Abstract
Abstract
The morphology of a collision cascade is an important aspect in understanding the formation of defects and their distribution. While the number of sub-cascades is an essential parameter to describe the cascade morphology, the methods to compute this parameter are limited. We present a method to compute the number of sub-cascades from the primary damage state of the collision cascade. Existing methods analyze peak damage state or the end of ballistic phase to compute the number of sub-cascades which is not always available in collision cascade databases. We use density based clustering algorithm from unsupervised machine learning domain to identify the sub-cascades from the primary damage state. To validate the results of our method we first carry out a parameter sensitivity study of the existing algorithms. The study shows that the results are sensitive to input parameters and the choice of the time-frame analyzed. On a database of 100 collision cascades in W, we show that the method we propose, which analyzes primary damage state to predict number of sub-cascades, is in good agreement with the existing method that works on the peak state. We also show that the number of sub-cascades found with different parameters can be used to classify and group together the cascades that have similar time-evolution and fragmentation. It is seen that the number of SIA and vacancies, % defects in clusters and volume of the cascade, decrease with increase in the number of sub-cascades.