Premium
STATISTICAL EVALUATION OF FRACTURE TOUGHNESS TEST DATA
Author(s) -
Hauge Mons,
Thaulow Christian
Publication year - 1993
Publication title -
fatigue and fracture of engineering materials and structures
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.887
H-Index - 84
eISSN - 1460-2695
pISSN - 8756-758X
DOI - 10.1111/j.1460-2695.1993.tb00733.x
Subject(s) - fracture toughness , parametric statistics , fracture (geology) , fracture mechanics , reliability (semiconductor) , sensitivity (control systems) , calibration , structural engineering , sample (material) , materials science , statistics , mathematics , engineering , composite material , physics , power (physics) , quantum mechanics , electronic engineering , thermodynamics
— Application of fracture mechanics for verification of the fracture resistance of structural components and pipelines is well established in the offshore industry, and development of reliability analysis methods and calibration of assessment procedures is now in progress. One important parameter in fracture mechanics evaluations is the fracture toughness—a parameter characterized by large scatter, sensitivity to fabrication conditions and large costs related to testing. This has raised the need for efficient methods for characterization of the fracture toughness test data in terms of characteristic values and parametric distribution functions. Two relevant data sets are investigated with respect to their statistical properties and the ability of different distribution functions to fit the data. Three methods for estimation of characteristic values for fracture toughness are described. Their relevance and capabilities are investigated by numerical simulations with sample data sets drawn randomly from a larger population. It is concluded that the distribution functions based on a physical model of the fracture have disadvantages such as unstable behaviour for small data sets and lack of capability in providing certain estimations. The log‐normal distribution showed a stable and predictable behaviour for sample sizes down to 3, the methods for calculation of distribution parameters and characteristic values with specified confidence levels are demonstrated to be satisfactory.