Extraction and Quantification of Features in XCT Datasets of Fibre Reinforced Polymers using Machine Learning Techniques

Abstract: This master’s thesis shows the extraction, quantification and visual analysis of pores and individual fibres in fibre reinforced polymer (FRP)materials. The core methods used and advanced for this purpose are tailored deep learning techniques, which are coupled with interactive visualisation.These techniques were applied to X-ray Computed Tomography(XCT) data to extract pores and fibres of carbon (CFRP) and glassfibre reinforced polymers (GFRP). Although segmentation is widely examined, there is still a high necessity to come up with improved methods ,especially given the huge potential of machine learning. In this thesis, we aimed at designing efficient and powerful segmentation models with reasonable performance on consumer-grade GPU systems. At the heart of the studied machine learning techniques for segmentation was U-Net [39], a deep convolutional neural network based segmentation technique. The main contributions of this thesis are seen in modifying and improving U-Net’s layer architecture to facilitate the segmentation of pores and fibres in 3D-XCT data of FRPs. Furthermore, a hyper-parameter optimization was completed through a parameter analysis with a tuning function. The results with highest accuracy from all suitable hyper-parameters were used for the final training process. The trained model (prediction model) has finally been implemented and integrated as a segmentation filter in open iA [13], in order to facilitate an efficient segmentation of XCT datasets with the fully trained model. Regarding evaluation, a quantitative feature analysis was carried out. Pores and fibres were counted, their shape and volumetric characteristics were computed to support the visual analysis of the generated results. We examined our findings on different real-world XCT datasets of carbon and glass fibre reinforced composites and tested the model on both high and low-resolution datasets: The proposed method shows in all test cases an efficient and robust segmentation of pores with different shapes and size, and that also barely visible features are detected correctly. We finally show evidence that the proposed modified U-Netis more suitable than other segmentation methods for pore segmentation .Even when trained only with a low number of datasets, it returns a reasonable prediction accuracy. In terms of computation time, this techniqueis suitable even on consumer-grade GPUs: For dataset sizes of over 1000x1000x1000 (4GB) voxels, the complete segmentation procedure only takes approximately 4 minutes.