DATA-DRIVEN CONTROLOF MECHANICAL PARAMETERS INCOMPOSITE STRUCTURES

The study presents a methodology to assess «material state» in polymer composites based on mathematical statistics and machine learning algorithms. The proposed framework transforms heterogeneous mechanical properties into an integrated digital material model, enabling quantitative characterization of both the reference state and state changes s caused by process induced defects, damage accumulation, or environmental degradation. The approach combines Pearsonian correlation analysis, hierarchical clustering, principal component analysis (PCA), and a multivariate statistical distance criterion suitable for detecting shifts in the material’s structural response. The methodology is validated on two case studies demonstrating how to arrange material properties into a low dimensional PCA-space, reflecting dominant deformation and failure mechanisms (fiber-dominated, matrixdominated, and bearing -dominated modes), along with a procedure for quantifying in-service deterioration. The resulting digital model provides a foundation for adaptive, data-driven material qualification, enabling the development of machine-learning-enhanced normalization and certification frameworks for composite structures.