No abstract

Scanning Electron Microscopy (SEM) is traditionally leveraged to image fracture surfaces and generate information for analysis. Conventionally, experts identify patterns of interest in SEM images and link them to fracture phenomena based on knowledge and experience. Such practice has substantial limitations. It relies on expert opinions for decision-making, which poses barriers for practitioners without relevant background; manual inspection must be done for individual SEM images, thus time-consuming and inapt for industrial automation. There is a genuine demand for a fast, automatic method for fractographic pattern recognition. Targeting the problem, this study proposes a two-stage data-driven approach based on clustering. In offline analysis (Stage 1), a clustering algorithm identifies the generic fractographic patterns on part. Each pattern corresponds to a cluster. Expert evaluation of the part’s crack status is leveraged to map individual patterns (clusters) to a crack type. In in-situ monitoring (Stage 2), SEM images of new parts are matched to the clusters from stage 1, which reveals the generic patterns on the part and indicates the potential crack status. The proposed approach enables automatic fractographic analysis without experts. It is demonstrated to be effective on real SEM images of additively manufactured Inconel-718 specimens subjected to high cycle fatigue.

No abstract

No abstract