The Effect of Central Flaw on the Unconfined Strength of Rock-Like Specimens: An Intelligent Approach


The strength of jointed rock mass is significantly affected by discontinuities. In particular, the impact of non-persistent joints on the strength of rock mass needs to be investigated further. In this paper, a vast number of uniaxial experiments were conducted on cylindrical specimens with a central flaw created from plaster. The effect of three parameters, namely the inclination of the flaw (θ), flaw length (2a) and flaw aperture (A), on the uniaxial strength of specimens have been investigated. Several intelligent methods, such as artificial neural network, adaptive neuro-fuzzy inference system and a combination of neuro-fuzzy inference system with particle swarm optimization and genetic algorithm along with statistical method, were adopted to estimate unconfined strength of flawed specimens. Statistical indices such as correlation variance accounting, mean square error, correlation coefficient and mean absolute error were employed to evaluate the models' performance. The result of experiments showed that at low flaw length, aperture affects uniaxial strength significantly, while increasing the flaw length the effect of flaw aperture decreases dramatically. Furthermore, shorter flaw lengths result in a drop in stress at a flaw angle of 45 degrees; however, at longer flaw lengths, an increasing flaw inclination results in a constant strength increase. Moreover, the comparison between heuristic methods revealed that the neuro-fuzzy inference system with particle swarm optimization model has better performance than the genetic algorithm-based model and both have priority over other models. Finally, sensitivity analyses revealed that the flaw length is the most effective parameter on the strength, while the flaw aperture is the least effective variable.


Mining Engineering

Keywords and Phrases


International Standard Serial Number (ISSN)

2364-1843; 2228-6160

Document Type

Article - Journal

Document Version


File Type





© 2022 Springer, All rights reserved.

Publication Date

02 Feb 2022