in over 50% of their airframe structures. Defects in composite materials frequently arise during manufacturing and processing phases, greatly affecting structural integrity and reliability. The intrinsic self-healing polymers are smart materials that have the capability to recuperate a crack partially or fully by intermolecular interaction such as disulfide, hydrogen bonding, hydrophobic interaction, ionic bonding. In this dissertation, we mainly have two contributions: (1) We used zero-bias deep neural networks to analyze a multisource dataset developed by Pragathi comprising grayscale images and strain contours during composite materials processing. We also utilized a Zero-Bias Deep Neural Network model to effectively detect defects in composite materials, thus enabling their repair or mitigation. In this approach, the last dense layer of a standard deep neural network is modified by incorporating a dimensionality reduction layer and a similarity-matching layer. The model identified multiple defects without individual training on any defective images. While grayscale images and strain data were successfully used independently for defect detection, the proposed approach achieved superior performance by combining both, resulting in a high anomaly detection accuracy of 99.09%. Our findings demonstrate a clear synergistic effect, where integrating strain profiles with grayscale images enhances defect detection effectiveness. (2) We also explored the study of self- healing polymer/carbon fiber composite, synthesized using a one-pot polycondensation method. This method combines amino-terminated poly(dimethylsiloxane) (Mn = 5000) with 4,4′- methylenebis (phenyl isocyanate) and isophorone diisocyanate, applied to carbon unidirectional fiber. We envisage applications where the self-healing polymer present in the composite interfaces heals the cracks on carbon fibers.