Mishal Thapa

View Video Presentation: https://doi.org/10.2514/6.2022-0805.vid Complex engineering systems are difficult to model using artificial neural networks (ANN) when the number of samples is limited due to the associated high computational cost. In addition, the distribution of samples impacts the ability of the ANN to capture the true response behavior. Therefore, an adaptive sampling-based ANN is presented to approximate the time-consuming responses. The adaptive D-optimal design, which minimizes the determinant of the inverse of information matrix, is used here as a sampling strategy. The information matrix is formulated by considering the samples and activation values of the neurons in the hidden layers. In addition, a sampling approach based on the max-min technique is used in conjunction with the optimality sampling criterion to avoid repetition of the sampling points and enhance the space-filling property. The proposed approach has been implemented to two non-linear analytical functions and an engineering problem: i) Modified Ishigami function(three-dimensional), ii)Modified Sobol function(eight-dimensional), and iii) Fiber-Metal Laminate(FML) problem(five-dimensional). The results thus obtained were also compared with the regular one-shot sampling and sequential sampling strategy. The proposed approach required fewer training samples and has better convergence than the sequential sampling approach. Finally, to show the efficiency and accuracy of the presented approach in surrogate-based optimization, the optimum results were compared to the true optimum of the function. Hence, the proposed method will aid in obtaining high-precision surrogate models with less computational effort.

This paper presents the study of repeatable self-healing capability of thermoplastic healing agent poly-caprolactone (CAPA-6506) and shape memory polymer (SMP) in a fiber reinforced polymer composite structure. This self-healing is inspired by the bio-mimetic process of ‘close then heal’ mechanism where the thermally actuated SMP causes the closing of cracks and the thermoplastic healing agent performs the healing. In this study, two types of sixteen layered unidirectional carbon fiber reinforced thermoset polymer (SC-780) composites were tested: (1) with SMP (containing 6.4 vol% of CAPA and 0.6 vol% of SMP), and (2) without SMP (containing 6.4 vol% of CAPA only). To quantify the healing efficiency in terms of Mode-I interlaminar fracture toughness, double-cantilever beam tests were carried out for seven separate healing cycles by following the ASTM-D5528-13 standard. The tests yielded promising results as maximum healing efficiency of 73%, and 63% were achieved for with SMP and without SMP specimens, respectively. In addition, increased surface roughness due to the crack deviation, SMP pullout, SMP and CAPA bridging between the layers, and plastic deformation of the matrix that are responsible for crack growth resistance were observed during fractography analysis. Also, the relationships between the composition and material properties of epoxy and healants were thoroughly investigated via Fourier transform infrared (FTIR) spectra and differential scanning calorimetry (DSC).