An adaptive neuro-fuzzy inference system for predicting unconfined compressive strength and Young’s modulus: a study on Main Range granite

• Published in: Bulletin of engineering geology and the environment Vol. 74; no. 4; pp. 1301 - 1319
• Main Authors: Jahed Armaghani, Danial, Tonnizam Mohamad, Edy, Momeni, Ehsan, Narayanasamy, Mogana Sundaram, Mohd Amin, Mohd For
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.11.2015; Springer Nature B.V
• Subjects: Adaptive systems; Artificial neural networks; Civil engineering; Compressive strength; Earth and Environmental Science; Earth Sciences; Foundations; Fuzzy logic; Geoecology/Natural Processes; Geoengineering; Geological engineering; Geotechnical Engineering & Applied Earth Sciences; Granite; Hydraulics; Laboratories; Learning theory; Mathematical models; Modulus of elasticity; Nature Conservation; Neural networks; Original Paper; Prediction models; Raw water; Regression analysis; Rock properties; Rocks; Water transfer; Malaysia; Young’s modulus; Multiple regression analysis; Adaptive neuro-fuzzy inference system; Unconfined compressive strength; Granite
• ISSN: 1435-9529; 1435-9537
• DOI: 10.1007/s10064-014-0687-4

Engineering properties of rocks such as unconfined compressive strength (UCS) and Young’s modulus ( E ) are among the essential parameters for the design of tunnel excavations. Many attempts have been made to develop indirect methods of estimating UCS and E . This is generally attributed to the difficulty of preparing and conducting the aforementioned tests in a laboratory. In essence, this study aims to present two predictive models of UCS and E for granite using an adaptive neuro-fuzzy inference system (ANFIS). The required rock samples for model development (45 granite sample sets) were obtained from site investigation work at the Pahang-Selangor raw water transfer tunnel, which was excavated across the Main Range of Peninsular Malaysia. In developing the predictive models, dry density, ultrasonic velocity, quartz content and plagioclase were set as model inputs. These parameters were selected based on simple and multiple regression analyses presented in the article. However, for the sake of comparison, the prediction performances of the ANFIS models were checked against multiple regression analysis (MRA) and artificial neural network (ANN) predictive models of UCS and E . The capacity performances of the predictive models were assessed based on the value account for (VAF), root mean squared error (RMSE) and coefficient of determination ( R 2 ). It was found that the ANFIS predictive model of UCS, with R 2 , RMSE and VAF equal to 0.985, 6.224 and 98.455 %, respectively, outperforms the MRA and ANN models. A similar conclusion was drawn for the ANFIS predictive model of E where the values of R 2 , RMSE and VAF were 0.990, 3.503 and 98.968 %, respectively.