PSI - Issue 71

Akash S.S. et al. / Procedia Structural Integrity 71 (2025) 180–187

186

fibers was doneusing the Hough Circle Transform. The mean, standard deviation and number of fibers were obtained from eachimage and given as input to the ANN model. One of the images with the detected fibers is shown in Figure 5. The fiberdiameter distributions of the 32 images are represented as Probability distribution functions (PDFs) in Figure 4(b) to visualize thevariations in the fiber diameter.The fiber diameter distributions obtained from the images were also usedto create FE models in ABAQUS to perform simulations and compare the results with the ANN results. The trained ANN model predicted the elastic properties of the material corresponding to each image with respect to the mean, standard deviation and number of fibers which was given as input to the model. The predicted elastic properties E 11 , E 22 , v 12 and G 12 were plotted as probability distribution functions with respect to the fiber diameter distributionstorepresent theuncertainty inthese properties. The resultsobtained using FE analysisarealsoplottedin the same plots and shown in Figure 6. As it can be observed from the plots, the model has predicted with good accuracy compared to the FEA results. The maximum deviation from the FEA results was found to be 0.22 %. The FE simulations required around 11 hoursto compute the uncertainty whereas the ANN model required only 4 seconds.

Fig. 5: Fibers detected from the image using image processing

Fig.6: Comparison of results in the form of Probabilitydistribution functionsofuncertaintiesin theelasticpropertiesobtained fromthe microstructure image using the trained ANN model and FE analysis 6. Conclusions A deep learning ANN model has been developed for the prediction of uncertainty in elastic properties of unidirectional composites due to the variability in fiber diameters. This is predicted from the fiber diameter distributionin SVEs obtained by splitting the microstructural image of the UD composite using image