Cluster Analysis in the Choice of Operating Modes in Durability Analysis of Random Time-History Records

Computational methods for assessing the structural durability of components under service loadings need, among other data, a detailed information about the service loading itself. Due to the usual non-stationary character of service loadings, the generalized loading block (GB) should be constructed based on the occurrence frequency of various service modes in the loading. A method for selecting the modes is proposed in this work, which proves to be particularly advantageous in those cases where it is difficult to distinguish the operating modes appropriately. The method is based on one of machine-learning tools called cluster analysis, which is here applied to representative loading time-histories recorded in a mountain-bike. To prove the correctness of the method, a posteriori comparison is made with known information about the service modes of the loading.