Structure Tensor Parameter Guidelines for Orientation Estimation in Unidirectional Fibre Reinforced Composites

Abstract. Structure tensor analysis is widely used to estimate local orientations in image data and has become a standard tool for fibre orientation characterisation in X-ray CT datasets. However, practical guidelines for selecting the gradient and integration scales that govern orientation estimation remain limited. In this work, quantitative parameter-selection guidelines are established using unidirectional GFRP and CFRP composites acquired with laboratory and synchrotron CT, respectively. Fibre orientations obtained from fibre tracking are used as independent reference data and compared with voxel-wise structure tensor orientation estimates across a broad parameter space. Parameter robustness is further evaluated using datasets degraded through controlled projection undersampling and noise addition prior to reconstruction. Across all investigated datasets and reconstruction conditions, similar low-error parameter regions are observed despite differences in fibre diameter, spatial resolution, image contrast, imaging modality, and reconstruction quality. Although image degradation increases orientation error and reduces the extent of the low-error region, the locations of the optimal and near-optimal parameter combinations remain largely unchanged. The results show that gradient scales corresponding to approximately 30–40% of the fibre radius and integration scales 4–8 times larger provide accurate and robust orientation estimates. Combined with computational cost analysis, these findings establish practical fibre-diameter-based guidelines for structure tensor parameter selection in CT-based orientation analysis of unidirectional fibre composites.