Abstract:

Laboratory test is the most basic method to determine uniaxial compressive strength of rock, but its value often differs significantly due to the heterogeneity of samples. Taking the sampling difficulty as well as the time and cost factors into account, it will be favorable to determine the uniaxial compressive strength with appropriate prediction models. Complete comprehension of the geological nature of rock is the bridge that leads to an accurate description of its physical and mechanical properties. The geological nature of rock consists of the intrinsic properties of rock substance, structure and occurrence state. Based on comprehensive consideration of those intrinsic properties of rock and their connection with uniaxial test, methods of regression and BP neural network are adopted to predict the uniaxial compressive strength of carbonate with basic index parameters of mineral composition, density, longitudinal wave velocity and saturation state, then grey correlation analysis is conducted to verify the rationality of chosen index parameters. Practice indicates that the regression method has a maximum error of 15.3% in prediction, while the BP neural network method shows a maximum of 8.5% error; The cause of prediction error is that the carbonate has a complex composition, whereas those selected parameters of mineral composition used in prediction are merely a simplification of the actual condition, meanwhile the expansion in marlstone is another reason that leads to the differences between the measured and predicted value of uniaxial compressive strength of carbonate.

Key words: geological nature of rock, uniaxial compressive strength, empirical formula, neural network, grey correlation analysis, carbonate rock