Point Cloud Failure Criterion for Composites using k-Nearest Neighbor Classification
This work addresses the problem of improving failure prediction accuracy for composite materials, which is crucial for structural engineering applications, but it appears incremental as it builds on existing multi-scale modeling and classification techniques.
The paper tackles the challenge of predicting failure in composite materials by using a multi-scale modeling scheme that generates a point cloud of failure surface data, which is then queried during finite element analysis using k-nearest neighbor classification to detect failure onset. A static finite element example with a unidirectional composite demonstrates that the framework is effective and efficient, with potential for extension to various composite architectures.
Numerous theories of failure have been postulated and implemented in various commercial programs for composite materials. Even the best theories have had limited success in predicting damage and failure in validation exercises. In view of this background, many researchers have started exploring the use of multiscale modeling to improve the fidelity of the modeling and simulation of various structural and materials systems. In this paper, a multi-scale modeling scheme is used to illustrate how a combination of virtual and laboratory testing programs can be used to generate a point cloud of failure surface data that can then be queried during finite element analysis at the continuum scale to ascertain if the onset of failure has occurred. The k-nearest neighbor (k-NN) classification concept is used to obtain the answer to the query. A linear, elastic, static finite element example using a unidirectional composite shows that the framework can be generated and used effectively and efficiently with the possibility to extend the approach for all types of composite architectures and behaviors.